Canonical Microcircuits for Predictive Coding

In the paper

“recent findings suggest that the superficial layers of cortex show neuronal synchronization and spike-field coherence predominantly in the gamma frequencies, while deep layers prefer lower (alpha or beta) frequencies (Roopun et al., 2006, 2008; Maier et al., 2010; Buffalo et al., 2011).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Roopun et al., 2006, 2008
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Maier et al., 2010
• Supported

  The Results (Coherence Differences Across Layers) report strong gamma-band spike-field coherence in superficial layers of V1, V2, and V4. Deep-layer coherence instead peaked around 10 Hz, within a measured band extending from alpha into low beta. (LLM, conf=0.91)

  cites Buffalo et al., 2011 source

In the paper

“the authors showed that deep L6 principal cells inhibited their extrinsic targets in the LGN and their intrinsic targets in cortical layers 2 to 5 (Olsen et al., 2012).”

What the sources say

• Supported

  The paper reports that photo-stimulation of excitatory L6 neurons in V1 suppressed visually evoked responses in layers 2/3, 4 and 5. It also reports that photo-stimulating L6 in V1 reduced visually evoked activity in dLGN relay neurons. Both passages are labelled “Abstract” in our copy; page numbers are not provided. (LLM, conf=0.94)

  cites Olsen et al., 2012 source

In the paper

“A case in point is the LGN to V1 connection: although it is only the sixth strongest connection to V1, LGN afferents have a substantial effect on V1 responses (Markov et al., 2011).”

What the sources say

• Partly supported

  The Results (Interareal Connectivity Profiles; page unavailable) describe a small LGN input to V1, and the Discussion (Connectivity Profiles: Local Versus Long Distance, and Subcortical; page unavailable) reports an LGN-to-V1 FLNt of 0.16% and discusses amplification of sparse thalamic signals by local circuits. Our copy does not establish that LGN is specifically the sixth strongest input or directly measure its effect on V1 responses. (LLM, conf=0.83)

  cites Markov et al., 2011 sourceExperimental — accuracy not yet measured

In the paper

“This suppression was powerful – in the LGN, visual responses were suppressed by 76%. Suppression was also high in V1, around 80-84% (Olsen et al., 2012).”

What the sources say

• Supported

  The paper reports a 76±4% reduction in visually evoked dLGN activity and reductions of 81±5%, 84±3%, and 84±3% in V1 layers 2/3, 4, and 5, respectively. Both results followed L6 photo-stimulation at the strongest illumination (passages under “Gain control by layer six in cortical circuits of vision > Abstract”; page not provided). (LLM, conf=0.94)

  cites Olsen et al., 2012 source

In the paper

“Driving connections tend to show a strong ionotropic component in their synaptic response, evoke large EPSPs, and respond to multiple EPSPs with depressing synaptic effects. Modulatory connections produce metabotropic and ionotropic responses when stimulated, evoke weak EPSPs, and show paired-pulse facilitation (Sherman and Guillery, 1998; 2011).”

What the sources say

• Partly supported

  Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  cites Sherman and Guillery, 1998 sourceExperimental — accuracy not yet measured
• ?Not checked — no full text of the sourcecites Sherman and Guillery, 2011 source

In the paper

“The remaining 5% represent cells giving rise to extrinsic connections, which – although sparse – can be extremely effective in driving their targets.”

What the sources say

• Partly supported

  The Results section (FLN Values: Local, Long Distance, and Subcortical; page unavailable) assigns about 5% of labeled neurons to long-range corticocortical connections, but also reports about 16% from adjoining cortical areas; 5% is therefore not the total for extrinsic connections. The Discussion (Connectivity Profiles: Local Versus Long Distance, and Subcortical; page unavailable) says weak cortical inputs can be robust and reliable and may transmit effectively, but does not establish that these connections are extremely effective at driving their targets. (LLM, conf=0.86)

  cites Markov et al., 2011 sourceExperimental — accuracy not yet measured

In the paper

“As in visual cortex, the strongest connections were intralaminar and the strongest interlaminar connections were the ascending L4 to L2, and descending L3 to L5.”

