Number detectors spontaneously emerge in a deep neural network designed for visual object recognition
Open accessThe DOI matches its Crossref record; 45 of the 45 references it cites were found in a registry or an index.
Abstract
A deep neural network trained only on visual object recognition develops tuned number detectors reminiscent of real neurons.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Figures

![Fig. 2 Numerosity-tuned units emerging in the HCNN. (A) Examples of the stimuli used to assess numerosity encoding. Standard stimuli contain dots of the same average radius. Dots in Area & Density stimuli have a constant total area and density across all numerosities. Dots in Shape & Convex hull stimuli have random shapes and a uniform pentagon convex hull (for numerosities >4). (B) Tuning curves for individual numerosity-selective network units. Colored curves show the average responses for each stimulus set. Black curves show the average responses over all stimulus set. Error bars indicate SE measure. PN, preferred numerosity. (C) Same as (B), but for neurons in monkey prefrontal cortex (20). Only the average responses over all stimulus sets are shown. (D) Distribution of preferred numerosities of the numerosity-selective network units. (E) Same as (D), but for real neurons recorded in monkey prefrontal cortex [data from (20)].](https://figures.cochat.ai/paper-figures/aaf715d51e0d474b8c325f99b3eb8af1/number-detectors-spontaneously-emerge-in-a-deep-neural-network-designed-for-figure-2.jpg)

![Fig. 4 Tuning properties of numerosity-selective network units. (A) Left: Average tuning curves for network units preferring each numerosity plotted on a linear scale. Right: Same tuning curves plotted on a logarithmic scale. (B) Average goodness-of-fit measure for fitting Gaussian functions to the tuning curves on different scales [Plinear-log = 0.009; Plinear-pow(1/2) = 0.003; Plinear-pow(1/3) = 0.001]. (C) SD of the best-fitting Gaussian function for each of the tuning curves of numerosity-selective network units for different scales.](https://figures.cochat.ai/paper-figures/627b66a232dc435c8d7b371f03af87de/number-detectors-spontaneously-emerge-in-a-deep-neural-network-designed-for-figure-4.jpg)
![Fig. 5 Relevance of numerosity-selective units to network performance. Average activity of numerosity-selective network units shown as a function of numerical distance between preferred numerosities and sample numerosities in the matching task. Left: Data from network units. Responses were average separately for correct trials (black) and error trials (gray). Responses during error trials are normalized to the maximum average response during correct trials (P = 0.019). Right: Same plot but for real neurons recorded from monkey prefrontal cortex [data from (20)].](https://figures.cochat.ai/paper-figures/ee8c0e1e56fb4bdbaaa7270a62d0c085/number-detectors-spontaneously-emerge-in-a-deep-neural-network-designed-for-figure-5.jpg)
![Fig. 6 Performance of the HCNN model in the numerosity matching task. (A) Left: Performance functions resulting from the discrimination of numerosities plotted on a linear scale. Each curve shows the probability of the model predicting that the sample image contains the same number of items as the test image (peak of the function). Sample numerosity is indicated above each curve. Right: Same performance functions but plotted on a logarithmic scale. (B) Average goodness-of-fit measure for fitting Gaussian functions to the performance tuning curves on different scales [Plinear-log = 0.003; Plinear-pow(1/2) = 0.049; Plinear-pow(1/3) = 0.016]. *P < 0.05. (C) SD of the best-fitting Gaussian function for each of the performance tuning curves for different scales.](https://figures.cochat.ai/paper-figures/f8bf766d1b534ddcaa16bedda91e1a87/number-detectors-spontaneously-emerge-in-a-deep-neural-network-designed-for-figure-6.jpg)
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
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?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Models trained to process images, for example, spontaneously learned how to count (30, 31) and differentially process central and peripheral image areas (32), as well as experience human-like optical illusions (33).”
From Evaluating large language models in theory of mind tasks (2024) · cites this paper as [30]
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 1×
“Here, we use a hierarchical convolutional neural network (HCNN), a class of biologically inspired models that have recently achieved great success in computer vision applications (14, 15) and in the modeling of the ventral visual stream (16, 17).”
MethodImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · cited 4×
“Network units that had the same receptive fields in convolutional layers competed with each other using a simple form of lateral inhibition (14).”
BackgroundMethodInfluentialImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380
(2014) · Cited by 33,419
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Ioffe, S., & Szegedy, C. (2015) · arXiv · doi:10.48550/arxiv.1502.03167 · Cited by 27,818 · cited 1×
“Before applying the activation function, the outputs of the convolution operation were normalized to have zero mean and unit SD [batch normalization; (40)].”
Method(2010) · Cited by 14,134
Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex
Hubel, D. H., & Wiesel, T. N. (1962) · The Journal of Physiology · doi:10.1113/jphysiol.1962.sp006837 · Cited by 13,680 · cited 1×
“The network model was an instance of HCNNs (18), originally inspired by the discovery of simple and complex cells in early visual cortex (19).”
