@article {5284, title = {Aligning Model and Macaque Inferior Temporal Cortex Representations Improves Model-to-Human Behavioral Alignment and Adversarial Robustness}, journal = {bioRxiv}, year = {2022}, month = {07/2022}, abstract = {

While some state-of-the-art artificial neural network systems in computer vision are strikingly accurate models of the corresponding primate visual processing, there are still many discrepancies between these models and the behavior of primates on object recognition tasks. Many current models suffer from extreme sensitivity to adversarial attacks and often do not align well with the image-by-image behavioral error patterns observed in humans. Previous research has provided strong evidence that primate object recognition behavior can be very accurately predicted by neural population activity in the inferior temporal (IT) cortex, a brain area in the late stages of the visual processing hierarchy. Therefore, here we directly test whether making the late stage representations of models more similar to that of macaque IT produces new models that exhibit more robust, primate-like behavior. We conducted chronic, large-scale multi-electrode recordings across the IT cortex in six non-human primates (rhesus macaques). We then use these data to fine-tune (end-to-end) the model {\textquotedblleft}IT{\textquotedblright} representations such that they are more aligned with the biological IT representations, while preserving accuracy on object recognition tasks. We generate a cohort of models with a range of IT similarity scores validated on held-out animals across two image sets with distinct statistics. Across a battery of optimization conditions, we observed a strong correlation between the models{\textquoteright} IT-likeness and alignment with human behavior, as well as an increase in its adversarial robustness. We further assessed the limitations of this approach and find that the improvements in behavioral alignment and adversarial robustness generalize across different image statistics, but not to object categories outside of those covered in our IT training set. Taken together, our results demonstrate that building models that are more aligned with the primate brain leads to more robust and human-like behavior, and call for larger neural data-sets to further augment these gains.

}, author = {Joel Dapello and Kohitij Kar and Martin Schrimpf and Robert Geary and Michael Ferguson and David D. Cox and James J. DiCarlo} } @conference {5298, title = {Primate Inferotemporal Cortex Neurons Generalize Better to Novel Image Distributions Than Analogous Deep Neural Networks Units}, booktitle = {NeurIPS}, year = {2022}, month = {10/2022}, abstract = {

Humans are successfully able to recognize objects in a variety of image distributions. Today{\textquoteright}s artificial neural networks (ANNs), on the other hand, struggle to recognize objects in many image domains, especially those different from the training distribution. It is currently unclear which parts of the ANNs could be improved in order to close this generalization gap. In this work, we used recordings from primate high-level visual cortex (IT) to isolate whether ANNs lag behind primate generalization capabilities because of their encoder (transformations up to the penultimate layer), or their decoder (linear transformation into class labels). Specifically, we fit a linear decoder on images from one domain and evaluate transfer performance on twelve held-out domains, comparing fitting on primate IT representations vs. representations in ANN penultimate layers. To fairly compare, we scale the number of each ANN{\textquoteright}s units so that its in-domain performance matches that of the sampled IT population (i.e. 71 IT neural sites, 73\% binary-choice accuracy). We find that the sampled primate population achieves, on average, 68\% performance on the held-out-domains. Comparably sampled populations from ANN model units generalize less well, maintaining on average 60\%. This is independent of the number of sampled units: models{\textquoteright} out-of-domain accuracies consistently lag behind primate IT. These results suggest that making ANN model units more like primate IT will improve the generalization performance of ANNs.

