CoChat
Sign Up
Cited by 404Neural dynamics and brain functionMemory and Neural MechanismsNeural Networks and Applications

Towards a mathematical theory of cortical micro-circuits

Open access
George, Dileep, Hawkins, Jeff·2009·PLoS Computational Biology
DOI 10.1371/journal.pcbi.1000532PMC PMC2749218
Open this paper in CoChat — free accountOpen PDF

Abstract

The theoretical setting of hierarchical Bayesian inference is gaining acceptance as a framework for understanding cortical computation. In this paper, we describe how Bayesian belief propagation in a spatio-temporal hierarchical model, called Hierarchical Temporal Memory (HTM), can lead to a mathematical model for cortical circuits. An HTM node is abstracted using a coincidence detector and a mixture of Markov chains. Bayesian belief propagation equations for such an HTM node define a set of functional constraints for a neuronal implementation. Anatomical data provide a contrasting set of organizational constraints. The combination of these two constraints suggests a theoretically derived interpretation for many anatomical and physiological features and predicts several others. We describe the pattern recognition capabilities of HTM networks and demonstrate the application of the derived circuits for modeling the subjective contour effect. We also discuss how the theory and the circuit can be extended to explain cortical features that are not explained by the current model and describe testable predictions that can be derived from the model.

AI snapshot

Written from the paper’s full text.

TL;DR

The paper formulates Hierarchical Temporal Memory (HTM) as a Bayesian belief-propagation model of cortical microcircuits and maps its abstract neuronal implementation onto laminar and columnar cortical anatomy. It also reports HTM vision experiments, including object recognition benchmarks and a feedback-based model of the subjective contour effect.

Key findings

  • The authors derive belief-propagation equations for an HTM node using coincidence patterns and Markov chains, then propose a cortical mapping: layer 4 stellate cells for coincidence likelihood, layer 2/3 pyramidal cells for sequence states and feed-forward/feedback computations, layer 5 cells for belief and timing, and layer 6 cells for feedback messages to child regions.
  • On the Pictures line-drawing dataset, an HTM with Markov-chain temporal pooling only at level 1 achieved 49% recognition accuracy; adding temporal pooling at higher levels increased accuracy to 72%, compared with 35% for a nearest-neighbor classifier using the same top-level training paradigm.
  • On Caltech-101, initial HTM experiments achieved 50% recognition with 15 training images and 62% with 30 training images; on a 4-category grayscale set with over 10,000 training images, the network gave 92% accuracy on a holdout set. Temporal inference improved frame-by-frame recognition by up to 30% over sliding-window averaging (window length = 4).
  • In a Kanizsa square test, the network classified the pattern as a rectangle despite not being trained on Kanizsa squares; a node with no feed-forward contour input showed nonzero feedback, modeling the subjective contour effect, and a corrupted Kanizsa figure produced a weaker response.

Methods

  • Theoretical/modeling study: starts from the HTM generative model, derives Bayesian belief-propagation equations for an HTM node, builds an abstract neuron-like circuit, and maps it onto cortical layers and columns by matching computational specifications with anatomical data.
  • Experiments used Numenta’s NuPIC HTM networks for visual object recognition on the Pictures dataset (48 categories of 32×32 line drawings), Caltech-101, a 4-category grayscale dataset, temporal inference on Pictures, top-down segmentation, and subjective contour/Kanizsa tests.
  • Analysis consisted of benchmark recognition-accuracy comparisons and qualitative inspection of node feed-forward and feedback states; no biological experiments are reported.

Limitations

  • The paper states the model is a starting point, the biological mapping is not unique, and it does not address learning, attention, motor action, asynchronous messages, detailed dendritic properties, inhibitory cells, cortical maps, or sparse distributed representations within a node; it also assumes singly connected non-loopy networks for simplicity.
  • The paper states the timing/duration mechanism is conjectural, the non-specific thalamic timing loop lacks direct evidence, no recordings from non-specific thalamic cells were found, and the subjective contour results are preliminary.
  • Evident from the text: the biological predictions are largely untested, the benchmark experiments are from Numenta/NuPIC and not independent, and no statistical tests or error bars are reported for the recognition results.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1371/journal.pcbi.1000532

Cites

79 of 79 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 99 cited sources checked · 0 supported · 0 flagged

  • Findingcitation

    A simple way to learn the mixture of Markov chains for real world cases is to learn a large transition matrix that is then partitioned using a graph partitioning algorithm.

    “We have found that a simple way to learn the mixture of Markov chains for real world cases is to learn a large transition matrix that is then partitioned using a graph partitioning algorithm [27].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [27] D George, 2007, The HTM Learning Algorithms

      Not checked in the free run.

  • Findingown data

    Storing a fixed number of a random selection of coincidence patterns is sufficient as long as multiple coincidence patterns are allowed to be active at the same time.

    “we have found that storing a fixed number of a random selection of the coincidence patterns is sufficient as long as we allow multiple coincidence patterns to be active at the same time.”
    Quote in the paperSubstantial claim
  • Findingown data

    Reasonable results can be achieved with a wide range of the number of coincidences stored.

    “We have found that reasonable results can be achieved with a wide range of the number of coincidences stored.”
    Quote in the paperNot a substantial claim
  • Findingown data

    Our literature search and private conversations with several thalamic anatomists have yielded no evidence of anyone ever recording from non-specific thalamic cells.

    “Our literature search and private conversations with several thalamic anatomists have yielded no evidence of anyone ever recording from non-specific thalamic cells.”
    Quote in the paperNot a substantial claim
  • Findingreasoning

    When looking for anatomical data satisfying the constraints, the cortical projections to and from non-specific thalamic cells were the best fit.

    “When we looked for anatomical data that satisfied these constraints, the cortical projections to and from non-specific thalamic cells were the best fit.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 106
  • May combine two claims. 55
  • The quote only partly states this claim. 1

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    HTMs have been used successfully in invariant pattern recognition on gray-scale images, in speaker identification in the auditory domain, and in learning a model for motion capture data in an unsupervised manner.

    “In our research, HTMs have been used successfully in invariant pattern recognition on gray-scale images, in the identification of speakers in the auditory domain and in learning a model for motion capture data in an unsupervised manner.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingnone given

    Researchers outside Numenta have had success training recognition systems using HTMs.

    “We are also happy to note that researchers outside Numenta have had success training recognition systems using HTMs.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    The model presented in the paper does not address most of the knowledge about properties and biophysics of dendrites, dendritic action potentials, and biochemical pathways related to synapses.

    “The model presented here does not address most of this knowledge.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingcitation

    HTM networks reported in previous work used Markov chains based temporal pooling only at level 1 of the hierarchy and gave 49% recognition accuracy on the test set.

    “The HTM networks reported in our previous work [9] used Markov chains based temporal pooling only at level 1 of the hierarchy and gave 49% recognition accuracy on this test set.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [9] D George, 2005, A hierarchical Bayesian model of invariant pattern recognition in the visual cortex.

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    On the standard Caltech-101 benchmark, initial experiments with the network achieved 50% recognition accuracy with 15 training images and 62% recognition accuracy with 30 training images.