What the sources say

• Supported

  The Discussion lists the five most significant connections in the mouse C2 barrel column: three intralaminar connections, followed by ascending L4→L2 and descending L3→L5B. The claim's L3→L5 refers to the L5B connection identified in the paper. (LLM, conf=0.92)

  cites Lefort et al., 2009 source

In the paper

“The traditional view holds that feedforward connections are strong and driving, capable of eliciting spiking activity in their targets and conferring classical receptive field properties – the prototypical example being the synaptic connection between LGN and V1 (Sherman and Guillery, 1998). Feedback connections are thought to modulate (extra-classical) receptive field characteristics according to”

What the sources say

• Partly supported

  Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  cites Sherman and Guillery, 1998 sourceExperimental — accuracy not yet measured

In the paper

“These L1 cells are almost all inhibitory and interconnect strongly with each other, via electrical connections and chemical synapses (Chu et al., 2003).”

What the sources say

• Partly supported

  The Abstract says histochemical studies suggest most layer 1 neurons are GABAergic, but it does not establish that almost all are inhibitory. The ‘Electrical coupling and inhibitory connections among LS cells’ section reports electrical coupling in 83% and chemical connections in 26.2% of tested late-spiking cell pairs. Those findings support interconnection within that cell class, not a strong interconnection among layer 1 cells generally. (LLM, conf=0.88)

  cites Chu et al., 2003 sourceExperimental — accuracy not yet measured

In the paper

“In addition, the thalamic reticular nucleus has been proposed to mediate the inhibition that might underlie cross-modal attention or top-down predictions (Yamaguchi and Knight, 1990; Crick, 1984; Wurtz et al., 2011).”

What the sources say

• ?Not checked — no full text of the sourcecites Yamaguchi and Knight, 1990 source
• ?Not checked — no full text of the sourcecites Crick, 1984 source
• Partly supported

  In “Modulation of Thalamic Reticular and Lateral Geniculate Neurons by Attention” (page unavailable), the paper describes inhibitory projections from the thalamic reticular nucleus and a proposed role in visual attention. It does not connect that inhibition to cross-modal attention or top-down predictions; its discussion of the attentional mechanism is limited to visual spatial attention. (LLM, conf=0.84)

  cites Wurtz et al., 2011 sourceExperimental — accuracy not yet measured

In the paper

“the spiking activity of neurons in the superficial layers of visual cortex are more coherent with gamma frequency oscillations in the local field potential, while neurons in deep layers are more coherent with alpha frequency oscillations (Buffalo et al., 2011).”

What the sources say

• Supported

  The Results report that spike-field coherence in superficial layers of visual areas V1, V2, and V4 peaked in the gamma band (about 40–60 Hz). In deep layers, gamma coherence was minimal and coherence peaked around 10 Hz in the alpha range in V1 and V2. No page numbers are supplied for these passages. (LLM, conf=0.94)

  cites Buffalo et al., 2011 source

In the paper

“The corticogeniculate feedback pathway from L6 maintains this segregation, as upper L6 cells preferentially synapse onto parvocellular cells in the LGN, while lower L6 cells target the magnocellular LGN layers (Fitzpatrick et al., 1994; Briggs and Usrey, 2009).”

What the sources say

• ?Not checked — no full text of the sourcecites Fitzpatrick et al., 1994 source
• Supported

  The Discussion (under “Feedforward and feedback pathways in the primate visual system”; page not provided) says upper-tier layer 6 corticogeniculate neurons project to parvocellular LGN layers, while lower-tier neurons project primarily to magnocellular layers. It describes these as anatomically segregated, stream-specific inputs, supporting the cited paraphrase. (LLM, conf=0.92)

  cites Briggs and Usrey, 2009 source

In the paper

“For example, feedback connections could arise directly from cells encoding conditional expectations in supragranular layers. Indeed, there is emerging evidence that feedback connections between proximate hierarchical levels originate from both deep and superficial layers (Markov et al 2011).”

What the sources say

• ?Could not verify

  Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.

  cites Markov et al 2011 sourceExperimental — accuracy not yet measured

In the paper

“This equation says that the contribution of any (angular) frequency ω in the prediction errors (encoded by superficial pyramidal cells) to the expectations (encoded by the deep pyramidal cells) is suppressed in proportion to that frequency (Friston 2008). In other words, high frequencies should be attenuated when passing from superficial to deep pyramidal cells.”

