Amr Farahat

The paper shows that representational drift in visual cortex and neural networks is driven by sensory input rather than elapsed time, with same-class acceleration observed during stimulus presentation.

The paper shows random-weight CNNs with sufficient complexity can predict early visual cortex activity as well as trained models, but not higher visual areas.

This study shows that VIM-DBS enhances motor learning for fixed sequential finger movements but not for random sequences, indicating its role in thalamo-cortical motor learning.

This paper evaluates CNNs for EEG decoding and introduces saliency maps to visualize model features, showing CNNs outperform LDA on high-dimensional data.