    “On the standard Caltech-101 benchmark [60], our initial experiments with the network achieved 50% recognition accuracy with 15 training images and 62% recognition accuracy with 30 training images.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [60] L Fei-Fei, 2006, One-shot learning of object categories., IEEE Transactions on Pattern Analysis and Machine Intelligence source

      Not checked in the free run.

  • The quote only partly states this claim.

    Findingown data

    A network trained with 4 categories of images with large intra-category variation and over 10000 training images gave 92% accuracy on a hold out set.

    “On a hold out set, this network gave 92% accuracy.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    The circuits derived in the paper do not incorporate a detailed mechanism for attention control.

    “the circuits derived in this paper do not incorporate a detailed mechanism for attention control.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingnone given

    The HTM model in the paper uses sparse-distributed representations when considering representations within an entire hierarchical level, but does not use sparse-distributed representations within a node.

    “The HTM model in this paper uses sparse-distributed representations [68],[69] when considering the representations within an entire hierarchical level. However, it does not use sparse-distributed representations within a node.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [68] P Sallee, 2004, Learning sparse multiscale image representations.

      Not checked in the free run.

    • cites [69] H Wersing, 2003, Learning optimized features for hierarchical models of invariant object recognition., Neural computation source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    The corrupted Kanizsa figure was recognized as a rectangle at the top of the network with a lower level of certainty, and this lower certainty was reflected in lower activation of the subjective contour.

    “In our experiment the corrupted figure was recognized as a rectangle at the top of the network, albeit with a lower level of certainty. This lower level of certainty is reflected in the lower activation level of the subjective contour.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Findingnone given

    The circuits discussed in the paper have been mapped to the belief propagation equations in a learned HTM node.

    “The circuits discussed in this paper have been mapped to the belief propagation equations in a learned HTM node.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingnone given

    The authors performed the mapping to biology in two stages: starting with a mathematical expression of how each node performs inference, they created an abstract neuronal implementation; next they mapped this abstract implementation onto observed anatomical data of cell types, cell layers, and micro-circuits in the cortex.

    “We performed the mapping to biology in two stages. Starting with a mathematical expression of how each node performs inference, we created an abstract neuronal implementation. Next we mapped this abstract implementation onto observed anatomical data of cell types, cell layers, and micro-circuits in the cortex.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

    Findingnone given

    There are many unknowns and variations in cortical anatomy, and many functions of the neocortex are not accounted for by the HTM model.

    “There are many unknowns and variations in cortical anatomy, and similarly there are many functions of the neocortex that are not accounted for by the HTM model.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The combination of the functional constraints from HTM belief propagation and the organizational constraints from anatomical data suggests a theoretically derived interpretation for many anatomical and physiological features and predicts several others.

    “The combination of these two constraints suggests a theoretically derived interpretation for many anatomical and physiological features and predicts several others.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given

    The cortical circuit shown in Figure 9 is organized as 5 columns corresponding to the 5 coincidence patterns in the reference HTM node.

    “The cortical circuit shown in Figure 9 is organized as 5 columns corresponding to the 5 coincidence patterns in the reference HTM node that we started with.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    Both of the above cases correspond to gating the bottom-up evidence using top-down activation.

    “Both of the above cases correspond to gating the bottom-up evidence using top-down activation.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    A cortical column can be thought of as encoding a particular value of the random variable that represents the coincidence patterns in the HTM node.

    “A cortical column can be thought of as encoding a particular value of the random variable that represents the coincidence patterns in the HTM node.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Maintaining this gating during further computations requires external control mechanisms that are not part of the standard belief propagation machinery.

    “However, maintaining this gating during further computations requires external control mechanisms that are not part of the standard belief propagation machinery.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    The Memory-Prediction Framework could be formulated in Bayesian terms and given a proper mathematical foundation.

    “George, recognized that the Memory-Prediction framework could be formulated in Bayesian terms and given a proper mathematical foundation [8],[9].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [8] D George, 2008, How the Brain Might Work: A Hierarchical and Temporal Model for Learning and Recognition.

      Not checked in the free run.

    • cites [9] D George, 2005, A hierarchical Bayesian model of invariant pattern recognition in the visual cortex.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    An area of cortex can be thought of as encoding a set of patterns and sequences in relation to the patterns and sequences in regions hierarchically above and below it.

    “An area of cortex can be thought of as encoding a set of patterns and sequences in relation to the patterns and sequences in regions hierarchically above and below it.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • May combine two claims.

    Interpretationreasoning

    The circle-neuron circuits shown in Figure 4 implement Equation 4, and the weights on the lateral time-delayed connections correspond to the transition matrix entries in each Markov chain.

    “Therefore, the circle-neuron circuits shown in Figure 4 implement Equation 4 and the weights on the lateral time-delayed connections correspond to the transition matrix entries in each Markov chain.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The equation for the output of a circle neuron assumes a simple step-wise notion of time.

  • May combine two claims.

    Interpretationreasoning

    Observed information anywhere in the cortex is propagated to other regions through these messages and can alter the probability values associated with the hypotheses maintained by other cortical columns.

    “Observed information anywhere in the cortex is propagated to other regions through these messages and can alter the probability values associated with the hypotheses maintained by other cortical columns.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    The operation of the pentagon neuron is different from that of the rectangle neuron in Figure 4.

    “Note that the operation of the pentagon neuron is different from that of the rectangle neuron in Figure 4.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    The gating signal at each node, corresponding to an activated top-level hypothesis, can be derived from the computed beliefs at that node.

    “The gating signal at each node, corresponding to an activated top-level hypothesis, can be derived from the computed beliefs at that node.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given

    The elements of the derived circuits are consistent with each other in that they operate under the same set of assumptions and work together in a hierarchy.

    “The elements of the circuits are also consistent with each other in that they operate under the same set of assumptions and work together in a hierarchy.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    It is straightforward to extend belief propagation in HTMs to include multiple parents for each node.

    “It is straightforward to extend belief propagation in HTMs to include multiple parents for each node.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    The patterns correspond to the coincidence patterns in an HTM node and the sequences correspond to the Markov chains.

    “The patterns correspond to the coincidence patterns in an HTM node and the sequences correspond to the Markov chains.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    The neurons in layer 2/3 represent the coincidence pattern in the context of different sequences, whereas the neurons in layer 6 represent the participation of the coincidence pattern in the calculation of feedback messages.

    “For example, the neurons in layer 2/3 represent the coincidence pattern in the context of different sequences, whereas the neurons in layer 6 represent the participation of the coincidence pattern in the calculation of feedback messages.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Anatomical data provide a contrasting set of organizational constraints.

    “Anatomical data provide a contrasting set of organizational constraints.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Hierarchical Bayesian inference is gaining acceptance as a framework for understanding cortical computation.

    “The theoretical setting of hierarchical Bayesian inference is gaining acceptance as the framework for understanding cortical computation [1]–[5].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [1]–[5]

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationcitation

    The theory presented in this paper can be useful in the modeling of physiological phenomena.

