Dissociable Neural Representations of Adversarially Perturbed Images in Convolutional Neural Networks and the Human Brain

Despite the remarkable similarities between convolutional neural networks (CNN) and the human brain, CNNs still fall behind humans in many visual tasks, indicating that there still exist considerable differences between the two systems. Here, we leverage adversarial noise (AN) and adversarial interference (AI) images to quantify the consistency between neural representations and perceptual outcomes in the two systems. Humans can successfully recognize AI images as the same categories as their corresponding regular images but perceive AN images as meaningless noise. In contrast, CNNs can recognize AN images similar as corresponding regular images but classify AI images into wrong categories with surprisingly high confidence. We use functional magnetic resonance imaging to measure brain activity evoked by regular and adversarial images in the human brain, and compare it to the activity of artificial neurons in a prototypical CNN—AlexNet. In the human brain, we find that the representational similarity between regular and adversarial images largely echoes their perceptual similarity in all early visual areas. In AlexNet, however, the neural representations of adversarial images are inconsistent with network outputs in all intermediate processing layers, providing no neural foundations for the similarities at the perceptual level. Furthermore, we show that voxel-encoding models trained on regular images can successfully generalize to the neural responses to AI images but not AN images. These remarkable differences between the human brain and AlexNet in representation-perception association suggest that future CNNs should emulate both behavior and the internal neural presentations of the human brain.

Download Full-text

Not so fast: Limited validity of deep convolutional neural networks as in silico models for human naturalistic face processing

10.1101/2021.11.17.469009 ◽

2021 ◽

Author(s):

Guo Jiahui ◽

Ma Feilong ◽

Matteo Visconti di Oleggio Castello ◽

Samuel A Nastase ◽

James V Haxby ◽

...

Keyword(s):

Neural Networks ◽

Human Brain ◽

Convolutional Neural Networks ◽

Face Processing ◽

In Silico ◽

Processing System ◽

Human Face ◽

Deep Convolutional Neural Networks ◽

Neural Representations ◽

Fully Connected

Deep convolutional neural networks (DCNNs) trained for face identification can rival and even exceed human-level performance. The relationships between internal representations learned by DCNNs and those of the primate face processing system are not well understood, especially in naturalistic settings. We developed the largest naturalistic dynamic face stimulus set in human neuroimaging research (700+ naturalistic video clips of unfamiliar faces) and used representational similarity analysis to investigate how well the representations learned by high-performing DCNNs match human brain representations across the entire distributed face processing system. DCNN representational geometries were strikingly consistent across diverse architectures and captured meaningful variance among faces. Similarly, representational geometries throughout the human face network were highly consistent across subjects. Nonetheless, correlations between DCNN and neural representations were very weak overall—DCNNs captured 3% of variance in the neural representational geometries at best. Intermediate DCNN layers better matched visual and face-selective cortices than the final fully-connected layers. Behavioral ratings of face similarity were highly correlated with intermediate layers of DCNNs, but also failed to capture representational geometry in the human brain. Our results suggest that the correspondence between intermediate DCNN layers and neural representations of naturalistic human face processing is weak at best, and diverges even further in the later fully-connected layers. This poor correspondence can be attributed, at least in part, to the dynamic and cognitive information that plays an essential role in human face processing but is not modeled by DCNNs. These mismatches indicate that current DCNNs have limited validity as in silico models of dynamic, naturalistic face processing in humans.

Download Full-text

Emergence of Visual Center-Periphery Spatial Organization in Deep Convolutional Neural Networks

10.1101/2020.02.19.956748 ◽

2020 ◽

Author(s):

Yalda Mohsenzadeh ◽

Caitlin Mullin ◽

Benjamin Lahner ◽

Aude Oliva

Keyword(s):

Neural Networks ◽

Human Brain ◽

Convolutional Neural Networks ◽

Spatial Organization ◽

Brain Regions ◽

Deep Convolutional Neural Networks ◽

Visual Center ◽

Neural Representations ◽

Parahippocampal Cortex ◽

Cortical Regions

AbstractResearch at the intersection of computer vision and neuroscience has revealed hierarchical correspondence between layers of deep convolutional neural networks (DCNNs) and cascade of regions along human ventral visual cortex. Recently, studies have uncovered emergence of human interpretable concepts within DCNNs layers trained to identify visual objects and scenes. Here, we asked whether an artificial neural network (with convolutional structure) trained for visual categorization would demonstrate spatial correspondences with human brain regions showing central/peripheral biases. Using representational similarity analysis, we compared activations of convolutional layers of a DCNN trained for object and scene categorization with neural representations in human brain visual regions. Results reveal a brain-like topographical organization in the layers of the DCNN, such that activations of layer-units with central-bias were associated with brain regions with foveal tendencies (e.g. fusiform gyrus), and activations of layer-units with selectivity for image backgrounds were associated with cortical regions showing peripheral preference (e.g. parahippocampal cortex). The emergence of a categorical topographical correspondence between DCNNs and brain regions suggests these models are a good approximation of the perceptual representation generated by biological neural networks.

