George A. Alvarez

The paper shows that leading neural network models of primate visual cortex differ in their ability to control neural firing using parametric stimulus perturbations, revealing discrepancies in how models and neurons parameterize visual space.

This paper uses deep learning to show that lightness illusions can emerge from edge-based image reconstruction, suggesting a simpler computational mechanism than previously thought.

This paper shows that feedforward CNNs can mimic human contour integration without lateral connections, recurrence, or feedback by leveraging receptive field progression and contour bias.

This paper shows that large language models can predict visual brain responses in humans and non-human primates, suggesting that language-aligned models capture general visual patterns rather than language-specific features.

This paper introduces contrastive coding as a framework explaining category-selective regions in the human brain using neural network analyses.

This paper develops models to explain memory errors in visual working memory tasks using natural images and finds that deep encoders predict difficulty better than baselines and replicate effects in artificial stimuli.

Across 224 vision models scored against fMRI responses, architecture and training objective barely mattered; the images a model was trained on mattered most, and linear re-weighting blurs differences.

This paper presents a self-supervised model that learns visual representations of individual images, showing category information emerges implicitly and matches category-supervised models in capturing brain response structures.

The paper benchmarks deep neural networks on mouse visual cortex data to assess their ability to model neural activity and resolve discrepancies in previous findings.

This paper introduces a self-supervised model that learns instance-level representations of images, achieving performance comparable to category-supervised models in predicting human ventral visual stream responses.

The paper shows that processing ambiguous images as meaningful (e.g., faces) predicts better memory, using EEG to identify this effect.