arXiv subjects

Agnish Kumar Behera

Publications and source records attributed to Agnish Kumar Behera.

5 recordsLinked to original sources

arXiv2026

Non-equilibrium active noise enhances generative memory in diffusion models

Generative diffusion models have emerged as powerful tools for sampling high-dimensional distributions, yet they typically rely on white gaussian noise and noise schedules to destroy and reconstruct information. Here, we demonstrate that driving the generative process out of equilibrium using active, temporally correlated noise sources fundamentally alters the information thermodynamics of the system. We show that coupling the data to an active non-Markovian bath creates a `memory effect' where high-level semantic information (such as class identity or molecular metastability) is stored in the temporal correlations of auxiliary degrees of freedom. Using Fisher information analysis, we prove that this active mechanism significantly retards the rate of information decay compared to passive Brownian motion. Crucially, this memory effect facilitates an earlier and more robust symmetry breaking (speciation) during the reverse generative process, allowing the system to resolve multi-scale structures, reminiscent of metastable states in molecular configurations that are washed out in the typical noising processes. Our results suggest that non-equilibrium protocols, inspired by active matter physics, offer a thermodynamically distinct and potentially advantageous pathway for recovering high-dimensional energy landscapes using generative diffusion.

cs.LG↗

arXiv2024

A correspondence between Hebbian unlearning and steady states generated by nonequilibrium dynamics

The classic paradigms for learning and memory recall focus on strengths of synaptic couplings and how these can be modulated to encode memories. In a previous paper [A. K. Behera, M. Rao, S. Sastry, and S. Vaikuntanathan, Physical Review X 13, 041043 (2023)], we demonstrated how a specific non-equilibrium modification of the dynamics of an associative memory system can lead to increase in storage capacity. In this work, using analytical theory and computational inference schemes, we show that the dynamical steady state accessed is in fact similar to those accessed after the operation of a classic unsupervised scheme for improving memory recall, Hebbian unlearning or ``dreaming". Together, our work suggests how nonequilibrium dynamics can provide an alternative route for controlling the memory encoding and recall properties of a variety of synthetic (neuromorphic) and biological systems.

cond-mat.dis-nn↗

arXiv2023

Active oscillatory associative memory

Traditionally, physical models of associative memory assume conditions of equilibrium. Here, we consider a prototypical oscillator model of associative memory and study how active noise sources that drive the system out of equilibrium, as well as nonlinearities in the interactions between the oscillators, affect the associative memory properties of the system. Our simulations show that pattern retrieval under active noise is more robust to the number of learned patterns and noise intensity than under passive noise. To understand this phenomenon, we analytically derive an effective energy correction due to the temporal correlations of active noise in the limit of short correlation decay time. We find that active noise deepens the energy wells corresponding to the patterns by strengthening the oscillator couplings, where the more nonlinear interactions are preferentially enhanced. Using replica theory, we demonstrate qualitative agreement between this effective picture and the retrieval simulations. Our work suggests that nonlinearity in the oscillator couplings can improve memory under nonequilibrium conditions.

cond-mat.dis-nn↗

arXiv2022

Enhancing associative memory recall in non-equilibrium materials through activity

Associative memory, a form of content-addressable memory, facilitates information storage and retrieval in many biological and physical systems. In statistical mechanics models, associative memory at equilibrium is represented through attractor basins in the free energy landscape. Here, we use the Hopfield model, a paradigmatic model to describe associate memory, to investigate the effect of non-equilibrium activity on memory retention and recall. We introduce activity into the system as gaussian-colored noise which breaks detailed balance and forces the system out of equilibrium. We observe that, under these non-equilibrium conditions, the Hopfield network has a higher storage capacity than that allowed at equilibrium. Using analytical and numerical techniques, we show that the rate of entropy production modifies the energy landscape and helps the system to access memory regions which were previously inaccessible.

cond-mat.dis-nn↗

arXiv2021

A mechanism for the generation of robust circadian oscillations through ultransensitivity and differential binding affinity

Biochemical circadian rhythm oscillations play an important role in many signalling mechanisms. In this work, we explore some of the biophysical mechanisms responsible for sustaining robust oscillations by constructing a minimal but analytically tractable model of the circadian oscillations in the KaiABC protein system found in the cyanobacteria $S. \ elongatus$. In particular, our minimal model explicitly accounts for two experimentally characterized biophysical features of the KaiABC protein system, namely, a differential binding affinity and an ultrasensitive response. Our analytical work shows how these mechanisms might be crucial for promoting robust oscillations even in sub optimal nutrient conditions. Our analytical and numerical work also identifies mechanisms by which biological clocks can stably maintain a constant time period under a variety of nutrient conditions. Finally, our work also explores the thermodynamic costs associated with the generation of robust sustained oscillations and shows that the net rate of entropy production alone might not be a good figure of merit to asses the quality of oscillations.

physics.bio-ph↗