Download Full-text

Modeling the Dynamics of Human Brain Activity with Recurrent Neural Networks

Frontiers in Computational Neuroscience ◽

10.3389/fncom.2017.00007 ◽

2017 ◽

Vol 11 ◽

Cited By ~ 36

Author(s):

Umut Güçlü ◽

Marcel A. J. van Gerven

Keyword(s):

Neural Networks ◽

Human Brain ◽

Recurrent Neural Networks ◽

Brain Activity

Download Full-text

Brain hierarchy score: Which deep neural networks are hierarchically brain-like?

10.1101/2020.07.22.216713 ◽

2020 ◽

Author(s):

Soma Nonaka ◽

Kei Majima ◽

Shuntaro C. Aoki ◽

Yukiyasu Kamitani

Keyword(s):

Neural Networks ◽

Image Recognition ◽

High Performance ◽

Deep Neural Networks ◽

Recognition Performance ◽

Brain Activity ◽

Spatial Integration ◽

Hierarchical Processing ◽

Visual Areas ◽

The Brain

SummaryAchievement of human-level image recognition by deep neural networks (DNNs) has spurred interest in whether and how DNNs are brain-like. Both DNNs and the visual cortex perform hierarchical processing, and correspondence has been shown between hierarchical visual areas and DNN layers in representing visual features. Here, we propose the brain hierarchy (BH) score as a metric to quantify the degree of hierarchical correspondence based on the decoding of individual DNN unit activations from human brain activity. We find that BH scores for 29 pretrained DNNs with varying architectures are negatively correlated with image recognition performance, indicating that recently developed high-performance DNNs are not necessarily brain-like. Experimental manipulations of DNN models suggest that relatively simple feedforward architecture with broad spatial integration is critical to brain-like hierarchy. Our method provides new ways for designing DNNs and understanding the brain in consideration of their representational homology.

Download Full-text

Neural representations of perceptual color experience in the human ventral visual pathway

Proceedings of the National Academy of Sciences ◽

10.1073/pnas.1911041117 ◽

2020 ◽

Vol 117 (23) ◽

pp. 13145-13150 ◽

Cited By ~ 2

Author(s):

Insub Kim ◽

Sang Wook Hong ◽

Steven K. Shevell ◽

Won Mok Shim

Keyword(s):

Light Stimulus ◽

Visual Pathway ◽

Cortical Processing ◽

Neural Responses ◽

Color Experience ◽

Retinal Stimulation ◽

Neural Representations ◽

Visual Areas ◽

Ventral Visual Pathway ◽

Higher Visual Areas

Color is a perceptual construct that arises from neural processing in hierarchically organized cortical visual areas. Previous research, however, often failed to distinguish between neural responses driven by stimulus chromaticity versus perceptual color experience. An unsolved question is whether the neural responses at each stage of cortical processing represent a physical stimulus or a color we see. The present study dissociated the perceptual domain of color experience from the physical domain of chromatic stimulation at each stage of cortical processing by using a switch rivalry paradigm that caused the color percept to vary over time without changing the retinal stimulation. Using functional MRI (fMRI) and a model-based encoding approach, we found that neural representations in higher visual areas, such as V4 and VO1, corresponded to the perceived color, whereas responses in early visual areas V1 and V2 were modulated by the chromatic light stimulus rather than color perception. Our findings support a transition in the ascending human ventral visual pathway, from a representation of the chromatic stimulus at the retina in early visual areas to responses that correspond to perceptually experienced colors in higher visual areas.

Download Full-text

Deep learning: Using machine learning to study biological vision

10.1101/178152 ◽

2017 ◽

Author(s):

Najib J. Majaj ◽

Denis G. Pelli

Keyword(s):

Machine Learning ◽

Neural Networks ◽

Deep Learning ◽

Deep Neural Networks ◽

Brain Activity ◽

Neural Responses ◽

Statistical Tool ◽

Biological Vision ◽

Vision Science ◽

Living Organisms

ABSTRACTToday many vision-science presentations employ machine learning, especially the version called “deep learning”. Many neuroscientists use machine learning to decode neural responses. Many perception scientists try to understand how living organisms recognize objects. To them, deep neural networks offer benchmark accuracies for recognition of learned stimuli. Originally machine learning was inspired by the brain. Today, machine learning is used as a statistical tool to decode brain activity. Tomorrow, deep neural networks might become our best model of brain function. This brief overview of the use of machine learning in biological vision touches on its strengths, weaknesses, milestones, controversies, and current directions. Here, we hope to help vision scientists assess what role machine learning should play in their research.

Download Full-text

Getting the gist faster: Blurry images enhance the early temporal similarity between neural signals and convolutional neural networks

10.1101/2021.08.22.451834 ◽

2021 ◽

Author(s):

David A. Tovar ◽

Tijl Grootswagers ◽

James Jun ◽

Oakyoon Cha ◽

Randolph Blake ◽

...