}, url = {https://openreview.net/forum?id=iPF7mhoWkOl}, author = {Ayu Marliawaty I Gusti Bagus and Tiago Marques and Sachi Sanghavi and James J. DiCarlo and Martin Schrimpf} } @article {5066, title = { Chemogenetic suppression of macaque V4 neurons produces retinotopically specific deficits in downstream IT neural activity patterns and core object recognition behavior}, journal = {Journal of Vision}, volume = {21}, year = {2021}, month = {09/2021}, chapter = {2489}, abstract = {

Distributed activity patterns across multiple brain areas (e.g., V4, IT) enable primates to accurately identify visual objects. To strengthen our inferences about the causal role of underlying brain circuits, it is necessary to develop targeted neural perturbation strategies that enable discrimination amongst competing models. To probe the role of area V4 in core object recognition, we expressed inhibitory DREADDs in neurons within a 5x5 mm subregion of V4 cortex via multiple viral injections (AAV8-hSyn-hM4Di-mCherry; two macaques). To assay for successful neural suppression, we recorded from a multi-electrode array implanted over the transfected V4. We also recorded from multi-electrode arrays in the IT cortex (the primary feedforward target of V4), while simultaneously measuring the monkeys{\textquoteright} behavior during object discrimination tasks. We found that systemic (intramuscular) injection of the DREADDs activator (CNO) produced reversible reductions (~20\%) in image-evoked V4 responses compared to the control condition (saline injections). Monkeys showed significant behavioral performance deficits upon CNO injections (compared to saline), which were larger when the object position overlapped with the RF estimates of the transfected V4 neurons. This is consistent with the hypothesis that the suppressed V4 neurons are critical to this behavior. Furthermore, we observed commensurate deficits in the linearly-decoded estimates of object identity from the IT population activity (post-CNO). To model the perturbed brain circuitry, we used a primate brain-mapped artificial neural network (ANN) model (CORnet-S) that supports object recognition. We {\textquotedblleft}lesioned{\textquotedblright} the model{\textquoteright}s corresponding V4 subregion by modifying its weights such that the responses matched a subset of our experimental V4 measurements (post-CNO). Indeed, the lesioned model better predicted the measured (held-out) V4 and IT responses (post-CNO), compared to the model{\textquoteright}s non-lesioned version, validating our approach. In the future, our approach allows us to discriminate amongst competing mechanistic brain models, while the data provides constraints to guide more accurate alternatives.

}, doi = {10.1167/jov.21.9.2489}, url = {https://jov.arvojournals.org/article.aspx?articleid=2777218}, author = {Kohitij Kar and Martin Schrimpf and Kailyn Schmidt and James J. DiCarlo} } @conference {5045, title = {Frivolous Units: Wider Networks Are Not Really That Wide}, booktitle = {AAAI 2021}, year = {2021}, month = {05/2021}, abstract = {

A remarkable characteristic of overparameterized deep neural networks (DNNs) is that their accuracy does not degrade when the network width is increased. Recent evidence suggests that developing compressible representations allows the complex- ity of large networks to be adjusted for the learning task at hand. However, these representations are poorly understood. A promising strand of research inspired from biology involves studying representations at the unit level as it offers a more granular interpretation of the neural mechanisms. In order to better understand what facilitates increases in width without decreases in accuracy, we ask: Are there mechanisms at the unit level by which networks control their effective complex- ity? If so, how do these depend on the architecture, dataset, and hyperparameters? We identify two distinct types of {\textquotedblleft}frivolous{\textquotedblright} units that prolifer- ate when the network{\textquoteright}s width increases: prunable units which can be dropped out of the network without significant change to the output and redundant units whose activities can be ex- pressed as a linear combination of others. These units imply complexity constraints as the function the network computes could be expressed without them. We also identify how the development of these units can be influenced by architecture and a number of training factors. Together, these results help to explain why the accuracy of DNNs does not degrade when width is increased and highlight the importance of frivolous units toward understanding implicit regularization in DNNs.