    “In addition to providing a template for understanding cortical circuits [15], the theory presented here can be useful in the modeling of physiological phenomena.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [15] RJ Douglas, 2004, Neuronal circuits of the neocortex., Annual Review of Neuroscience source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The belief calculation in Equation 6 has almost the same form as the forward dynamic programming Equations 4.

    “The belief calculation, described in Equation 6, has almost the same form as the forward dynamic programming Equations 4.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    This kind of columnar organization is typical in biology.

    “This kind of columnar organization is typical in biology [38],[39].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [38] K Tsunoda, 2001, Complex objects are represented in macaque inferotemporal cortex by the combination of feature columns., Nat Neurosci source

      Not checked in the free run.

    • cites [39] PS Churchland, 1992, The Computational Brain source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    Information can also be propagated down in the hierarchy for attention, segmentation, and filling in missing inputs.

    “Information can also be propagated down in the hierarchy for attention, segmentation, and filling in missing inputs.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    The 5 columnar structures shown represent a set of 5 mutually exclusive hypotheses about the same input space.

    “The 5 columnar structures shown represent a set of 5 mutually exclusive hypotheses about the same input space.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    The external timing circuit achieves the effect of a specific duration model whose tempo can be changed.

    “The external timing circuit achieves the effect of a specific duration model whose tempo can be changed.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Durations are signaled to a node by an external timing unit that determines the rate of change of the signals using some system-level measurements.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationnone given

    The circuits derived in this paper can provide a hypothesis-driven framework for examining neural connectivity.

    “The circuits derived in this chapter can provide a hypothesis-driven framework for examining the neural connectivity.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisenone given

    HTM networks infer on time-varying inputs, and inference on a static input is a special case of this computation.

    “In general, HTM networks infer on time-varying inputs. Inference on a static input is a special case of this computation.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    The resultant biological circuit supports all the Bayesian computations required for temporal, feed-forward, and feedback inference.

    “The resultant biological circuit supports all the Bayesian computations required for temporal, feed-forward, and feedback inference.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Layer 4 is generally accepted as the primary feed-forward input layer to cortical regions.

    “Layer 4 is generally accepted as the primary feed-forward input layer to cortical regions [2].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [2] J Hegde, 2007, Reappraising the functional implications of the primate visual anatomical hierarchy., Neuroscientist source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given

    Layer 3 cells can also act as coincidence detectors of inputs from thalamus that make proximal connections, and layer 3 cells can take the full burden of coincidence detection.

    “Layer 3 cells can also act as coincidence detectors of inputs from thalamus that make proximal connections, and layer 3 cells can take the full burden of coincidence detection.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    The second proposal does not require a second set of neurons but instead relies on long lasting metabotropic responses of neurons.

    “The second proposal does not require a second set of neurons but instead relies on long lasting metabotropic responses of neurons.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    Strong evidence suggesting which of these two, or other, mechanisms is used is lacking.

    “Strong evidence suggesting which of these two, or other, mechanisms is used is lacking.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    These neurons correspond to the 'circle neurons' described in the Model section and implement the dynamic programming Equation 4 in Table 1.

    “These neurons correspond to the ‘circle neurons’ that we described in the Model section and implement the dynamic programming Equation 4 in Table 1.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    Yellow neurons need to have apical dendrites in layer 1 to receive feedback from higher cortical areas.

    “The yellow neurons need to have apical dendrites in layer 1 to receive feedback from higher cortical areas.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    The cells in layer 3 which represent the individual elements in Markov chains will become active in turn as sequences are learned and recalled.

    “The cells in layer 3 which represent the individual elements in Markov chains will become active in turn as sequences are learned and recalled.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    A second set of dynamic programming states were required for the calculation of the belief of coincidence patterns and as an intermediate step in deriving the feedback messages to be sent to the children.

    “In the Model section, we saw that a second set of dynamic programming states were required for the calculation of the belief of coincidence patterns and as an intermediate step in deriving the feedback messages to be sent to the children.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • May combine two claims.

    Premisecitation

    Mini-columns are developmental units that contain about 80 to 100 neurons.

    “Mini-columns are developmental units that contain about 80 to 100 neurons.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [40] P Rakic, 1988, Specification of cerebral cortical areas., Science source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    The pyramidal neurons shown in green in Figure 9 implement the Markov chain sequences and the dynamic programming computations for feed-forward sequential inference.

    “The pyramidal neurons shown in green in Figure 9 implement the Markov chain sequences and the dynamic programming computations for feed-forward sequential inference.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Classified as probably not a substantial claim.

    Interpretationnone given

    In the idealized cortical column model, each different aspect that needs to be represented for a coincidence pattern is represented using a single neuron.

    “In the idealized cortical column model, each different aspect that needs to be represented for a coincidence pattern is represented using a single neuron.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    These pyramidal neurons receive 'vertical' excitatory inputs from the outputs of layer 4 stellate neurons and 'lateral' inputs from other pyramidal cells within layer 2/3.

    “These pyramidal neurons receive ‘vertical’ excitatory inputs from the outputs of layer 4 stellate neurons and ‘lateral’ inputs from other pyramidal cells within layer 2/3.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    These neurons correspond to the filled-circle neurons in Figure 5.

    “These neurons correspond to the filled-circle neurons in Figure 5.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • May combine two claims.

    Interpretationreasoning

    Any computation that requires major inputs from layer 4 will need to be done at layer 3 and should match the general characteristics of layer 3.

    “This means that any computation that requires major inputs from layer 4 will need to be done at layer 3 and should match the general characteristics of layer 3.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    If the layer 3 cells that represent the sequence elements project to metabotropic receptors in higher cortical regions, those destination neurons could stay active for the duration of sequences.

    “If the layer 3 cells that represent the sequence elements project to metabotropic receptors in higher cortical regions, those destination neurons could stay active for the duration of sequences.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    The lateral connections of these neurons are similar to the lateral connections of the layer 2/3 green pyramids.

    “The lateral connections of these neurons are similar to the lateral connections of the layer 2/3 green pyramids that we just described.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Green feed-forward neurons should not make connections with the feedback signal.

    “But the green feed-forward neurons should not make connections with the feedback signal.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    Nothing about the computation or the representation changes if we replicate each neuron in this circuit a few times, while maintaining their connectivity.

    “Nothing about the computation or the representation changes if we replicate each neuron in this circuit a few times, while maintaining their connectivity.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    A second set of pyramidal cells in layer 2/3 corresponds to the Markov chain identities and gets excitatory inputs from the layer 2/3 pyramidal cells that belong to the same Markov chain.

    “These pyramidal cells correspond to the Markov chain identities and get excitatory inputs from the layer 2/3 pyramidal cells that belong to the same Markov chain.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    These neurons are represented using yellow colored pyramidal neurons in layer 2/3.

    “In Figure 9, these neurons are represented using yellow colored pyramidal neurons in layer 2/3.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • May combine two claims.

    Interpretationreasoning

    These neurons do the sequence computations while integrating feedback information from the higher layers.

    “These neurons do the sequence computations while integrating feedback information from the higher layers.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The major bottom-up input required for the above calculations is the feed-forward probability over coincidence patterns that was assigned to layer 4 neurons.

