Alkan Kabakçıoğlu
Koç University
“Out-of-equilibrium dynamics in biology and machine learning”
Abstract
Dynamical trajectories in high-dimensional systems can exhibit complex and sometimes counterintuitive behaviors driven by noise. We here provide two examples: one from biology, involving the folding and melting dynamics of nucleic acids under thermal drive, and another from machine learning, where the training process can be seen as evolution under nonthermal noise in a complex loss landscape. In the first case, we demonstrate the time-reversal asymmetry between the folding and melting processes and find a type of Mpemba effect in the melting dynamics of the RNA hairpin, both underlining the importance of the molecule’s helical ground state in its nonequilibrium response to thermal effects. In the second case, the show how the nontrivial correlations in the gradient noise induced by minibatching result in a non-Boltzmanian stationary state with finite angular momentum and entropy production which, nonetheless, are in excellent agreement with detailed and integral fluctuation theorems.
Alkan Kabakçıoğlu is an Assistant Professor of Physics at Koç University. His research focuses on statistical mechanics and its applications to biological systems and complex networks. He previously held postdoctoral positions at the Weizmann Institute of Science and the University of Padova. He received his Ph.D. in Physics from MIT, and holds M.Sc. and B.Sc. degrees from Bilkent University.
Date: 26 November 2025, Wednesday
Time: 15:30
Place: SA-240
All interested are cordially invited.