}, url = {https://dblp.org/rec/conf/aaai/CasperBDGSVK21.html}, author = {Stephen Casper and Xavier Boix and Vanessa D{\textquoteright}Amario and Ling Guo and Martin Schrimpf and Vinken, Kasper and Gabriel Kreiman} } @article {5079, title = {Multi-scale hierarchical neural network models that bridge from single neurons in the primate primary visual cortex to object recognition behavior}, journal = {bioRxiv}, year = {2021}, month = {08/2021}, abstract = {

Primate visual object recognition relies on the representations in cortical areas at the top of the ventral stream that are computed by a complex, hierarchical network of neural populations. While recent work has created reasonably accurate image-computable hierarchical neural network models of those neural stages, those models do not yet bridge between the properties of individual neurons and the overall emergent behavior of the ventral stream. One reason we cannot yet do this is that individual artificial neurons in multi-stage models have not been shown to be functionally similar to individual biological neurons. Here, we took an important first step by building and evaluating hundreds of hierarchical neural network models in how well their artificial single neurons approximate macaque primary visual cortical (V1) neurons. We found that single neurons in certain models are surprisingly similar to their biological counterparts and that the distributions of single neuron properties, such as those related to orientation and spatial frequency tuning, approximately match those in macaque V1. Critically, we observed that hierarchical models with V1 stages that better match macaque V1 at the single neuron level are also more aligned with human object recognition behavior. Finally, we show that an optimized classical neuroscientific model of V1 is more functionally similar to primate V1 than all of the tested multi-stage models, suggesting room for further model improvements with tangible payoffs in closer alignment to human behavior. These results provide the first multi-stage, multi-scale models that allow our field to ask precisely how the specific properties of individual V1 neurons relate to recognition behavior.

}, author = {Tiago Marques and Martin Schrimpf and James J. DiCarlo} } @article {4969, title = {The neural architecture of language: Integrative modeling converges on predictive processing}, journal = {Proceedings of the National Academy of Sciences}, volume = {118}, year = {2021}, month = {11/2021}, pages = {e2105646118}, abstract = {

Significance

Language is a quintessentially human ability. Research has long probed the functional architecture of language in the mind and brain using diverse neuroimaging, behavioral, and computational modeling approaches. However, adequate neurally-mechanistic accounts of how meaning might be extracted from language are sorely lacking. Here, we report a first step toward addressing this gap by connecting recent artificial neural networks from machine learning to human recordings during language processing. We find that the most powerful models predict neural and behavioral responses across different datasets up to noise levels. Models that perform better at predicting the next word in a sequence also better predict brain measurements{\textemdash}providing computationally explicit evidence that predictive processing fundamentally shapes the language comprehension mechanisms in the brain.

Abstract

The neuroscience of perception has recently been revolutionized with an integrative modeling approach in which computation, brain function, and behavior are linked across many datasets and many computational models. By revealing trends across models, this approach yields novel insights into cognitive and neural mechanisms in the target domain. We here present a systematic study taking this approach to higher-level cognition: human language processing, our species{\textquoteright} signature cognitive skill. We find that the most powerful {\textquotedblleft}transformer{\textquotedblright} models predict nearly 100\% of explainable variance in neural responses to sentences and generalize across different datasets and imaging modalities (functional MRI and electrocorticography). Models{\textquoteright} neural fits ({\textquotedblleft}brain score{\textquotedblright}) and fits to behavioral responses are both strongly correlated with model accuracy on the next-word prediction task (but not other language tasks). Model architecture appears to substantially contribute to neural fit. These results provide computationally explicit evidence that predictive processing fundamentally shapes the language comprehension mechanisms in the human brain.

}, issn = {0027-8424}, doi = {10.1073/pnas.2105646118}, url = {http://www.pnas.org/lookup/doi/10.1073/pnas.2105646118}, author = {Martin Schrimpf and Blank, Idan Asher and Tuckute, Greta and Kauf, Carina and Hosseini, Eghbal A. and Nancy Kanwisher and Joshua B. Tenenbaum and Fedorenko, Evelina} } @conference {4527, title = {Hierarchical neural network models that more closely match primary visual cortex tend to better explain higher level visual cortical responses }, booktitle = {COSYNE}, year = {2020}, month = {02/2020}, address = {Denver, Colorado, USA}, author = {Tiago Marques and Martin Schrimpf and James J. DiCarlo} } @article {4810, title = {Integrative Benchmarking to Advance Neurally Mechanistic Models of Human Intelligence}, journal = {Neuron}, volume = {108}, year = {2020}, month = {11/2020}, pages = {413 - 423}, issn = {08966273}, doi = {10.1016/j.neuron.2020.07.040}, url = {https://linkinghub.elsevier.com/retrieve/pii/S089662732030605X}, author = {Martin Schrimpf and Kubilius, Jonas and Lee, Michael J. and N. Apurva Ratan Murty and Ajemian, Robert and James J. DiCarlo} } @proceedings {4692, title = {Simulating a Primary Visual Cortex at the Front of CNNs Improves Robustness to Image Perturbations}, year = {2020}, month = {12/2020}, abstract = {

