Performance-dependent network evolution for enhanced computational capacity?

Abstract:

The quest to understand structure–function relationships in networks has intensified, yet optimal architectures for complex information processing remain poorly understood. In this work, Prof. Sinha investigates how task-specific network structures emerge through performance-dependent network evolution based on reservoir computing principles. She shows that the resulting minimal networks consistently outperform those produced by alternative growth strategies and Erdős–Rényi random networks. These evolved networks exhibit pronounced sparsity, follow scaling laws in node-density space, and display a distinctive asymmetry in the distribution of input and readout nodes. Her findings provide new insights into process-specific network evolution and inform the design of efficient information-processing systems, particularly in machine learning.

​​

Bio: 

Prof. Sudeshna Sinha is a professor at the Indian Institute of Science Education and Research Mohali. She has previously served as dean of faculty affairs and deputy director. She has

authored over a hundred research papers and holds several patents, with her work featured in Nature News & Views, Scientific American, The Economist, and MIT Technology Review. She is a recipient of the Birla Award for Physics and the J.C. Bose National Fellowship, and is a fellow of the Indian Academy of Sciences, the Indian National Science Academy, and The World Academy of Sciences. She also serves on the editorial boards of leading journals, including Chaos, Communications in Nonlinear Science and Numerical Simulation, and Proceedings of the Royal Society A.