What the sources say

• Partly supported

  The Results section (Neuronal Implementation; page unavailable) identifies superficial pyramidal cells with prediction errors and deep pyramidal cells with expected causes. Its Summary says cells elaborating predictions may have low-pass dynamics relative to error-encoding cells, supporting high-frequency attenuation as a possibility. Our copy does not state that each frequency’s contribution is suppressed in proportion to its angular frequency. (LLM, conf=0.84)

  cites Friston 2008 sourceExperimental — accuracy not yet measured

In the paper

“In terms of implementation, predictive coding is, arguably, the most plausible neurobiological candidate for making these inferences (Srinivasan et al., 1982; Mumford, 1992; Rao and Ballard, 1999).”

What the sources say

• ?Not checked — no full text of the sourcecites Srinivasan et al., 1982 source
• ?Not checked — no full text of the sourcecites Mumford, 1992 source
• ?Not checked — no full text of the sourcecites Rao and Ballard, 1999 source

In the paper

“Since feedforward connections originate predominately from superficial layers and feedback connections from deep layers”

What the sources say

• ?Not checked — no full text of the sourcecites Felleman and Van Essen, 1991 source

In the paper

“Their model contained superficial and deep pyramidal cells with a common pool of inhibitory cells. All three neuronal populations received thalamic drive and were fully interconnected.”

What the sources say

• ?Not checked — no full text of the sourcecites Douglas and Martin, 1991 source

In the paper

“This segregation can only arise from local neuronal computations that are structured and precisely interconnected.”

In the paper

“This notion builds on a long history of proposals that the brain uses internal or generative models to make inferences about the causes of its sensorium (Helmholtz, 1860; Gregory 1968, 1980; Dayan et al., 1995).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Helmholtz, 1860
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Gregory 1968, 1980
• ?Not checked — no full text of the sourcecites Dayan et al., 1995 source

In the paper

“The idea that the brain actively constructs explanations for its sensory inputs is now generally accepted.”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Helmholtz, 1860
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Gregory 1968, 1980
• ?Not checked — no full text of the sourcecites Dayan et al., 1995 source

In the paper

“Their circuit, although based on recordings from cat visual cortex, was also proposed as a basic theme that might be present and replicated, with minor variations, throughout the cortical sheet (Douglas et al., 1989).”

What the sources say

• ?Not checked — no full text of the sourcecites Douglas et al., 1989 source

In the paper

“first, L4 inhibitory connections are more restricted in their lateral extent, relative to other layers.”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Kätzel et al., 2010

In the paper

“This evidence is in line with classical studies of corticogeniculate contributions to length tuning in the LGN, showing that cortical feedback contributes to the surround suppression of feline LGN cells: without feedback, LGN cells are disinhibited and show weaker surround suppression (Murphy and Sillito, 1987; Sillito et al., 1993; but see Alitto and Usrey, 2008).”

What the sources say

• ?Not checked — no full text of the sourcecites Murphy and Sillito, 1987 source
• ?Not checked — no full text of the sourcecites Sillito et al., 1993 source
• ?Not checked — no full text of the sourcecites Alitto and Usrey, 2008 source

In the paper

“Feedforward connections originate largely from superficial pyramidal cells and target L4, while feedback connections originate largely from deep pyramidal cells and terminate outside of L4 (Felleman and van Essen 1991).”

What the sources say

• ?Not checked — no full text of the sourcecites Felleman and van Essen 1991 source

In the paper

“Furthermore, normal subjects show a rapid adaptation to deviant stimuli as they become predictable – an effect not seen in prefrontal patients.”

In the paper

“In short, predictive coding requires feedback connections to drive cells in lower levels in a context sensitive fashion, which necessitates a modulatory aspect to their postsynaptic effects.”

In the paper

“The most frequently connected cells were located in the same cortical layer, where the largest interlaminar projections were the ‘feedforward’ connections from L4 to L3 and from L3 to L5.”