Current state-of-the-art object recognition models are largely based on convolutional neural network (CNN) architectures, which are loosely inspired by the primate visual system. However, these CNNs can be fooled by imperceptibly small, explicitly crafted perturbations, and struggle to recognize objects in corrupted images that are easily recognized by humans. Here, by making comparisons with primate neural data, we first observed that CNN models with a neural hidden layer that better matches primate primary visual cortex (V1) are also more robust to adversarial attacks. Inspired by this observation, we developed VOneNets, a new class of hybrid CNN vision models. Each VOneNet contains a fixed weight neural network front-end that simulates primate V1, called the VOneBlock, followed by a neural network back-end adapted from current CNN vision models. The VOneBlock is based on a classical neuroscientific model of V1: the linear-nonlinear-Poisson model, consisting of a biologically-constrained Gabor filter bank, simple and complex cell nonlinearities, and a V1 neuronal stochasticity generator. After training, VOneNets retain high ImageNet performance, but each is substantially more robust, outperforming the base CNNs and state-of-the-art methods by 18\% and 3\%, respectively, on a conglomerate benchmark of perturbations comprised of white box adversarial attacks and common image corruptions. Finally, we show that all components of the VOneBlock work in synergy to improve robustness. While current CNN architectures are arguably brain-inspired, the results presented here demonstrate that more precisely mimicking just one stage of the primate visual system leads to new gains in ImageNet-level computer vision applications.

Github: https://github.com/dicarlolab/vonenet

}, url = {https://proceedings.neurips.cc/paper/2020/hash/98b17f068d5d9b7668e19fb8ae470841-Abstract.html}, author = {Joel Dapello and Tiago Marques and Martin Schrimpf and Franziska Geiger and David Cox and James J. DiCarlo} } @conference {4528, title = {Temporal information for action recognition only needs to be integrated at a choice level in neural networks and primates }, booktitle = {COSYNE}, year = {2020}, month = {02/2020}, address = {Denver, CO, USA}, author = {Martin Schrimpf and Fukushi Sato and Sachi Sanghavi and James J. DiCarlo} } @article {4632, title = {ThreeDWorld: A Platform for Interactive Multi-Modal Physical Simulation}, journal = {arXiv}, year = {2020}, month = {07/2020}, type = {Preprint}, abstract = {

We introduce ThreeDWorld (TDW), a platform for interactive multi-modal physical simulation. With TDW, users can simulate high-fidelity sensory data and physical interactions between mobile agents and objects in a wide variety of rich 3D environments. TDW has several unique properties: 1) realtime near photo-realistic image rendering quality; 2) a library of objects and environments with materials for high-quality rendering, and routines enabling user customization of the asset library; 3) generative procedures for efficiently building classes of new environments 4) high-fidelity audio rendering; 5) believable and realistic physical interactions for a wide variety of material types, including cloths, liquid, and deformable objects; 6) a range of "avatar" types that serve as embodiments of AI agents, with the option for user avatar customization; and 7) support for human interactions with VR devices. TDW also provides a rich API enabling multiple agents to interact within a simulation and return a range of sensor and physics data representing the state of the world. We present initial experiments enabled by the platform around emerging research directions in computer vision, machine learning, and cognitive science, including multi-modal physical scene understanding, multi-agent interactions, models that "learn like a child", and attention studies in humans and neural networks. The simulation platform will be made publicly available.