    “The major bottom-up input required for the above calculations is the feed-forward probability over coincidence patterns that was assigned to layer 4 neurons in a previous section.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    The layer 4 stellate cells implement the probability calculation described in Equation 2 and shown in Figure 3.

    “We propose that the layer 4 stellate cells implement the probability calculation described in Equation 2 and shown in Figure 3.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • May combine two claims.

    Interpretationnone given

    The other class of layer 2/3 neuron will not form synapses with feedback axons, but will form synapses with axons from non-specific thalamic cells needed for timing.

    “The other class of layer 2/3 neuron will not form synapses with feedback axons, but will form synapses with axons from non-specific thalamic cells needed for timing (discussed more in a later section).”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationreasoning

    Green and yellow layer 2/3 neurons are needed to learn two sets of sequences: one set for feed-forward calculations and one set for feedback calculations.

    “We show green and yellow layer 2/3 neurons in Figure 9 because we need to learn two sets of sequences. One set of sequences is used in feed-forward calculations and the other set of sequences is used in feedback calculations.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    This second set of pyramidal neurons in layer 2/3 corresponds to the rectangle neurons in Figure 4.

    “This second set of pyramidal neurons in layer 2/3 corresponds to the rectangle neurons in Figure 4.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given

    The horizontal connections between layer 4 cells can implement spatial pooling or temporal pooling without timing.

    “The horizontal connections between layer 4 cells can implement spatial pooling or temporal pooling without timing.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    A mechanism must exist in each region of cortex for forming constant representations for sequences of activity.

    “It is a strong theoretical prediction that a mechanism must exist in each region of cortex for forming constant representations for sequences of activity.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    These neurons send their outputs to higher-level cortical regions.

    “These neurons send their outputs to higher-level cortical regions.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • May combine two claims.

    Interpretationreasoning

    This is consistent with the implementation of sequence memories and dynamic programming computations.

    “This is akin to the blue and magenta axons that we show in Figure 9 and is consistent with the implementation of sequence memories and dynamic programming computations.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Green neurons may also have apical dendrites in layer 1 to receive timing information.

    “The green neurons may also have apical dendrites in layer 1 to receive timing information.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    These connections correspond to learned sequence memories and will be very precise about which columns and which specific neurons within these columns they target.

    “Since these connections correspond to learned sequence memories, they will be very precise about which columns and which specific neurons within these columns they target.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The cortical circuits implement the HTM belief propagation equations described in Table 1.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given

    Cells in layer 2/3 are known to be 'complex' cells that respond to sequence of motion or cells that respond invariantly to different translations of the same feature.

    “Cells in layer 2/3 are known to be ‘complex’ cells that respond to sequence of motion or cells that respond invariantly to different translations of the same feature.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    In the proposed circuit, layer 5 pyramidal cells remember the precise time at which a belief is going to be active as measured as a duration from the previous element in the sequence.

    “In the proposed circuit, layer 5 pyramidal cells remember the precise time at which a belief is going to be active as measured as a duration from the previous element in the sequence.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The layer 2/3 input, plus the apical dendrite extending to layer 1, and the bursting behavior, suggest the IB cells are ideally situated for both the pooling of evidence and encoding the beginning (and hence timing) of a sequence element.

    “This layer 2/3 input, plus the apical dendrite extending to layer 1, and the bursting behavior, suggest the IB cells are ideally situated for both the pooling of evidence and encoding the beginning (and hence timing) of a sequence element.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    LaBerge's research provides a summary of arguments for the hypothesis that a recurrent cortico-thalamo-cortical circuit as proposed here can provide stable levels of modulatory activity at the soma of cortical pyramidal neurons that can persist over extended periods of time.

    “LaBerge's research [55] provides a summary of arguments for the hypothesis that a recurrent cortico-thalamo-cortical circuit as proposed here can provide stable levels of modulatory activity at the soma of cortical pyramidal neurons that can persist over extended periods of time.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [55] D LaBerge, 2005, Sustained attention and apical dendrite activity in recurrent circuits., Brain Research Reviews source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisereasoning

    Non-specific thalamic cells were recommended for the timing role because they have the required connectivity, receiving input from layer 5 cells and projecting broadly back to layer 1.

    “Non-specific thalamic cells were recommended for this role because they have the required connectivity, receiving input from layer 5 cells and projecting broadly back to layer 1.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    The feedback message sent to one child is not the same as the feedback message sent to the other child.

    “The feedback message sent to one child is not the same as the feedback message sent to the other child.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    RS cells in layer 5 can be thought of as belief cells that anticipate the occurrence of the belief, whereas IB cells represent the same belief at a precise time point.

    “The RS cells in layer 5 can be thought of as belief cells that 'anticipate' the occurrence of the belief, whereas the IB cells represent the same belief at a precise time point.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: There is a need for a neuron representing a belief in a coincidence pattern that does not incorporate precise timing information.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    An alternative implementation is for these neurons to be located in layer 2 of the respective child nodes.

    “An alternative implementation is for these neurons to be located in layer 2 of the respective child nodes.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    Some layer 6 cortical circuits already identified by other neuroscientists as possible candidates for the gating of feedforward activation (control of attention) are compatible with our model.

    “However, it is worth mentioning that some of the layer 6 cortical circuits already identified by other neuroscientists as possible candidates for the gating of feedforward activation (control of attention) [56]–[59] are compatible with our model.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [56] R Guillery, 1998, Paying attention to the thalamic reticular nucleus., Trends in Neurosciences source

      Not checked in the free run.

    • cites [57] SM Sherman, 2002, The role of the thalamus in the flow of information to the cortex., Philos Trans R Soc Lond B Biol Sci source

      Not checked in the free run.

    • cites [58] AM Sillito, 2006, Always returning: feedback and sensory processing in visual cortex and thalamus., Trends Neurosci source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    These feedback connections target the distal dendrites of relay cells and also contact inhibitory interneurons.

    “These feedback connections target the distal dendrites of relay cells and also contact inhibitory interneurons.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [56] R Guillery, 1998, Paying attention to the thalamic reticular nucleus., Trends in Neurosciences source

      Not checked in the free run.

    • cites [57] SM Sherman, 2002, The role of the thalamus in the flow of information to the cortex., Philos Trans R Soc Lond B Biol Sci source

      Not checked in the free run.

    • cites [58] AM Sillito, 2006, Always returning: feedback and sensory processing in visual cortex and thalamus., Trends Neurosci source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S1 (asserted), S4 (based).

    Interpretationreasoning

    For the conjecture that non-specific thalamic cells generate a variable timing signal to be true, we would expect to see something like a cascade of non-specific cells become active in sequence in response to a burst on a layer 5 IB cell, lasting for several hundred milliseconds.

    “For this conjecture to be true we would expect to see something like a cascade of non-specific cells become active in sequence in response to a burst on a layer 5 IB cell. The cascade would last for several hundred milliseconds.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationreasoning

    These signals need to be passed through control mechanisms that will maintain the gating while further bottom-up and top-down propagation for the attended to stimuli alter the belief neuron responses.

    “These signals need to be passed through control mechanisms that will maintain the gating while further bottom-up and top-down propagation for the attended to stimuli alter the belief neuron responses.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    These layer 6 neurons are appropriately situated for calculating the feedback messages and their connectivity is consistent with the proposals for other layers.