Method(2017) · Cited by 11,625
An overview of gradient descent optimization algorithms
Ruder, S. (2016) · arXiv · doi:10.48550/arxiv.1609.04747 · Cited by 4,962 · cited 1×
“The minimization was performed using mini-batch gradient descent (43) with a batch size of 256 images, a learning rate of 0.”
MethodPerformance-optimized hierarchical models predict neural responses in higher visual cortex
Yamins, D. L. K., Hong, H., Cadieu, C. F., Solomon, E. A., Seibert, D., & DiCarlo, J. J. (2014) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1403112111 · Cited by 1,924
Research works printed but not linked: 3
- S. Dehaene The Number Sense (Oxford Univ. Press 1999).
- Y. LeCun Y. Bengio Convolutional networks for images speech and time-series in The Handbook of Brain Theory and Neural Networks (MIT Press 1995) vol. 3361 pp. 255–258.
- S. Carey The Origin of Concepts: Oxford Series in Cognitive Development (Oxford Univ. Press 2009).
Cited by
Evaluating large language models in theory of mind tasks
Kosinski, M. (2024) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2405460121 · Cited by 339
“Models trained to process images, for example, spontaneously learned how to count ( 30 , 31 ) and differentially process central and peripheral image areas ( 32 ), as well as experience human - like optical illusions ( 33 ).”
BackgroundUsing artificial neural networks to ask ‘why’ questions of minds and brains
Kanwisher, N., Khosla, M., & Dobs, K. (2023) · Trends in Neurosciences · doi:10.1016/j.tins.2022.12.008 · Cited by 182
“Analyses in this vein have revealed the spontaneous emergence of face-selective [88,89] and number-selective units [90,91] in ANNs.”
BackgroundThe Adaptive Value of Numerical Competence
Nieder, A. (2020) · Trends in Ecology & Evolution · doi:10.1016/j.tree.2020.02.009 · Cited by 162
“In support of a putatively innate ANS, artificial number neurons obeying Weber’s and Fechner’s law spontaneously emerged also in a deep network that was merely trained on visual object recognition, never on numbers [133].”
BackgroundSelf-citationBrain-like functional specialization emerges spontaneously in deep neural networks
Dobs, K., Martinez, J., Kell, A. J. E., & Kanwisher, N. (2022) · Science Advances · doi:10.1126/sciadv.abl8913 · Cited by 129
“…then the same should be true for any computational system, including CNNs. Prior work suggests that this may not necessarily be the case: CNNs optimized for generic object recognition transfer well to many other tasks ( 12 – 15 ), including fine-grained discrimination within a category ( 13 ).”
BackgroundEvaluating Large Language Models in Theory of Mind Tasks
Kosinski, M. (2023) · arXiv · doi:10.48550/arxiv.2302.02083 · Cited by 339
“Models trained to process images, for example, spontaneously learned how to count ( 30 , 31 ) and differentially process central and peripheral image areas ( 32 ), as well as experience human - like optical illusions ( 33 ).”
BackgroundStructure learning and the posterior parietal cortex
Summerfield, C., Luyckx, F., & Sheahan, H. (2020) · Progress in Neurobiology · doi:10.1016/j.pneurobio.2019.101717 · Cited by 118
Shared functional specialization in transformer-based language models and the human brain
Kumar, S., Sumers, T. R., Yamakoshi, T., Goldstein, A., Hasson, U., Norman, K. A., Griffiths, T. L., Hawkins, R. D., & Nastase, S. A. (2024) · Nature Communications · doi:10.1038/s41467-024-49173-5 · Cited by 86
Zhao, H., Zhang, Y., Han, L., Qian, W., Wang, J., Wu, H., Li, J., Dai, Y., Zhang, Z., Bowen, C., et al. (2023) · Nano-Micro Letters · doi:10.1007/s40820-023-01216-0 · Cited by 104
“In general, robot vision technologies have made substantial progress in rapidly recognizing objects [2, 6–10], and some have even been commercialized.”
Paul, J. M., van Ackooij, M., ten Cate, T., & Harvey, B. M. (2022) · Nature Communications · doi:10.1038/s41467-022-29030-z · Cited by 92
Zhao, X., Sun, Z., & Lee, C. (2024) · Advanced Functional Materials · doi:10.1002/adfm.202409558 · Cited by 85
Visual number sense in untrained deep neural networks
Kim, G., Jang, J., Baek, S., Song, M. J., & Paik, S. (2021) · Science Advances · doi:10.1126/sciadv.abd6127 · Cited by 79
“Similar to the method used to find number-selective neurons in monkeys (27) and to detect number-selective network units (21), a two-way analysis of variance (ANOVA) with two factors (numerosity and stimulus set) was used.”
