Understanding how living systems dissipate energy is a central challenge in nonequilibrium biophysics. Traditional approaches to quantify entropy production often rely on detailed knowledge of microscopic dynamics, which is rarely accessible in experiments. This project aims to develop a machine learning-based framework to infer entropy production directly from experimental data, such as time-lapse movies of cellular processes or single-molecule trajectories. By leveraging advanced ML techniques—including neural networks, variational inference, and physics-informed models—we will train models to recognize patterns of irreversible dynamics and quantify dissipation without requiring full knowledge of underlying interactions. This approach will enable the systematic mapping of energy flows in complex biological systems, providing new insights into the thermodynamics of living matter. Our work will establish a scalable, data-driven methodology that bridges experimental observation and theoretical nonequilibrium physics, opening avenues for predictive modeling and control of biomolecular processes.