    “Hence they are appropriately situated for calculating the feedback messages and their connectivity is consistent with our proposals for other layers.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The dark cyan timing neurons have their apical dendrites in layer 1 to access the timing signal.

    “The dark cyan timing neurons have their apical dendrites in layer 1 to access this timing signal.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The timing signal is broadly available in layer 1.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    These layer 5 belief neurons receive inputs from the yellow neurons in layer 2/3.

    “These neurons receive inputs from the yellow neurons in layer 2/3.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationreasoning

    The connections that a set of layer 6 neurons makes to layer 4 and the thalamus are ideally suited for the attention control mechanism outlined for the HTM belief propagation.

    “The connections that a set of layer 6 neurons makes to layer 4 and the thalamus are ideally suited for the attention control mechanism that we outlined for the HTM belief propagation.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    RS cells are known to project to sub-cortical areas like the striatum and the superior colliculus where the anticipation signal could be used to anticipate actions.

    “The RS cells are known to project to sub-cortical areas like the striatum and the superior colliculus [42] where the anticipation signal could be used to anticipate actions.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [42] AM Thomson, 2003, Interlaminar connections in the neocortex., Cerebral cortex (New York, NY : 1991)

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    The layer 2/3 circuit described can be thought of as a minimal set of circuits needed for temporal inference on multiple scales.

    “The layer 2/3 circuit that we described can be thought of as a minimal set of circuits that are needed for temporal inference on multiple scales.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationcitation

    The timing loop requires the projection of the IB neurons to an external timing circuit.

    “The timing loop requires the projection of the IB neurons to an external timing circuit.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    Humans have the ability to speed up and slow down memorized sequences during recall, which suggests a partially centralized mechanism that can influence the rate of recall over multiple elements in a sequence.

    “Humans also have the ability to speed up and slow down memorized sequences during recall, which suggests a partially centralized mechanism that can influence the rate of recall over multiple elements in a sequence.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Humans have the ability to speed up and slow down memorized sequences during recall.

  • May combine two claims.

    Interpretationnone given

    The result of this paper, the mapping of a hierarchical belief propagation model onto cortical anatomy, does not depend on how duration is encoded, but it does require it is encoded by some means that can handle time scales varying in orders of magnitude.

    “Again the result of this paper, the mapping of a hierarchical belief propagation model onto cortical anatomy, does not depend on how duration is encoded, but it does require it is encoded by some means that can handle time scales varying in orders of magnitude.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S1 (asserted).

    Premisecitation

    The observed nonlinear summation is attributable to intrinsic membrane properties of the pyramidal cells or synaptic properties of the inputs, rather than properties of the global neuronal circuitry.

    “They concluded that this indicated multiplicative supralinear summation of EPSPs caused by simultaneous activation of long range horizontal and vertical inputs under depolarized conditions, and suggested that the observed nonlinear summation is attributable to the intrinsic membrane properties of the pyramidal cells or the synaptic properties of the inputs, rather than the properties of the global”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [49] Y Yoshimura, 2000, Properties of horizontal and vertical inputs to pyramidal cells in the superficial layers of the cat visual cortex., Journal of Neuroscience source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    A layer 6 neuron that receives input from layer 5 belief cells is ideally situated for the attention control purpose.

    “A layer 6 neuron that receives input from layer 5 belief cells is ideally situated for this purpose.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    These connections will need to be learned through the simultaneous activation of the bottom-up outputs from children with top-down outputs from the parent.

    “These connections will need to be learned through the simultaneous activation of the bottom-up outputs from children with top-down outputs from the parent.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Interpretationreasoning

    The layer 1 synapses of layer 5 IB cells appear to be in a more efficient and logical location for storing duration in that these synapses are directly connected to the beliefs being calculated in layer 5.

    “However, the layer 1 synapses of layer 5 IB cells appear to be in a more efficient and logical location for storing duration in that these synapses are directly connected to the beliefs being calculated in layer 5.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S1 (asserted).

    Premisecitation

    Projections of layer 4 spiny neurons to layer 2/3 pyramidal neurons act as a gate for the lateral spread of excitation in layer 2/3.

    “Another study [50] suggested that the projections of layer 4 spiny neurons to layer 2/3 pyramidal neurons act as a gate for the lateral spread of excitation in layer 2/3.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [50] D Feldmeyer, 2002, Synaptic connections between layer 4 spiny neurone-layer 2/3 pyramidal cell pairs in juvenile rat barrel cortex: physiology and anatomy of interlaminar signalling within a cortical column., The Journal of Physiology source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The connections through the matrix thalamus have also been proposed as a mechanism for thalamo-cortical synchrony.

    “The connections through the matrix thalamus have also been proposed as a mechanism for thalamo-cortical synchrony [54].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [54] E Jones, 2001, The thalamic matrix and thalamocortical synchrony., Trends in Neurosciences source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Cells in layer 6 of V1 provide extensive feedback connections to the LGN of the thalamus.

    “According to these studies, cells in layer 6 of V1 provide extensive feedback connections to the LGN of the thalamus.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [56] R Guillery, 1998, Paying attention to the thalamic reticular nucleus., Trends in Neurosciences source

      Not checked in the free run.

    • cites [57] SM Sherman, 2002, The role of the thalamus in the flow of information to the cortex., Philos Trans R Soc Lond B Biol Sci source

      Not checked in the free run.

    • cites [58] AM Sillito, 2006, Always returning: feedback and sensory processing in visual cortex and thalamus., Trends Neurosci source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Both green and yellow neurons in Figure 9 take part in the sequence memory circuits, with yellow neurons receiving feedback information as well.

    “Both the green neurons and the yellow neurons in Figure 9 take part in these circuits, with the yellow neurons receiving feedback information as well.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Horizontal, trans-columnar connections target other layer 2/3 pyramidal cells and correspond to the sequence memory circuits described above.

    “The horizontal, trans-columnar connections target other layer 2/3 pyramidal cells [52],[53] and correspond to the sequence memory circuits that were described above.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [52] D Feldmeyer, 2006, Efficacy and connectivity of intracolumnar pairs of layer 2/3 pyramidal cells in the barrel cortex of juvenile rats., J Physiol source

      Not checked in the free run.

    • cites [53] C Holmgren, 2003, Pyramidal cell communication within local networks in layer 2/3 of rat neocortex., J Physiol source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationnone given

    If intrinsic properties of neurons are not adequate to represent the longer time scales required by the model, it can be achieved via additional network mechanisms.

    “If intrinsic properties of neurons are not adequate to represent the longer time scales required by our model, it can be achieved via additional network mechanisms.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The model requires that sequences at higher levels of the hierarchy represent longer durations of time.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    In the alternative implementation, the higher-level node can send the same feedback signal to all the child nodes.

    “This implementation has the advantage that the higher-level node can send the same feedback signal to all the child nodes.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationnone given

    In the mapping in Figure 9, RS cells are the light-cyan colored neurons in layer 5 and IB cells are the dark-cyan colored neurons in layer 5 with their apical dendrites in layer 1.