InfluentialVisual sense of number vs. sense of magnitude in humans and machines
Testolin, A., Dolfi, S., Rochus, M., & Zorzi, M. (2020) · Scientific Reports · doi:10.1038/s41598-020-66838-5 · Cited by 71
“Although sensitivity to numerosity has been found at the level of individual neurons in deep networks 39,40,54 , with tuning functions resembling those of real neurons in the dorsal visual stream of non-human primates 55 , the modulatory effect of non-numerical features in the response profiles of…”
BackgroundNumber neurons in the nidopallium of young domestic chicks
Kobylkov, D., Mayer, U., Zanon, M., & Vallortígara, G. (2022) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2201039119 · Cited by 63
“Indeed, recent studies have found that in a deep neural network designed to analyse images, a similar amount of number detectors can emerge spontaneously (Nasr et al. 2019), even without pre-training (Kim et al. 2021).”
BackgroundInfluentialModeling personalized cognition of customers in online shopping
Wu, X., & Liao, H. (2021) · Omega · doi:10.1016/j.omega.2021.102471 · Cited by 57
The Evolutionary History of Brains for Numbers
Nieder, A. (2021) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2021.03.012 · Cited by 63
“Irrespective of which territory evolves into the highest integration center (the ventral pallium in birds versus the dorsal pallium in mammals), something about the circuit properties of the pallium invites the representation of abstract categories and concepts [44,45].”
BackgroundSelf-citationWill We Ever Have Conscious Machines?
Krauß, P., & Maier, A. (2020) · Frontiers in Computational Neuroscience · doi:10.3389/fncom.2020.556544 · Cited by 61
“For instance, in deep neural networks trained on visual object recognition, the spontaneous emergence of number detectors (Nasr et al., 2019), solid shape coding (Srinath et al.”
BackgroundRajalingham, R., Kar, K., Sanghavi, S., Dehaene, S., & DiCarlo, J. J. (2020) · Nature Communications · doi:10.1038/s41467-020-17714-3 · Cited by 57
Numerosity and cumulative surface area are perceived holistically as integral dimensions.
Aulet, L. S., & Lourenco, S. F. (2020) · Journal of Experimental Psychology General · doi:10.1037/xge0000874 · Cited by 52
Born Knowing: Imprinting and the Origins of Knowledge
Vallortígara, G., & Losi, C. (2021) · The MIT Press eBooks · doi:10.7551/mitpress/14091.001.0001 · Cited by 50
Sanders, C. A., & Nosofsky, R. M. (2020) · Computational Brain & Behavior · doi:10.1007/s42113-020-00073-z · Cited by 59
“Such CNNs learn representations of data in a hierarchical fashion inspired by the visual cortex and have been shown to spontaneously extract fundamental characteristics associated with perceptual and cognitive processing of natural objects (e.g., Nasr et al. 2019).”
BackgroundBrain-inspired models for visual object recognition: an overview
Yang, X. N., Yan, J., Wang, W., Li, S., Hu, B., & Lin, J. L. (2022) · Artificial Intelligence Review · doi:10.1007/s10462-021-10130-z · Cited by 54
Neuroethology of number sense across the animal kingdom
Nieder, A. (2021) · Journal of Experimental Biology · doi:10.1242/jeb.218289 · Cited by 52
“Third, and much like the behavioural data (Merten and Nieder, 2009), the tuning curves are mildly asymmetrical when plotted on a linear number scale: only when plotted on a logarithmic scale do they become symmetrical (Nieder and Miller, 2003), an effect also emerging in deep-learning networks (Nasr et al., 2019).”
BackgroundSelf-citationAulet, L. S., Chiu, V., Prichard, A., Spivak, M., Lourenco, S. F., & Berns, G. S. (2019) · Biology Letters · doi:10.1098/rsbl.2019.0666 · Cited by 43
Numerosities and Other Magnitudes in the Brains: A Comparative View
Lorenzi, E., Perrino, M., & Vallortígara, G. (2021) · Frontiers in Psychology · doi:10.3389/fpsyg.2021.641994 · Cited by 41
“, 2020), comparison tasks for relationships of sameness (Nasr et al., 2019), or inequality (Testolin et al.”
BackgroundMethodThe role of neural tuning in quantity perception
Tsouli, A., Harvey, B. M., Hofstetter, S., Cai, Y., van der Smagt, M. J., Pas, S. F. T., & Dumoulin, S. O. (2021) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2021.10.004 · Cited by 41
A theory of perceptual number encoding.
Lourenco, S. F., & Aulet, L. S. (2022) · Psychological Review · doi:10.1037/rev0000380 · Cited by 41
Nieder, A. (2024) · Physiological Reviews · doi:10.1152/physrev.00014.2024 · Cited by 41
Self-citationHofstetter, S., Cai, Y., Harvey, B. M., & Dumoulin, S. O. (2021) · Nature Communications · doi:10.1038/s41467-020-20567-5 · Cited by 36
Neural constraints on human number concepts
Nieder, A. (2019) · Current Opinion in Neurobiology · doi:10.1016/j.conb.2019.10.003 · Cited by 33
Self-citationViswanathan, P., & Nieder, A. (2020) · Journal of Cognitive Neuroscience · doi:10.1162/jocn_a_01548 · Cited by 32
“Whether they use a small number of dots or a dense display of numerosity, the relevant layers of neural network models have access to the entire stimulus (Nasr, Viswanathan, & Nieder, 2019; Zorzi & Testolin, 2018; Stoianov & Zorzi, 2012).”