What the sources say

• ?Not checked — no full text of the sourcecites Thomson et al 2002 source

In the paper

“In contrast, they show reduced-amplitude P300 ERPs in response to novel stimuli – as if there were a failure to communicate top-down predictions to sensory cortex (Knight, 1984).”

What the sources say

• ?Not checked — no full text of the sourcecites Knight, 1984 source

In the paper

“Excitatory reciprocal ‘feedback’ connections were not observed (L3 to L4) or less commonly observed (L5 to L3), suggesting that excitation spreads within the column in a feedforward fashion.”

What the sources say

• ?Not checked — no full text of the sourcecites Thomson et al 2002 source

In the paper

“Third, several interlaminar motifs appeared to be general – at least in granular cortex: principally, a strong inhibitory connection from L4 onto supragranular L2/3 and from infragranular layers onto L4.”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Kätzel et al., 2010

In the paper

“The largest fraction of synaptic input arrived in L5A/B, consistent with its key role in accumulating information from a wide range of afferents – before sending its output to the corticospinal tract.”

What the sources say

• ?Not checked — no full text of the sourcecites Weiler et al., 2008 source

In the paper

“Current dogma holds that the cortex is hierarchically organized.”

What the sources say

• ?Not checked — no full text of the sourcecites Felleman and Van Essen, 1991 source

In the paper

“The idea of a cortical hierarchy rests on the distinction between three types of extrinsic connections: feedforward connections, that link an earlier area to a higher area, feedback connections, that link a higher to an earlier area, and lateral connections, that link areas at the same level (reviewed in Felleman and Van Essen, 1991).”

What the sources say

• ?Not checked — no full text of the sourcecites Felleman and Van Essen, 1991 source

In the paper

“in primates, parvocellular input enters the cortex primarily in layer 4Cβ, whereas magnocellular inputs enter in 4Cα.”

In the paper

“Roopun et al. (2006) showed that L2/3 of rat somatomotor cortex shows prominent gamma oscillations that are co-expressed with beta oscillations in L5. Both rhythms persisted, when superficial and deep layers were disconnected at the level of L4.”

What the sources say

• ?Not checked — no full text of the sourcecites Roopun et al. (2006) source

In the paper

“Entropy is the average of surprise over time, which means biological systems minimise the surprise associated with their sensory states at each point in time.”

What the sources say

• ?Not checked — no full text of the sourcecites Friston, 2010 source

In the paper

“The established view is that extrinsic corticocortical connections are exclusively excitatory (using glutamate as their excitatory neurotransmitter) – although recent evidence suggests that inhibitory extrinsic connections exist and may play an important role in synchronizing distant regions (Melzer et al., 2012).”

What the sources say

• ?Not checked — no full text of the sourcecites Melzer et al., 2012 source

In the paper

“Computational modelling studies indicate that cortical columns with structured connectivity are computationally more efficient than a network containing the same number of neurons but with random connectivity (Haeusler and Maass, 2007).”

What the sources say

• ?Not checked — no full text of the sourcecites Haeusler and Maass, 2007 source

In the paper

“Maier et al., found that supragranular sites had higher broadband gamma power than infragranular sites. This pattern was reversed in the alpha and beta range; with greater power in the infragranular and granular layers.”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Maier et al. (2010)

In the paper

“Superficial and deep layers are anatomically interconnected, but mounting evidence suggests that they constitute functionally distinct processing streams”

What the sources say

• ?Not checked — no full text of the sourcecites Roopun et al. (2006) source
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Maier et al. (2010)

In the paper

“However, when the authors used a stimulus that spanned the extra-classical RF the responses of V1 neurons were – on average – enhanced after cooling area V5, consistent with the suppressive role of feedback connections.”

What the sources say

• ?Not checked — no full text of the sourcecites Hupé et al., 1998 source

In the paper

“Maier and colleagues (2010) used multilaminar recordings to show strong LFP coherence amongst sites within the superficial layers (the superficial compartment), as well as strong coherence amongst sites in deep layers (the deep compartment), but weak inter-compartment coherence.”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Maier et al. (2010)

In the paper

“Rao and Ballard (1999) trained a hierarchical predictive coding network to recognise natural images. They showed that higher levels in the hierarchy learn to predict visual features that extend across many CRFs in the lower levels (e.g. tree trunks or horizons).”