    “In our mapping in Figure 9, the RS cells are the light-cyan colored neurons in layer 5. The IB cells are the dark-cyan colored neurons in layer 5 with their apical dendrites in layer 1.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    The duration a human can store, up to about a second, is too long for typical process delay times in neurons, suggesting the existence of a separate duration encoding mechanism.

    “This is too long for typical process delay times in neurons, suggesting the existence of a separate duration encoding mechanism.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: A human can store specific durations, such as duration of notes in music, that last up to about a second.

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing).

    Premisecitation

    Hawkins has proposed the projections of IB cells to the non-specific thalamus as the mechanism for generating a variable timing signal.

    “Hawkins [7] has proposed the projections of IB cells to the non-specific thalamus as the mechanism for generating a variable timing signal.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [7] J Hawkins, 2004, On Intelligence

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    The activation of the global hypothesis, at areas V2 and above, in turn constrains the activity of lower level neurons through the feedback messages.

    “The activation of this global hypothesis, at areas V2 and above, in turn constrains the activity of lower level neurons through the feedback messages.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [1] TS Lee, 2003, Hierarchical Bayesian inference in the visual cortex., Journal of the Optical Society of America

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    The attention control mechanism can be thought of as analogous to the operator proposed by Pearl to model the effect of actions in causal Bayesian networks.

    “the attention control mechanism can also be thought of as analogous to the operator proposed by Pearl [74] to model the effect of actions in causal Bayesian networks.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [74] J Pearl, 2000, Causality : Models, Reasoning, and Inference

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    In domains where a node is exposed to data with rich characteristics, the model would require modifications to include sparse-distributed representations within an HTM node.

    “In domains where a node is exposed to data that has rich characteristics, the model would require modifications to include sparse-distributed representations within an HTM node.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given

    In belief propagation, the messages required for attention control are different from standard feedback messages; attention control messages instantiate variables at intermediate levels and affect feed-forward propagation, whereas standard feedback messages do not interact with feed-forward messages.

    “In belief propagation, the messages required for attention control are different from those of standard feedback messages. The attention control messages instantiate variables at intermediate levels and therefore affect the results of feed-forward propagation, whereas the standard feedback messages in belief propagation do not interact with feed-forward messages.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The mapping of the model to cortical anatomy is sufficiently constrained to suggest its validity and provide testable predictions, but not so constrained to forbid useful variations in its biological implementation.

    “However, we suggest that the mapping of our model to cortical anatomy is sufficiently constrained to suggest its validity and provide testable predictions, but not so constrained to forbid useful variations in its biological implementation.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Excitatory cells connect long distance in both vertical and lateral dimensions and their activity is molded by local inhibitory neurons.

    “It is the excitatory cells that connect long distance in both vertical and lateral dimensions and their activity is then molded by local inhibitory neurons [15].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [15] RJ Douglas, 2004, Neuronal circuits of the neocortex., Annual Review of Neuroscience source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Guillery and Sherman found that layer 5 pyramidal cells that project to the pulvinar of the thalamus also project to motor structures.

    “Guillery and Sherman [75] found that the layer 5 pyramidal cells that project to the pulvinar of the thalamus also project to motor structures.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [75] RW Guillery, 2002, The thalamus as a monitor of motor outputs., Philos Trans R Soc Lond B Biol Sci source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Hypothesis-driven attention is an important aspect of perception and plays an important role in belief propagation.

    “Hypothesis-driven attention is an important aspect of perception and plays an important role in belief propagation as well [29].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [29] J Pearl, 1988, Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S1 (asserted).

    Premisecitation

    The presented stimulus, according to the statistics of the visual world, is adequate to create a high-level hypothesis of the rectangle even though the edges are missing.

    “Their argument was that the presented stimulus, according to the statistics of the visual world, is adequate to create a high-level hypothesis of the rectangle even though the edges are missing.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [1] TS Lee, 2003, Hierarchical Bayesian inference in the visual cortex., Journal of the Optical Society of America

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Inhibitory cells are needed to implement learning.

    “Inhibitory cells are needed to implement learning.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    Variations and omissions in cortical architecture are not necessarily at odds with the model presented in the paper.

    “These variations and omissions are not necessarily at odds with the model presented in this paper.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationreasoning

    Learning these connections requires mechanisms that support competition, inhibition and online learning.

    “Learning these connections requires mechanisms that support competition, inhibition and online learning.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    The six-layered cortical architecture described is most typical of sensory regions of cortex.

    “The six-layered cortical architecture we have described so far is most typical of sensory regions of cortex.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    The responses to illusory contours in area V2 occurred earlier, at 70 milliseconds in the superficial layers and at 95 milliseconds in the deep layers.

    “The responses to illusory contours in area V2 occurred earlier, at 70 milliseconds in the superficial layers and at 95 milliseconds in the deep layers.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [64] TS Lee, 2001, Dynamics of subjective contour formation in the early visual cortex., Proc Natl Acad Sci U S A source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    The population averaged response to illusory contours emerged 100 milliseconds after stimulus onset in the superficial layers of V1 and at approximately 120 to 190 millisecond in the deep layers.

    “The summary of their findings is that the population averaged response to illusory contours emerged 100 milliseconds after stimulus onset in the superficial layers of V1 and at approximately 120 to 190 millisecond in the deep layers.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [64] TS Lee, 2001, Dynamics of subjective contour formation in the early visual cortex., Proc Natl Acad Sci U S A source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Lee and Mumford suggested that the subjective contour effect could be the result of Bayesian computations.

    “In [1], Lee and Mumford suggested this could be the result of Bayesian computations.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [1] TS Lee, 2003, Hierarchical Bayesian inference in the visual cortex., Journal of the Optical Society of America

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    These experiments show how top-down propagation in the current model can be used for segmentation, for the assignment of border-ownership, and for the binding of features corresponding to a top-level hypothesis.

    “These experiments show how top-down propagation in the current model can be used for segmentation, for the assignment of border-ownership, and for the 'binding' of features corresponding to a top-level hypothesis [62].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [62] A Bartels, 2009, Visual Perception: Converging Mechanisms of Attention, Binding, and Segmentation?, Current Biology source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationown data

    The cortical circuit described in the paper can be used for studying and modeling physiological phenomena.

    “The cortical circuit described in this paper can be used for studying and modeling physiological phenomena.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Most inter-laminar and long-distance connections within a cortical area are provided by spiny excitatory neurons, whereas smooth inhibitory interneurons more prevalently connect locally within their layer of origin.

    “most of the inter-laminar and long distance connections within a cortical area are provided by the spiny excitatory neurons, whereas the smooth inhibitory interneurons more prevalently connect locally within their layer of origin.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [15] RJ Douglas, 2004, Neuronal circuits of the neocortex., Annual Review of Neuroscience source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Not all functions of the belief propagation equations have to be implemented exclusively in one layer.

    “Not all functions of the belief propagation equations have to be implemented exclusively in one layer.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    Some forms of attention can be considered as an internal motor action because the attention control mechanism activates parts of the network and blocks other parts.

    “Some forms of attention can also be considered as an internal motor action because the attention control mechanism activates parts of the network and blocks the other parts.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    Inhibitory mechanisms are required for competition during learning.