Self-citationAdriano, A., Rinaldi, L., & Girelli, L. (2021) · Journal of Experimental Psychology Human Perception & Performance · doi:10.1037/xhp0000844 · Cited by 30
Emerged human-like facial expression representation in a deep convolutional neural network
Zhou, L., Yang, A., Meng, M., & Zhou, K. (2022) · Science Advances · doi:10.1126/sciadv.abj4383 · Cited by 30
Is there an innate sense of number in the brain?
Lorenzi, E., Kobylkov, D., & Vallortígara, G. (2025) · Cerebral Cortex · doi:10.1093/cercor/bhaf004 · Cited by 27
Park, J., & Huber, D. E. (2022) · eLife · doi:10.7554/elife.80990 · Cited by 31
“Unlike prior models, this 333 eliminated the effect of size in the early visual response, without requiring subsequent pooling 334 layers (Creatore et al., 2021; Kim et al., 2021; Nasr et al., 2019; Stoianov and Zorzi, 2012; 335 Testolin et al., 2020) or a decision making process that compares high versus low spatial…”
BackgroundMethodInfluentialDNNBrain: A Unifying Toolbox for Mapping Deep Neural Networks and Brains
Chen, X. S., Zhou, M. Q., Gong, Z., Xu, W., Liu, X., Huang, T., Zhen, Z., & Liu, J. (2020) · Frontiers in Computational Neuroscience · doi:10.3389/fncom.2020.580632 · Cited by 26
“Finally, DNNs designated for object recognition spontaneously generate many well-known behavioral and neurophysiological signatures of cognitive phenomena such as shape tuning (Pospisil et al., 2018), numerosity (Nasr et al., 2019), and visual illusions (Watanabe et al., 2018).”
BackgroundMistry, P. K., Strock, A., Liu, R., Young, G., & Menon, V. (2023) · Nature Communications · doi:10.1038/s41467-023-39548-5 · Cited by 25
“Contrary to Nasr, et al.7, they proposed that visual object training was not relevant for the initial emergenceofnumerosity.”
ResultAtomic Quantum Technologies for Quantum Matter and Fundamental Physics Applications
Malo, J. Y., Lepori, L., Gentini, L., & Chiofalo, M. L. (2024) · Technologies · doi:10.3390/technologies12050064 · Cited by 31
Karami, A., Castaldi, E., Eger, E., & Piazza, M. (2025) · Communications Biology · doi:10.1038/s42003-025-08395-z · Cited by 22
Categorical representation of abstract spatial magnitudes in the executive telencephalon of crows
Wagener, L., & Nieder, A. (2023) · Current Biology · doi:10.1016/j.cub.2023.04.013 · Cited by 19
“51,52 How learned magnitude categories emerge in the avian brain and the neuronal mechanisms underlying them is currently unknown.”
BackgroundSelf-citationArtificial neural network modelling of the neural population code underlying mathematical operations
Nakai, T., & Nishimoto, S. (2023) · NeuroImage · doi:10.1016/j.neuroimage.2023.119980 · Cited by 20
Hancock, P. A. (2023) · Ergonomics · doi:10.1080/00140139.2023.2236822 · Cited by 18
“The evidence we have seems to be most prominent in certain areas of neuroscientific specialisation (e.g. Mohsenzadeh et al. 2020; Nasr, Viswanathan, and Nieder 2019; Stoianov and Zorzi 2012).”
BackgroundAutomated calibration of 3D-printed microfluidic devices based on computer vision
Wang, J., Liang, K., Zhang, N., Yao, H., Ho, T.-Y., & Sun, L. (2021) · Biomicrofluidics · doi:10.1063/5.0037274 · Cited by 18
“Among all CV techniques, Convolutional Neural Networks (CNNs) have been shown as a powerful tool undertaking image classification and recognition tasks, 30,31 such as medical image analysis, 32,33 object recognition, 34 – 36 action recognition, 37,38 or even face recognition with a mask on it.”
MethodGeneral object-based features account for letter perception
Janini, D., Hamblin, C. J., Deza, A., & Konkle, T. (2022) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1010522 · Cited by 20
“These include investigations into the perception of a wide variety of domains of sensory input including faces [54,81]; objects, scenes, and nearscale “reachspaces” [82]; approximate number [83]; and language and music [84].”
