Search for: All records

Abstract Anterior regions of the ventral visual stream encode substantial information about object categories. Are top-down category-level forces critical for arriving at this representation, or can this representation be formed purely through domain-general learning of natural image structure? Here we present a fully self-supervised model which learns to represent individual images, rather than categories, such that views of the same image are embedded nearby in a low-dimensional feature space, distinctly from other recently encountered views. We find that category information implicitly emerges in the local similarity structure of this feature space. Further, these models learn hierarchical features which capture the structure of brain responses across the human ventral visual stream, on par with category-supervised models. These results provide computational support for a domain-general framework guiding the formation of visual representation, where the proximate goal is not explicitly about category information, but is instead to learn unique, compressed descriptions of the visual world. 

Brain responses in visual cortex are typically modeled as a positively and negatively weighted sum of all features within a deep neural network (DNN) layer. However, this linear fit can dramatically alter a given feature space, making it unclear whether brain prediction levels stem more from the DNN itself, or from the flexibility of the encoding model. As such, studies of alignment may benefit from a paradigm shift toward more constrained and theoretically driven mapping methods. As a proof of concept, here we present a case study of face and scene selectivity, showing that typical encoding analyses do not differentiate between aligned and misaligned tuning bases in model-to-brain predictivity. We introduce a new alignment complexity measure -- tuning reorientation -- which favors DNNs that achieve high brain alignment via minimal distortion of the original feature space. We show that this measure helps arbitrate between models that are superficially equal in their predictivity, but which differ in alignment complexity. Our experiments broadly signal the benefit of sparse, positive-weighted encoding procedures, which directly enforce an analogy between the tuning directions of model and brain feature spaces. 

According to the efficient coding hypothesis, neural populations encode information optimally when representations are high-dimensional and uncorrelated. However, such codes may carry a cost in terms of generalization and robustness. Past empirical studies of early visual cortex (V1) in rodents have suggested that this tradeoff indeed constrains sensory representations. However, it remains unclear whether these insights generalize across the hierarchy of the human visual system, and particularly to object representations in high-level occipitotemporal cortex (OTC). To gain new empirical clarity, here we develop a family of object recognition models with parametrically varying dropout proportion , which induces systematically varying dimensionality of internal responses (while controlling all other inductive biases). We find that increasing dropout produces an increasingly smooth, low-dimensional representational space. Optimal robustness to lesioning is observed at around 70% dropout, after which both accuracy and robustness decline. Representational comparison to large-scale 7T fMRI data from occipitotemporal cortex in the Natural Scenes Dataset reveals that this optimal degree of dropout is also associated with maximal emergent neural predictivity. Finally, using new techniques for achieving denoised estimates of the eigenspectrum of human fMRI responses, we compare the rate of eigenspectrum decay between model and brain feature spaces. We observe that the match between model and brain representations is associated with a common balance between efficiency and robustness in the representational space. These results suggest that varying dropout may reveal an optimal point of balance between the efficiency of high-dimensional codes and the robustness of low dimensional codes in hierarchical vision systems. 

Artificial neural networks (ANNs) inspired by biology are beginning to be widely used to model behavioural and neural data, an approach we call ‘neuroconnectionism’. ANNs have been not only lauded as the current best models of information processing inthe brain butalsocriticized for failing to account for basic cognitive functions. In this Perspective article, we propose that arguing about the successes and failures of a restricted set of current ANNs is the wrong approach to assess the promise of neuroconnectionism for brain science. Instead, we take inspiration from the philosophy of science, and in particular from Lakatos, who showed that the core of a scientific research programme is often not directly falsifiable but should be assessed by its capacity to generate novel insights. Following this view, we present neuroconnectionism as a general research programme centred around ANNs as a computational language for expressing falsifiable theories about brain computation. We describe the core of the programme, the underlying computational framework and its tools for testing specific neuroscientific hypotheses and deriving novel understanding. Taking a longitudinal view, we review past and present neuroconnectionist projects and their responses to challenges and argue that the research programme is highly progressive, generating new and otherwise unreachable insights into the workings of the brain. 

Self-organizing principles provide a computational account for the topographic organization of the high-level visual system. 

After years of experience, humans become experts at perceiving letters. Is this visual capacity attained by learning specialized letter features, or by reusing general visual features previously learned in service of object categorization? To explore this question, we first measured the perceptual similarity of letters in two behavioral tasks, visual search and letter categorization. Then, we trained deep convolutional neural networks on either 26-way letter categorization or 1000-way object categorization, as a way to operationalize possible specialized letter features and general object-based features, respectively. We found that the general object-based features more robustly correlated with the perceptual similarity of letters. We then operationalized additional forms of experience-dependent letter specialization by altering object-trained networks with varied forms of letter training; however, none of these forms of letter specialization improved the match to human behavior. Thus, our findings reveal that it is not necessary to appeal to specialized letter representations to account for perceptual similarity of letters. Instead, we argue that it is more likely that the perception of letters depends on domain-general visual features. 

Abstract Responses to visually presented objects along the cortical surface of the human brain have a large-scale organization reflecting the broad categorical divisions of animacy and object size. Emerging evidence indicates that this topographical organization is supported by differences between objects in mid-level perceptual features. With regard to the timing of neural responses, images of objects quickly evoke neural responses with decodable information about animacy and object size, but are mid-level features sufficient to evoke these rapid neural responses? Or is slower iterative neural processing required to untangle information about animacy and object size from mid-level features, requiring hundreds of milliseconds more processing time? To answer this question, we used EEG to measure human neural responses to images of objects and their texform counterparts—unrecognizable images that preserve some mid-level feature information about texture and coarse form. We found that texform images evoked neural responses with early decodable information about both animacy and real-world size, as early as responses evoked by original images. Furthermore, successful cross-decoding indicates that both texform and original images evoke information about animacy and size through a common underlying neural basis. Broadly, these results indicate that the visual system contains a mid-level feature bank carrying linearly decodable information on animacy and size, which can be rapidly activated without requiring explicit recognition or protracted temporal processing.