    “Inhibitory mechanisms are required for competition during learning.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationnone given

    The HTM nodes described in the paper are shown as discrete entities with abrupt boundaries, which does not correspond to biology where overlapping receptive fields and imprecise boundaries are commonly found.

    “The HTM nodes described in this paper are shown as discrete entities with abrupt boundaries, which does not correspond to biology where overlapping receptive fields and imprecise boundaries are commonly found.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    A layer 4 neuron that projects to layer 3 can do the same computation even if moved to layer 3 and receives direct bottom-up input from the thalamus.

    “a layer 4 neuron that projects to layer 3 can do the same computation even if that neuron is moved to layer 3 and if it receives direct bottom-up input from the thalamus.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The finding that illusory contour response occurs first in superficial then deep layers is consistent with the cortical circuit model because feedback information first reaches the Markov chain neurons in layer 2/3 and then is integrated into the layer 5 neurons.

    “This is also consistent with the cortical circuit model in Figure 9 because the feedback information first reaches the Markov chain neurons in layer 2/3 (the yellow neurons in Figure 9) and then is integrated into the layer 5 neurons.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    If the model required a precise and unwavering mapping onto many unique cell types and their connections, it is unlikely such a system could evolve.

    “If our model required a precise and unwavering mapping onto many unique cell types and their connections it is unlikely such a system could evolve.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Lee and Mumford suggested that the formation of illusory contours is primarily a top-down mechanism, in combination with lateral mechanisms.

    “Lee and Mumford [1] suggested that the formation of illusory contours is primarily a top-down mechanism, as also suggested in our experiments, in combination with lateral mechanisms proposed in [66].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [1] TS Lee, 2003, Hierarchical Bayesian inference in the visual cortex., Journal of the Optical Society of America

      Not checked in the free run.

    • cites [66] L Williams, 1997, Stochastic completion fields: A neural model of illusory contour shape and salience., Neural Computation source

      Not checked in the free run.

  • May combine two claims.

    Interpretationnone given

    HTM theory provides a mechanism for training a visual cortical hierarchy and the HTM circuit model gives a detailed anatomical circuit that can be used to test the hypothesis that top-down feedback generates illusory contours.

    “The HTM theory provides a mechanism for training a visual cortical hierarchy and the HTM circuit model gives a detailed anatomical circuit that can be used to test this hypothesis.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The thalamus plays an important role in cortico-cortical communication, acting as a dynamic control of information passed from one cortical area to another.

    “It is known that thalamus plays an important role in cortico-cortical communication, acting as a dynamic control of information passed from one cortical area to another [57],[73].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [57] SM Sherman, 2002, The role of the thalamus in the flow of information to the cortex., Philos Trans R Soc Lond B Biol Sci source

      Not checked in the free run.

    • cites [73] BA Olshausen, 1993, A neurobiological model of visual attention and pattern recognition based on dynamic routing of information., Journal of Neuroscience source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationown data

    Preliminary investigations indicate that relaxing the synchronous child-node assumption may require additional communication between hierarchical levels, which may explain the role of some layer 6 cells.

    “Our preliminary investigations indicate that relaxing this assumption may require additional communication between hierarchical levels which may explain the role of some of the layer 6 cells.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Child-node states change synchronously.

  • May combine two claims.

    Interpretationreasoning

    A similar degree of freedom exists between different levels of the hierarchy; for example, a neuron in layer 6 that sends feedback information to layer 2 of a child region can be moved to layer 2 of the child region.

    “A similar degree of freedom exists between different levels of the hierarchy. For example, a neuron in layer 6 that sends feedback information to layer 2 of a child region can actually be moved to layer 2 of the child region.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    These variations do not violate the computational principles described in the paper.

    “These variations do not violate the computational principles we described”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The work done, including that in the paper, suggests HTM is a good starting point for a biologically grounded neocortical model.

    “The work we have done, including that in this paper, suggests HTM is a good starting point for such a biologically grounded neocortical model.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    The circuit derived in the paper provides a template to explore such variations.

    “The circuit derived in this paper provides a template to explore such variations.”
    Quote in the paperNot a substantial claim
139 of 139

  • The authors state this as an assumption.

    The equation for the output of a circle neuron assumes a simple step-wise notion of time.

    “(Again, the above equation assumes a simple step-wise notion of time which is insufficient for encoding duration and for non-discrete time problems.”

  • The authors state this as an assumption.

    The cortical circuits implement the HTM belief propagation equations described in Table 1.

    “Our working hypothesis can be stated simply: The cortical circuits implement the HTM belief propagation equations described in Table 1.”
    “Under the assumption that the cortical circuits are implementing these equations, what remains to be explained is how the abstract neuronal implementation of the previous section is physically organized in the layers and columns of actual cortical anatomy.”

  • The authors state this as an assumption.

    A timing signal is available at layer 1.

    “In addition, the availability of a timing signal at layer 1 is assumed.”

  • The authors state this as an assumption.

    Durations are signaled to a node by an external timing unit that determines the rate of change of the signals using some system-level measurements.

    “Instead, we assume that durations are signaled to a node by an external timing unit that determines the rate of change of the signals using some system-level measurements.”

  • The authors state this as an assumption.

    In real brains the basic computational column will consist of many redundant cells bound together using common input and short-range intra-laminar connections.

    “Therefore, we assume that in real brains the basic computational column will consist of many redundant cells bound together using common input and short-range intra-laminar connections resulting in a column as shown in Figure 10(B) [41].”

  • The authors state this as an assumption.

    The green neurons and the yellow neurons in Figure 9 encode states within sequences.

    “The green neurons and the yellow neurons in Figure 9 correspond to this description and are assumed to encode states within sequences.”

  • The paper relies on this without support.

    The number of non-exclusive patterns that needs to be maintained decreases as you ascend in the hierarchy.

    “Typically, the number of non-exclusive patterns that needs to be maintained decreases as you ascend in the hierarchy.”

  • The paper relies on this without support.

    The model requires that sequences at higher levels of the hierarchy represent longer durations of time.

    “Our model requires that sequences at higher levels of the hierarchy represent longer durations of time.”

  • The authors state this as an assumption.

    Cortical circuits store duration of individual elements within sequences and the mechanism used to store duration must be present in many if not all cortical areas.

    “Our model makes the assumption that cortical circuits store duration of individual elements within sequences and that the mechanism used to store duration must be present in many if not all cortical areas.”

  • The paper relies on this without support.

    A human can store specific durations, such as duration of notes in music, that last up to about a second.

    “Further, a human can store specific durations, such as duration of notes in music, that last up to about a second.”

  • The paper relies on this without support.

    Humans have the ability to speed up and slow down memorized sequences during recall.

    “Humans also have the ability to speed up and slow down memorized sequences during recall, which suggests a partially centralized mechanism that can influence the rate of recall over multiple elements in a sequence.”

  • The paper relies on this without support.

    Duration information must be available over broad areas of cortex so that duration information can be associated between any subsequent elements in a Markov chain.

    “Duration information must also be available over broad areas of cortex so that duration information can be associated between any subsequent elements in a Markov chain.”

