Stefano Martiniani

This paper shows that divisive normalization can stabilize recurrent neural networks with strong recurrent connections, preventing instability through critical slowing down.

The paper identifies molecular traces of long-term memory using gene expression and protein analysis, revealing shadow proteins like PKMζ that are not detected by transcriptome profiling.

This paper develops a hierarchical neural circuit theory that explains normalization and inter-areal communication through feedback connections and validates it with simulations and spectral analysis.

This paper predicts and interprets protein developability using transfer learning on a high-throughput dataset of engineered protein variants.