}, url = {https://arxiv.org/abs/2007.04954}, author = {Chuang Gen and Jeremy Schwartz and Seth Alter and Martin Schrimpf and James Traer and Julian De Freitas and Jonas Kubilius and Abhishek Bhandwaldar and Nick Haber and Megumi Sano and Kuno Kim and Elias Wang and Damian Mrowca and Michael Lingelbach and Aidan Curtis and Kevin Feigleis and Daniel Bear and Dan Gutfreund and David Cox and James J. DiCarlo and Josh H. McDermott and Joshua B. Tenenbaum and Daniel L K Yamins} } @article {4633, title = {ThreeDWorld (TDW): A High-Fidelity, Multi-Modal Platform for Interactive Physical Simulation}, year = {2020}, month = {07/2020}, abstract = {

TDW is a 3D virtual world simulation platform, utilizing state-of-the-art video game engine technology

A TDW simulation consists of two components: a) the Build, a compiled executable running on the Unity3D Engine, which is responsible for image rendering, audio synthesis and physics simulations; and b) the Controller, an external Python interface to communicate with the build.

Researchers write Controllers that send commands to the Build, which executes those commands and returns a broad range of data types representing the state of the virtual world.

TDW provides researchers with:

TDW is being used on a daily basis in multiple labs, supporting research that sits at the nexus of neuroscience, cognitive science and artificial intelligence.

Find out more about ThreeDWorld on the project weobsite using the link below.

}, url = {http://www.threedworld.org/}, author = {Jeremy Schwartz and Seth Alter and James J. DiCarlo and Josh H. McDermott and Joshua B. Tenenbaum and Daniel L K Yamins and Dan Gutfreund and Chuang Gan and James Traer and Jonas Kubilius and Martin Schrimpf and Abhishek Bhandwaldar and Julian De Freitas and Damian Mrowca and Michael Lingelbach and Megumi Sano and Daniel Bear and Kuno Kim and Nick Haber and Chaofei Fan} } @proceedings {4379, title = {Brain-Like Object Recognition with High-Performing Shallow Recurrent ANNs}, year = {2019}, month = {10/2019}, address = {Vancouver, Canada}, abstract = {

Deep convolutional artificial neural networks (ANNs) are the leading class of candidate models of the mechanisms of visual processing in the primate ventral stream. While initially inspired by brain anatomy, over the past years, these ANNs have evolved from a simple eight-layer architecture in AlexNet to extremely deep and branching architectures, demonstrating increasingly better object categorization performance, yet bringing into question how brain-like they still are. In particular, typical deep models from the machine learning community are often hard to map onto the brain{\textquoteright}s anatomy due to their vast number of layers and missing biologically-important connections, such as recurrence. Here we demonstrate that better anatomical alignment to the brain and high performance on machine learning as well as neuroscience measures do not have to be in contradiction. We developed CORnet-S, a shallow ANN with four anatomically mapped areas and recurrent connectivity, guided by Brain-Score, a new large-scale composite of neural and behavioral benchmarks for quantifying the functional fidelity of models of the primate ventral visual stream. Despite being significantly shallower than most models, CORnet-S is the top model on Brain-Score and outperforms similarly compact models on ImageNet. Moreover, our extensive analyses of CORnet-S circuitry variants reveal that recurrence is the main predictive factor of both Brain- Score and ImageNet top-1 performance. Finally, we report that the temporal evolution of the CORnet-S "IT" neural population resembles the actual monkey IT population dynamics. Taken together, these results establish CORnet-S, a compact, recurrent ANN, as the current best model of the primate ventral visual stream.