  • The authors state this as an assumption.

    The timing signal is broadly available in layer 1.

    “The timing signal, assumed to be broadly available in layer 1, is shown as blue colored axons.”

  • The authors state this as an assumption.

    Belief-timing neurons project to non-specific thalamic regions (the thalamic matrix) which implement a variable delay mechanism that projects back to layer 1 to complete a timing loop.

    “It is assumed that the belief-timing neurons project to non-specific thalamic regions (the thalamic matrix) [54] which implement a variable delay mechanism that projects back to layer 1 to complete a timing loop, as shown in Figure 7.”

  • The paper relies on this without support.

    There is a need for a neuron representing a belief in a coincidence pattern that does not incorporate precise timing information.

    “Introspection leads us to believe that there is indeed the need for such a neuron.”

  • The paper relies on this without support.

    Layer 3 pyramidal cells that connect to layer 5 cells have their apical dendrites in layer 1.

    “It is also known that layer 3 pyramidal cells that connect to layer 5 cells have their apical dendrites in layer 1.”

  • The paper relies on this without support.

    There are primarily two kinds of pyramidal neurons in layer 5 of the cortex: regular-spiking (RS) neurons and intrinsically bursting (IB) neurons.

    “There are primarily two kinds of pyramidal neurons in layer 5 of the cortex. The first type are called 'regular-spiking' (RS) neurons and the second type are called 'intrinsically bursting' (IB) neurons.”

  • The paper relies on this without support.

    IB cells are larger, extend apical dendrites into layer 1, and exhibit a burst of action potentials when they become active.

    “The IB cells are larger, they extend apical dendrites into layer 1, and as their name suggests they exhibit a burst of action potentials when they become active.”

  • The paper relies on this without support.

    RS cells are smaller, their apical dendrites are mostly in superficial layer 4, and they exhibit a stream of action potentials when active.

    “The RS cells are smaller, their apical dendrites are mostly in superficial layer 4, and they exhibit a stream of action potentials when active.”

  • The paper relies on this without support.

    IB neurons of layer 5 also project to sub-cortical areas and to motor areas.

    “The IB neurons of layer 5 also project to sub-cortical areas and to motor areas.”

  • The paper relies on this without support.

    The cerebellum is known to encode specific timing and its connectivity to the thalamus suggests it could play a role in encoding duration.

    “For example, the cerebellum is known to encode specific timing and its connectivity to the thalamus suggests it could play this role.”

  • The paper relies on this without support.

    Caltech-101 images have low intra-category variation, and most images are centered and approximately of the same size.

    “Caltech-101 images have low intra-category variation. Most of the images are centered and approximately of the same size.”

  • The authors state this as an assumption.

    Excitatory neurons provide the prominent information-processing pathway and inhibitory neurons largely play a supporting role in implementing the learning algorithms.

    “we have adopted the common assumption that excitatory neurons provide the prominent information-processing pathway and that inhibitory neurons largely play a supporting role in implementing the learning algorithms.”

  • The authors state this as an assumption.

    Child-node states change synchronously.

    “The belief propagation messages in this paper were derived under the simplifying assumption that child-node states change synchronously.”

  • The authors state this as an assumption.

    A node represents a set of mutually exclusive hypotheses.

    “relaxing the assumption that a node represents a set of mutually exclusive hypotheses.”

  • The authors state this as an assumption.

    The HTM model assumes a hierarchy of nodes where each node learns spatial coincidences and then learns a mixture of Markov models over the set of coincidences.

    “The model, called Hierarchical Temporal Memory (HTM), is a type of Bayesian network which assumes a hierarchy of nodes where each node learns spatial coincidences and then learns a mixture of Markov models over the set of coincidences.”
26 of 26

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • This coding scheme is Bayesian in implementing the kind of model selection involved in hierarchical hidden Markov models, or hierarchical Kalman filtering.

    “This coding scheme is Bayesian in implementing the kind of model selection involved in hierarchical hidden Markov models (George and Hawkins, 2009), or hierarchical Kalman filtering.”

    From An Integrated World Modeling Theory (IWMT) of Consciousness (2020) · cites this paper as George and Hawkins, 2009

  • Biologically plausible theories are available for how cortex could learn using prediction errors by exploiting temporal differences or top-down feedback.

    “and biologically plausible theories are available for how cortex could learn using prediction errors by exploiting temporal differences (O'Reilly et al., 2014b) or top-down feedback (George and Hawkins, 2009).”

    From Toward an integration of deep learning and neuroscience (2016) · cites this paper as George and Hawkins, 2009

  • We also know that objects tend to undergo predictable sequences of transformations, and it is possible to build this assumption into unsupervised neural learning systems.

    “We also know that objects tend to undergo predictable sequences of transformations, and it is possible to build this assumption into unsupervised neural learning systems (George and Hawkins, 2009).”

    From Toward an integration of deep learning and neuroscience (2016) · cites this paper as George and Hawkins, 2009

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Hardware-implementation citation points to an object-categorization paper

    Minor

    The text cites reference 19 for hardware implementations and parallel architectures, but reference 19 is titled as an object-categorization paper. The reconfigurable-logic implementation is listed as reference 17.

  • Object-categorization citation points to a hardware paper

    Minor

    The text cites reference 17 for object categorization, but reference 17 is titled as a reconfigurable-logic implementation. The object-categorization paper is listed as reference 19.

2 of 2

Cites

Flagged first, then by citations

2 of the 81 references couldn’t be matched to a known paper.

Research works printed but not linked: 1

  1. RPN Rao, 2005, Hierarchical Bayesian Inference in Networks of Spiking Neurons, MIT Press., 1113

Cited by

Sorted by citations

Related

Authors
George, D., & Hawkins, J.
ORCID
Hawkins, Jeff
Venue
PLoS Computational Biology
Publisher
Public Library of Science (PLoS)
Published
2009-10-09
Volume
5
Issue
10
Pages
e1000532
ISSN
  • 1553-734X (ISSN-L)
  • 1553-7358
DOI
10.1371/journal.pcbi.1000532
PMC
PMC2749218
OpenAlex
W2010124835
Semantic Scholar
2f53caf8443a28ea47f7dafe422c06d132b253de
Cited by
404
References
81
Citations by year
2026: 7 · 2025: 17 · 2024: 13 · 2023: 15 · 2022: 13 · 2021: 24 · 2020: 29 · 2019: 25 · 2018: 28 · 2017: 28 · 2016: 31 · 2015: 24 · 2014: 35 · 2013: 22 · 2012: 34
FWCI
14.68
Type
journal article
Language
en
Open access
Licence
Locations
  • PLoS Computational Biology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Figshare (submittedVersion, cc-by)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural dynamics and brain function
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Memory and Neural Mechanisms
  • Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Applications
Keywords
hierarchical temporal memory, hierarchical Bayesian inference, coincidence detectors, cortical anatomy, pattern recognition, cortical computation
MeSH terms
Artificial Intelligence, Bayes Theorem, Cerebral Cortex, Feedback, Markov Chains, Memory, Models, Neurological, Pattern Recognition, Automated, Pyramidal Cells

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review