BackgroundSymbolic and non-symbolic representations of numerical zero in the human brain
Barnett, B., & Fleming, S. M. (2024) · Current Biology · doi:10.1016/j.cub.2024.06.079 · Cited by 17
Personalized deep neural networks reveal mechanisms of math learning disabilities in children
Strock, A., Mistry, P. K., & Menon, V. (2025) · Science Advances · doi:10.1126/sciadv.adq9990 · Cited by 18
“DNNs have demonstrated notable success in modeling a variety of cognitive functions including number sense ( 13 , 43 , 44 ), word reading ( 45 ), object recognition ( 46 , 47 ), and sentence processing ( 48 ), yet their application in understanding learning disabilities has been scant, primarily…”
Spontaneous representation of numerosity zero in a deep neural network for visual object recognition
Nasr, K., & Nieder, A. (2021) · iScience · doi:10.1016/j.isci.2021.103301 · Cited by 18
“This indicates a bias toward monotonic units and was in contrast to the distribution of peak-tuned units in our network trained on object classification (Nasr et al., 2019).”
BackgroundMethodResultInfluentialSelf-citationLearning exact enumeration and approximate estimation in deep neural network models
Creatore, C., Sabathiel, S., & Solstad, T. (2021) · Cognition · doi:10.1016/j.cognition.2021.104815 · Cited by 15
“The single-unit response profiles were distinct from the Gaussian-like tuning curves of number-selective neurons re-ported in experimental data (Viswanathan & Nieder, 2013) and models of approximate number perception (Nasr et al., 2019; Zorzi & Testolin, 2018).”
BackgroundMethodResultInfluentialThe Challenge of Modeling the Acquisition of Mathematical Concepts
Testolin, A. (2020) · Frontiers in Human Neuroscience · doi:10.3389/fnhum.2020.00100 · Cited by 19
Attention drives human numerosity-selective responses
Cai, Y., Hofstetter, S., Harvey, B. M., & Dumoulin, S. O. (2022) · Cell Reports · doi:10.1016/j.celrep.2022.111005 · Cited by 14
Non-symbolic numerosity encoding escapes spatial frequency equalization
Adriano, A., Girelli, L., & Rinaldi, L. (2021) · Psychological Research · doi:10.1007/s00426-020-01458-2 · Cited by 14
Visual enumeration remains challenging for multimodal generative AI
Testolin, A., Hou, K., & Zorzi, M. (2025) · PLoS ONE · doi:10.1371/journal.pone.0331566 · Cited by 16
“…rudimentary visual number sense has been shown to emerge in small-scale generative models trained with the goal of reconstructing images with a varying number of items [19, 20] and number-selective neurons have been observed in generic convolutional networks trained for object recognition [21, 22].”
Brain-like functional specialization emerges spontaneously in deep neural networksPreprint
Dobs, K., Martinez, J., Kell, A. J. E., & Kanwisher, N. (2021) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2021.07.05.451192 · Cited by 15
Lorenzi, E., Perrino, M., Messina, A., Zanon, M., & Vallortígara, G. (2024) · Heliyon · doi:10.1016/j.heliyon.2024.e34162 · Cited by 12
“Artificial neural modeling suggests that units tuned to numerousness may emerge spontaneously in deep neural networks without any supervised learning [12].”
BackgroundEffects of spatial frequency cross-adaptation on the visual number sense
Bonn, C. D., & Odic, D. (2023) · Attention Perception & Psychophysics · doi:10.3758/s13414-023-02798-y · Cited by 13
Numerosity representation in a deep convolutional neural network
Zhou, C., Xu, W., Liu, Y., Xue, Z., Chen, R., Zhou, K., & Liu, J. (2021) · Journal of Pacific Rim Psychology · doi:10.1177/18344909211012613 · Cited by 11
“To do so, like previous studies (Kim et al., 2021; Nasr et al., 2019), we first identified the potential number-selective units in the AlexNet, a typical DCNN pre-trained for object classification, by using dot-displays without any connections.”
BackgroundMethodResultInfluentialFundamental units of numerosity estimation
Chakravarthi, R., Nordqvist, A., Poncet, M., & Adamian, N. (2023) · Cognition · doi:10.1016/j.cognition.2023.105565 · Cited by 11
“Such numerosity tracking ‘neurons ’ have been observed to spontaneously develop in convolutional neural networks, whether trained for object recognition or untrained (Kim, Jang, Baek, Song, & Paik, 2021; Nasr, Viswanathan, & Nieder, 2019).”
BackgroundKobylkov, D., Zanon, M., Perrino, M., & Vallortígara, G. (2023) · Biosystems · doi:10.1016/j.biosystems.2023.104999 · Cited by 12
Electrophysiological Signatures of Numerosity Encoding in a Delayed Match-to-Sample Task
Fu, W., Dolfi, S., Decarli, G., Spironelli, C., & Zorzi, M. (2022) · Frontiers in Human Neuroscience · doi:10.3389/fnhum.2021.750582 · Cited by 11
A number sense as an emergent property of the manipulating brain
Kondapaneni, N., & Perona, P. (2024) · Scientific Reports · doi:10.1038/s41598-024-56828-2 · Cited by 10
“Thus, it is likely that multiple visual maps are simultaneously learned, and it is possible that our ‘latent representation’ is shared with other visual modalities [12].”