}, author = {Jonas Kubilius and Martin Schrimpf and Kohitij Kar and Rishi Rajalingham and Ha Hong and Najib J. Majaj and Elias B. Issa and Pouya Bashivan and Jonathan Prescott-Roy and Kailyn Schmidt and Aran Nayebi and Daniel Bear and Daniel L K Yamins and James J. DiCarlo} } @conference {4321, title = {Large-scale hyperparameter search for predicting human brain responses in the Algonauts challenge}, booktitle = { The Algonauts Project: Explaining the Human Visual Brain Workshop 2019 }, year = {2019}, month = {8/14/2019}, address = {MIT, Cambridge MA}, doi = {10.1101/689844 }, url = {https://www.biorxiv.org/content/10.1101/689844v2.full}, author = {K.M. Jozwik and Lee, M. and Marques, T. and Martin Schrimpf and Pouya Bashivan} } @conference {4322, title = {To find better neural network models of human vision, find better neural network models of primate vision}, booktitle = {BioRxiv}, year = {2019}, abstract = {

Specific deep artificial neural networks (ANNs) are the current best models of ventral visual processing and object recognition behavior in monkeys. We here explore whether models of non-human primate vision generalize to visual processing in the human primate brain. Specifically, we asked if model match to monkey IT is a predictor of model match to human IT, even when scoring those matches on different images. We found that the model match to monkey IT is a positive predictor of the model match to human IT (R = 0.36), and that this approach outperforms the current standard predictor of model accuracy on ImageNet. This suggests a more powerful approach for pre-selecting models as hypotheses of human brain processing.

}, url = {https://www.biorxiv.org/content/10.1101/688390v1.full}, author = {K.M. Jozwik and Martin Schrimpf and Nancy Kanwisher and James J. DiCarlo} } @article {4294, title = {Brain-Score: Which Artificial Neural Network for Object Recognition is most Brain-Like?}, journal = {bioRxiv preprint}, year = {2018}, abstract = {

The internal representations of early deep artificial neural networks (ANNs) were found to be remarkably similar to the internal neural representations measured experimentally in the primate brain. Here we ask, as deep ANNs have continued to evolve, are they becoming more or less brain-like? ANNs that are most functionally similar to the brain will contain mechanisms that are most like those used by the brain. We therefore developed Brain-Score {\textendash} a composite of multiple neural and behavioral benchmarks that score any ANN on how similar it is to the brain{\textquoteright}s mechanisms for core object recognition {\textendash} and we deployed it to evaluate a wide range of state-of-the-art deep ANNs. Using this scoring system, we here report that: (1) DenseNet-169, CORnet-S and ResNet-101 are the most brain-like ANNs. There remains considerable variability in neural and behavioral responses that is not predicted by any ANN, suggesting that no ANN model has yet captured all the relevant mechanisms. (3) Extending prior work, we found that gains in ANN ImageNet performance led to gains on Brain-Score. However, correlation weakened at >= 70\% top-1 ImageNet performance, suggesting that additional guidance from neuroscience is needed to make further advances in capturing brain mechanisms. (4) We uncovered smaller (i.e. less complex) ANNs that are more brain-like than many of the best-performing ImageNet models, which suggests the opportunity to simplify ANNs to better understand the ventral stream. The scoring system used here is far from complete. However, we propose that evaluating and tracking model-benchmark correspondences through a Brain-Score that is regularly updated with new brain data is an exciting opportunity: experimental benchmarks can be used to guide machine network evolution, and machine networks are mechanistic hypotheses of the brain{\textquoteright}s network and thus drive next experiments. To facilitate both of these, we release Brain-Score.org: a platform that hosts the neural and behavioral benchmarks, where ANNs for visual processing can be submitted to receive a Brain-Score and their rank relative to other models, and where new experimental data can be naturally incorporated.

