Abhranil Das

Principles of Local and Global Grouping that Underlie Segmentation of Natural Texture Images

This paper introduces a hierarchical Bayesian observer model for texture segmentation that mimics human visual processing and outperforms deep networks on specific tasks.

Systematic errors in connectivity inferred from activity in strongly coupled recurrent circuits

The paper shows that connectivity inferred from neural activity in strongly coupled circuits is biased toward false positive connections due to model mismatch, and suggests noisy or perturbed activity states may reduce this bias.