Numerical Representation for Action in Crows Obeys the Weber-Fechner Law
Kirschhock, M. E., & Nieder, A. (2023) · Psychological Science · doi:10.1177/09567976231201624 · Cited by 10
“Moreover, computational networks representing numerosity also show logarithmic scaling for the representation of sensed numerosity (Dehaene & Changeux, 1993; Nasr et al., 2019; Nasr & Nieder, 2021).”
Self-citationZero-shot counting with a dual-stream neural network model
Thompson, J. A. F., Sheahan, H., Dumbalska, T., Sandbrink, J. D., Piazza, M., & Summerfield, C. (2024) · Neuron · doi:10.1016/j.neuron.2024.10.008 · Cited by 16
“Such number detectors have been found in networks trained on visual object recognition, 60,61 unsupervised objectives, 62,63 action prediction, 64 or networks that are not trained at all (randomly initialized networks).”
Spontaneous emergence of rudimentary music detectors in deep neural networks
Kim, G., Kim, D.-K., & Jeong, H. (2024) · Nature Communications · doi:10.1038/s41467-023-44516-0 · Cited by 12
Visual Search Asymmetry: Deep Nets and Humans Share Similar Inherent Biases
Gupta, S. K., Zhang, M., Wu, C.-C., Wolfe, J. M., & Kreiman, G. (2021) · PubMed · doi:10.48550/arxiv.2106.02953 · Cited by 9
“These observations are consistent with, and build bridges between, psychophysics observations in visual search studies [35, 40, 7, 2, 8] and analyses of behavioral biases of deep networks [28, 53, 33].”
ResultA neural theory for counting memories
Dasgupta, S., Hattori, D., & Navlakha, S. (2022) · Nature Communications · doi:10.1038/s41467-022-33577-2 · Cited by 10
Verma, B. K., & Sengupta, R. (2023) · Scientific Reports · doi:10.1038/s41598-023-44535-3 · Cited by 8
Extracting conservative equations from nonconservative state data
Huang, Z. L., Huang, S., Li, J., Wang, Y., & Jiang, H. (2022) · Journal of the Mechanics and Physics of Solids · doi:10.1016/j.jmps.2022.105127 · Cited by 8
Structure Learning and the Parietal CortexPreprint
Summerfield, C., Luyckx, F., & Sheahan, H. (2019) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/zfxj2 · Cited by 9
Numerosity coding in the brain: from early visual processing to abstract representations
Nieder, A. (2025) · Cerebral Cortex · doi:10.1093/cercor/bhaf180 · Cited by 7
Self-citationOhki, T., Kunii, N., & Chao, Z. C. (2023) · Reviews in the Neurosciences · doi:10.1515/revneuro-2022-0137 · Cited by 8
Emergent Symbol-like Number Variables in Artificial Neural Networks
Grant, S., Goodman, N. D., & McClelland, J. L. (2025) · Transactions on machine learning research · doi:10.48550/arxiv.2501.06141 · Cited by 8
“Many publications explore ANNs’ abilities to perform counting tasks (Di Nuovo & McClelland, 2019; Fang et al., 2018; Sabathiel et al., 2020; Kondapaneni & Perona, 2020; Nasr et al., 2019; Zhang et al., 2018; Trott et al., 2018) and closely related tasks (Csordás et al., 2024).”
Hofstetter, S., & Dumoulin, S. O. (2022) · iScience · doi:10.1016/j.isci.2022.105267 · Cited by 7
“…similar for the different images used in this study, these proposals would not work on the raw natural images per se, but may still be feasible in combination with object segmentation (Dehaene and Changeux, 1993; Nasr et al., 2019) and/or more basic second-order filters (Johnson and Baker, 2004).”
Hou, K., Zorzi, M., & Testolin, A. (2024) · Psychological Research · doi:10.1007/s00426-024-02064-2 · Cited by 6
“Again, this is unlikely to reflect the reality of the developmental environment that infants are stimulated with, and might tune onto.”
InfluentialChapalain, T., Thirion, B., & Eger, E. (2025) · Neural Computation · doi:10.1162/neco.a.30 · Cited by 7
Comparison of visual quantities in untrained neural networks
Lee, H., Choi, W., Lee, D. I., & Paik, S. (2023) · Cell Reports · doi:10.1016/j.celrep.2023.112900 · Cited by 7
Understanding the Ability of Deep Neural Networks to Count Connected Components in Images
Guan, S., & Loew, M. H. (2020) · IEEE Applied Imagery Pattern Recognition Workshop · doi:10.1109/aipr50011.2020.9425331 · Cited by 6
“In specific, (T-15) Triangle set: size ∈ [10, 20]; (T-27.”