Keyword(s):

Neural Networks ◽

Convolutional Neural Networks ◽

Brain Activity ◽

Visual Image ◽

High Temporal Resolution ◽

Neural Signals ◽

Time Frequency ◽

Brain Signals ◽

Spatial Frequencies ◽

Full Picture

Humans are able to recognize objects under a variety of noisy conditions, so models of the human visual system must account for how this feat is accomplished. In this study, we investigated how image perturbations, specifically reducing images to their low spatial frequency (LSF) components, affected correspondence between convolutional neural networks (CNNs) and brain signals recorded using magnetoencephalography (MEG). Using the high temporal resolution of MEG, we found that CNN-Brain correspondence for deeper and more complex layers across CNN architectures emerged earlier for LSF images than for their unfiltered broadband counterparts. The early emergence of LSF components is consistent with the coarse-to-fine theoretical framework for visual image processing, but surprisingly shows that LSF signals from images are more prominent when high spatial frequencies are removed. In addition, we decomposed MEG signals into oscillatory components and found correspondence varied based on frequency bands, painting a full picture of how CNN-Brain correspondence varies with time, frequency, and MEG sensor locations. Finally, we varied image properties of CNN training sets, and found marked changes in CNN processing dynamics and correspondence to brain activity. In sum, we show that image perturbations affect CNN-Brain correspondence in unexpected ways, as well as provide a rich methodological framework for assessing CNN-Brain correspondence across space, time, and frequency.

Download Full-text

Application Research of Deep Convolutional Neural Network in Computer Vision

Journal of Networking and Telecommunications ◽

10.18282/jnt.v2i2.886 ◽

2020 ◽

Vol 2 (2) ◽

pp. 23

Author(s):

Lei Wang

Keyword(s):

Neural Network ◽

Neural Networks ◽

Computer Vision ◽

Face Recognition ◽

Human Brain ◽

Image Classification ◽

Convolutional Neural Networks ◽

Convolution Neural Network ◽

Data Set ◽

Deep Convolution Neural Network

<p>As an important research achievement in the field of brain like computing, deep convolution neural network has been widely used in many fields such as computer vision, natural language processing, information retrieval, speech recognition, semantic understanding and so on. It has set off a wave of neural network research in industry and academia and promoted the development of artificial intelligence. At present, the deep convolution neural network mainly simulates the complex hierarchical cognitive laws of the human brain by increasing the number of layers of the network, using a larger training data set, and improving the network structure or training learning algorithm of the existing neural network, so as to narrow the gap with the visual system of the human brain and enable the machine to acquire the capability of "abstract concepts". Deep convolution neural network has achieved great success in many computer vision tasks such as image classification, target detection, face recognition, pedestrian recognition, etc. Firstly, this paper reviews the development history of convolutional neural networks. Then, the working principle of the deep convolution neural network is analyzed in detail. Then, this paper mainly introduces the representative achievements of convolution neural network from the following two aspects, and shows the improvement effect of various technical methods on image classification accuracy through examples. From the aspect of adding network layers, the structures of classical convolutional neural networks such as AlexNet, ZF-Net, VGG, GoogLeNet and ResNet are discussed and analyzed. From the aspect of increasing the size of data set, the difficulties of manually adding labeled samples and the effect of using data amplification technology on improving the performance of neural network are introduced. This paper focuses on the latest research progress of convolution neural network in image classification and face recognition. Finally, the problems and challenges to be solved in future brain-like intelligence research based on deep convolution neural network are proposed.</p>

Download Full-text

Integrating Flexible Normalization into Midlevel Representations of Deep Convolutional Neural Networks

Neural Computation ◽

10.1162/neco_a_01226 ◽

2019 ◽

Vol 31 (11) ◽

pp. 2138-2176 ◽

Cited By ~ 2

Author(s):

Luis Gonzalo Sánchez Giraldo ◽

Odelia Schwartz

Keyword(s):

Neural Networks ◽

Visual Cortex ◽

Convolutional Neural Networks ◽

Neural Responses ◽

Spatial Normalization ◽

Deep Convolutional Neural Networks ◽

Cortical Areas ◽

Classical Receptive Field ◽

Divisive Normalization ◽

Spatial Dependencies

Deep convolutional neural networks (CNNs) are becoming increasingly popular models to predict neural responses in visual cortex. However, contextual effects, which are prevalent in neural processing and in perception, are not explicitly handled by current CNNs, including those used for neural prediction. In primary visual cortex, neural responses are modulated by stimuli spatially surrounding the classical receptive field in rich ways. These effects have been modeled with divisive normalization approaches, including flexible models, where spatial normalization is recruited only to the degree that responses from center and surround locations are deemed statistically dependent. We propose a flexible normalization model applied to midlevel representations of deep CNNs as a tractable way to study contextual normalization mechanisms in midlevel cortical areas. This approach captures nontrivial spatial dependencies among midlevel features in CNNs, such as those present in textures and other visual stimuli, that arise from tiling high-order features geometrically. We expect that the proposed approach can make predictions about when spatial normalization might be recruited in midlevel cortical areas. We also expect this approach to be useful as part of the CNN tool kit, therefore going beyond more restrictive fixed forms of normalization.

Download Full-text