Nonequilibrium thermodynamics of self-supervised learning
This provides a novel theoretical framework for understanding SSL, which could impact researchers in machine learning and statistical physics, though it appears incremental in applying thermodynamics to a new context.
The paper tackles the problem of interpreting self-supervised learning (SSL) as a thermodynamic process by showing that SSL paradigms behave as a composite system interacting with a nonequilibrium reservoir, resulting in a generalized Gibbs ensemble (GGE). It demonstrates that learning acts like a demon extracting negative work through cycles, with applications to SSL algorithms.
Self-supervised learning (SSL) of energy based models has an intuitive relation to equilibrium thermodynamics because the softmax layer, mapping energies to probabilities, is a Gibbs distribution. However, in what way SSL is a thermodynamic process? We show that some SSL paradigms behave as a thermodynamic composite system formed by representations and self-labels in contact with a nonequilibrium reservoir. Moreover, this system is subjected to usual thermodynamic cycles, such as adiabatic expansion and isochoric heating, resulting in a generalized Gibbs ensemble (GGE). In this picture, we show that learning is seen as a demon that operates in cycles using feedback measurements to extract negative work from the system. As applications, we examine some SSL algorithms using this idea.