What the sources say

• ?Not checked — no full text of the sourcecites Rao and Ballard (1999) source

In the paper

“Similar effects were observed when area V2 was cooled and neurons were measured in V1: when stimuli were presented only to the CRF, cooling V2 decreased V1 spiking activity; however, when stimuli were present in the CRF and the surround, cooling V2 increased V1 activity (Bullier et al., 1996).”

What the sources say

• ?Not checked — no full text of the sourcecites Bullier et al., 1996 source

In the paper

“This accommodates the fact that (in macaque V1) a significant percentage of superficial L2/3 cells (about half) and deep L5 excitatory cells (about 80%) do not project outside the cortical column (Callaway and Wiser, 1996; Briggs and Callaway, 2005).”

What the sources say

• ?Not checked — no full text of the sourcecites Callaway and Wiser, 1996 source
• ?Not checked — no full text of the sourcecites Briggs and Callaway, 2005 source

In the paper

“Having said this, there are many variations on the mapping between computational and neuronal architectures: Even if predictive coding is an appropriate implementation of Bayesian filtering, there are many variations on the arrangement shown in Figure 5.”

In the paper

“The circuitry in Figure 5 appears consistent with the broad scheme of ascending (feedforward) and descending (feedback) intrinsic connections: feedforward prediction errors from a lower cortical level arrive at granular layers and are passed forward to excitatory and inhibitory interneurons in supragranular layers, encoding expectations.”

In the paper

“The remaining populations are the excitatory and inhibitory interneurons in the supragranular layer, to which we assign expectations about hidden causes and states respectively.”

In the paper

“This arrangement accommodates the fact that the dependencies among hidden states are confined to each node (by the nature of graphical models), which means that their expectations and prediction errors should be encoded by interneurons. Furthermore, the splitting of excitatory cells in the upper layers into two populations (encoding expectations and prediction errors on hidden causes) is sensible,”

In the paper

“In summary, predictions are formed by a linear accumulation of prediction errors. Conversely, prediction errors are nonlinear functions of predictions.”

In the paper

“This means that superficial pyramidal cells must encode and broadcast prediction errors on hidden causes ξv(i+1), while deep pyramidal cells must encode conditional expectations (μ~v(i),μ~x(i)) so that they can elaborate feedback predictions.”

In the paper

“This smoothing – inherent in Bayesian filtering – leads to an asymmetry in frequency content of superficial and deep cells: for example, superficial cells should express more gamma relative to beta, and deep cells should express more beta relative to gamma (Roopun et al., 2006, 2008; Maier et al., 2010).”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Roopun et al., 2006, 2008
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Maier et al., 2010

In the paper

“This means that the conversion of prediction errors into predictions (Bayesian filtering) necessarily entails a loss of high frequencies. However, the nonlinearity in the mapping from predictions to prediction errors means that high frequencies can be created (consider the effect of squaring a sine wave, which would convert beta into gamma).”

In the paper

“This leaves the inhibitory interneurons in the granular layer, which, for symmetry, we associate with prediction errors on the hidden states.”

In the paper

“Figure 5 illustrates a remarkable correspondence between the form of Equation (1) and the connectivity of the canonical microcircuit. Furthermore, the resulting scheme corresponds almost exactly to the computational architecture proposed by Mumford (1992).”

What the sources say

• ?Not checked — no full text of the sourcecites Mumford (1992) source

In the paper

“Another key feature of Equation (1) is that the top-down predictions produce prediction errors through subtraction. In other words, feedback connections should exert inhibitory effects, of the sort seen empirically.”

In the paper

“These arguments mean that by minimising surprise, through selecting appropriate sensations, the brain is implicitly maximising the evidence for its own existence – this is known as active inference.”

In the paper

“However, there are two exceptions that both involve connections to the inhibitory cells in the granular layer (shown as dotted lines in Figure 5). Predictive coding requires that these cells (that encode prediction errors on hidden states) compare the expected changes in hidden states with the actual changes. This suggests that there should be interlaminar projections from supragranular (inhibitor”