BackgroundMethodInfluentialA connectome-based neuromarker of nonverbal number acuity and arithmetic skills
Zhang, D., Zhou, L., Yang, A., Li, S., Chang, C., Liu, J., & Zhou, K. (2022) · Cerebral Cortex · doi:10.1093/cercor/bhac108 · Cited by 7
“2016), and even deep neural networks (Nasr et al. 2019; Kim et al. 2021; Zhou et al. 2021), which encodes quantity in an approximate, nonsymbolic manner (i.”
Quantum spin models for numerosity perception
Yago Malo, J., Cicchini, G. M., Morrone, M. C., & Chiofalo, M. L. (2023) · PLOS ONE · doi:10.1371/journal.pone.0284610 · Cited by 5
“More recently, Nasr et al. [14] have found that some units of a deep convolutional neural network (CNN) designed and trained for objects recognition, could also perform number discrimination on static displays.”
The making of number: from content to representation
Nieder, A. (2026) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2025.12.011 · Cited by 6
Self-citationVisual number sense for real-world scenes shared by deep neural networks and humans
Wu, W., Ge, Y., Zuo, Z., Chen, L., Xu, Q., & Liu, Z. (2023) · Heliyon · doi:10.1016/j.heliyon.2023.e18517 · Cited by 5
“For numerosity perception, deep neural networks can perform visual number distinction [24], and demonstrate number-selective units, similar to number-selective neurons [25] designed for object categorization [26].”
BackgroundBayesian estimation yields anti-Weber variability
Prat-Carrabin, A., & Gershman, S. J. (2025) · PNAS Nexus · doi:10.1093/pnasnexus/pgaf275 · Cited by 9
“…properties of the brain’s processing of visual input, at least when numerosities are presented as visual arrays of multiple items; for instance, neural-network models of the visual stream, which are not trained for numerosity discrimination, nonetheless exhibit numerosity-tuned responses (54 – 56).”
Emergence of Emotion Selectivity in Deep Neural Networks Trained to Recognize Visual Objects
Liu, P., Bo, K., Ding, M., & Fang, R. (2024) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1011943 · Cited by 5
“For instance, (28) showed that neurons in a CNN trained on 58 ImageNet became selective for numbers without having been trained on any "number" datasets.”
The malleable impact of non-numeric features in visual number perception
Dramkin, D., Bonn, C. D., Baer, C., & Odic, D. (2022) · Acta Psychologica · doi:10.1016/j.actpsy.2022.103737 · Cited by 5
On Numerosity of Deep Neural NetworksPreprint
Zhang, X., & Wu, X. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.08674 · Cited by 5
“In the first part of this paper, we prove that the conclusion of [22] is incorrect and maintain that the pure data-driven, black box DCNN methodology cannot learn the abstract notion of numbers.”
BackgroundMethodInfluentialMachts, T., Grüb, J., & Nieder, A. (2025) · Progress in Neurobiology · doi:10.1016/j.pneurobio.2025.102794 · Cited by 4
Self-citationCroteau, J., Fornaciai, M., Huber, D. E., & Park, J. (2024) · Cerebral Cortex · doi:10.1093/cercor/bhae418 · Cited by 5
Role of Lateral Inhibition on Visual Number Sense
Zhou, Y., Chen, H., & Wang, Y. (2022) · Frontiers in Computational Neuroscience · doi:10.3389/fncom.2022.810448 · Cited by 4
“…sense models based on artificial neural networks directly or indirectly use lateral inhibition function and can qualitatively or quantitatively reproduce the results of biological experiments (Dehaene and Changeux, 1993; Sengupta et al., 2014; Nieder, 2016; Hannagan et al., 2017; Nasr et al., 2019).”
BackgroundMethodResultInfluentialDisparate processing of numerosity and associated continuous magnitudes in rats
Liang, T., Peng, R., Rong, K., Li, J., Ke, Y., & Yung, W. (2024) · Science Advances · doi:10.1126/sciadv.adj2566 · Cited by 4
From Perils to Possibilities: Understanding how Human (and AI) Biases affect Online ForaPreprint
Morini, V., Pansanella, V., Abramski, K., Cau, E., Failla, A., Citraro, S., & Rossetti, G. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2403.14298 · Cited by 4
“…presented evidence that GPT models can solve traditional false-belief tasks, leading the authors to hypothesize that ToM may develop as an emerging behavior in LLMs during the training phase, similar to the case of the ability to count [72] (which emerged spontaneously in image processing models).”
BackgroundA Computational Theory for the Emergence of Grammatical Categories in Cortical Dynamics
Demattíes, D., Rizzi, S., Thiruvathukal, G. K., Perez, M. D., Wainselboim, A. J., & Zanutto, B. S. (2020) · Frontiers in Neural Circuits · doi:10.3389/fncir.2020.00012 · Cited by 3
“For instance, it has been seen that biologically-inspired deep neural networks spontaneously gain number sense even when trained only to recognize objects in an image (Nasr et al., 2019).”