}, keywords = {computational neuroscience, deep learning, Neural Networks, object recognition, ventral stream}, doi = {10.1101/407007}, url = {https://www.biorxiv.org/content/10.1101/407007v1}, author = {Martin Schrimpf and Jonas Kubilius}, editor = {Ha Hong and Najib J. Majaj and Rishi Rajalingham and Elias B. Issa and Kohitij Kar and Pouya Bashivan and Jonathan Prescott-Roy and Kailyn Schmidt and Daniel L K Yamins and James J. DiCarlo} } @article {3764, title = {Recurrent computations for visual pattern completion}, journal = {Proceedings of the National Academy of Sciences}, year = {2018}, month = {08/2018}, abstract = {

Making inferences from partial information constitutes a critical aspect of cognition. During visual perception, pattern completion enables recognition of poorly visible or occluded objects. We combined psychophysics, physiology, and computational models to test the hypothesis that pattern completion is implemented by recurrent computations and present three pieces of evidence that are consistent with this hypothesis. First, subjects robustly recognized objects even when they were rendered \<15\% visible, but recognition was largely impaired when processing was interrupted by backward masking. Second, invasive physiological responses along the human ventral cortex exhibited visually selective responses to partially visible objects that were delayed compared with whole objects, suggesting the need for additional computations. These physiological delays were correlated with the effects of backward masking. Third, state-of-the-art feed-forward computational architectures were not robust to partial visibility. However, recognition performance was recovered when the model was augmented with attractor-based recurrent connectivity. The recurrent model was able to predict which images of heavily occluded objects were easier or harder for humans to recognize, could capture the effect of introducing a backward mask on recognition behavior, and was consistent with the physiological delays along the human ventral visual stream. These results provide a strong argument of plausibility for the role of recurrent computations in making visual inferences from partial information.

}, keywords = {Artificial Intelligence, computational neuroscience, Machine Learning, pattern completion, Visual object recognition}, issn = {0027-8424}, doi = {10.1073/pnas.1719397115}, url = {http://www.pnas.org/lookup/doi/10.1073/pnas.1719397115}, author = {Hanlin Tang and Martin Schrimpf and William Lotter and Moerman, Charlotte and Paredes, Ana and Ortega Caro, Josue and Hardesty, Walter and David Cox and Gabriel Kreiman} } @article {3881, title = {Single units in a deep neural network functionally correspond with neurons in the brain: preliminary results}, year = {2018}, month = {11/2018}, abstract = {

Deep neural networks have been shown to predict neural responses in higher visual cortex. The mapping from the model to a neuron in the brain occurs through a linear combination of many units in the model, leaving open the question of whether there also exists a correspondence at the level of individual neurons. Here we show that there exist many one-to-one mappings between single units in a deep neural network model and neurons in the brain. We show that this correspondence at the single- unit level is ubiquitous among state-of-the-art deep neural networks, and grows more pronounced for models with higher performance on a large-scale visual recognition task. Comparing matched populations{\textemdash}in the brain and in a model{\textemdash}we demonstrate a further correspondence at the level of the population code: stimulus category can be partially decoded from real neural responses using a classifier trained purely on a matched population of artificial units in a model. This provides a new point of investigation for phenomena which require fine-grained mappings between deep neural networks and the brain.

}, author = {Luke Arend and Yena Han and Martin Schrimpf and Pouya Bashivan and Kohitij Kar and Tomaso Poggio and James J. DiCarlo and Xavier Boix} } @article {2681, title = {On the Robustness of Convolutional Neural Networks to Internal Architecture and Weight Perturbations}, year = {2017}, month = {03/2017}, abstract = {

Deep convolutional neural networks are generally regarded as robust function approximators. So far, this intuition is based on perturbations to external stimuli such as the images to be classified. Here we explore the robustness of convolutional neural networks to perturbations to the internal weights and architecture of the network itself. We show that convolutional networks are surprisingly robust to a number of internal perturbations in the higher convolutional layers but the bottom convolutional layers are much more fragile. For instance, Alexnet shows less than a 30\% decrease in classification performance when randomly removing over 70\% of weight connections in the top convolutional or dense layers but performance is almost at chance with the same perturbation in the first convolutional layer. Finally, we suggest further investigations which could continue to inform the robustness of convolutional networks to internal perturbations.

}, author = {Nicholas Cheney and Martin Schrimpf and Gabriel Kreiman} }