ferrarikazu/adversarial-cognitive-model

This repository is a research platform for studying adversarial robustness across human vision and multiple vision-model families, with RHAN-Next as its current architectural focus. The central workflow is to train or select models, evaluate them under adversarial attacks, and inspect comparative behavioral and human-study results. A Streamlit application exposes inference, robustness experiments, model internals, datasets, and reports. The graph preserves documented capabilities; some dashboard backend wiring and unsampled analysis implementations are represented only where paths or explicit relationships are supported.