BackgroundNoda, K., Soda, T., & Yamashita, Y. (2024) · Frontiers in Neuroscience · doi:10.3389/fnins.2024.1330512 · Cited by 3
“To address this potential confounding factor, future studies should consider controlling such low-level visual features by drawing on the methodology of previous computational simulation studies (Nasr et al., 2019; Testolin et al., 2020).”
BackgroundMethodInfluentialAn, N. H., Roh, H., Kim, S., Kim, J. H., & Im, M. (2025) · Advanced Science · doi:10.1002/advs.202405789 · Cited by 4
Gamal, M., & Eldawlatly, S. (2025) · Journal of Neuroscience Methods · doi:10.1016/j.jneumeth.2025.110429 · Cited by 2
Space, Time, Categories, Mechanics, and Consciousness: On Kant and Neuroscience
Hepp, K. (2020) · Journal of Statistical Physics · doi:10.1007/s10955-020-02551-x · Cited by 3
“Theoretical neuroscience is still waiting for her Newton; at present she is seduced by recent successes in ‘Deep Learning’ [60], as exemplified by [31].”
BackgroundThe "Fraction Sense" Emerges from a Deep Convolutional Neural Network.
Chuang, Y., Hubbard, E. M., & Austerweil, J. L. (2020) · eScholarship (California Digital Library) · Cited by 2
Zou, J., & Ding, N. (2021) · Research Square · doi:10.21203/rs.3.rs-813994/v1 · Cited by 2
Memory preservation and cooperative shielding in complex quantum networks
Ausilio, S., Borgonovi, F., Celardo, G. L., Malo, J. Y., & Chiofalo, M. L. (2025) · Physical review. B./Physical review. B · doi:10.1103/dt4n-8sqg · Cited by 2
“In [5, 14], the mapping of the problem into an out-of-equilibrium XXZ network with all-to-all connectivity has been shown to account for the Weber’s law in effective and efficient manner, whereas conventional artificial neural networks poorly perform [82].”
Decision-level processes in rapid numerosity estimation
Solstad, T., Kaspersen, E., Romijn, E. I., & Hodgen, J. (2026) · Cognition · doi:10.1016/j.cognition.2026.106511 · Cited by 1
Recognition ability of untrained neural networks to symbolic numbers
Zhou, Y., Chen, H., & Wang, Y. (2022) · Frontiers in Neuroinformatics · doi:10.3389/fninf.2022.973010 · Cited by 1
“A non-symbolic numerosity dataset mimicking the dataset of Nasr et al. (2019) was constructed to test whether the model could reproduce the non-symbolic numerosity recognition abilities that animals had prior to training.”
MethodШевченко, А. А., Shevchenko, A. A., Ульянов, С. В., Ulyanov, S. V., Шевченко, А. А., Shevchenko, A. V., Зрелова, Д. П., & П., З. Д. (2023) · Международный журнал Программные продукты и системы · doi:10.15827/0236-235x.141.0 · Cited by 1
Vers l’XAI sémantique – la troisième vague de l’intelligence artificielle explicable
Bollaert, M., & Coppin, G. (2024) · Ingénierie cognitique · doi:10.21494/iste.op.2024.1147 · Cited by 1
Metadata
- Authors
- Nasr, K., Viswanathan, P., & Nieder, A.
- Affiliations
- University of Tübingen (DE)
- Corresponding author
- Nieder, Andreas
- Venue
- Science Advances
- Published
- 2019-05-03
- Volume
- 5
- Issue
- 5
- Pages
- eaav7903-eaav7903
- ISSN
- 2375-2548 (ISSN-L)
- PMC
- PMC6506249
- OpenAlex
- W2944629072
- Cited by
- 110
- References
- 41
- Citations by year
- 2026: 4 · 2025: 11 · 2024: 18 · 2023: 19 · 2022: 19 · 2021: 20 · 2020: 15 · 2019: 4
- FWCI
- 11.34
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by-nc
- Locations
- Science Advances (publishedVersion, cc-by-nc)
- PubMed Central (publishedVersion, cc-by-nc)
- PubMed (publishedVersion)
- Funders
- Deutsche Forschungsgemeinschaft (NI 618/10-1)
- Topics
- Social Sciences › Psychology › Developmental and Educational Psychology › Cognitive and developmental aspects of mathematical skills
- Life Sciences › Neuroscience › Cognitive Neuroscience › Visual perception and processing mechanisms
- Social Sciences › Psychology › Experimental and Cognitive Psychology › Multisensory perception and integration
- Keywords
- number sense, numerosity, spontaneous emergence, deep neural network, visual object recognition, number discrimination, Weber-Fechner law, biologically inspired models, visual system
- MeSH terms
- Animals, Humans, Models, Biological, Neurons, Visual Perception, Neural Networks, Computer
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.
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
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.