Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The authors provide formal identifiability guarantees for predictive self-supervised learning (SSL), resolving the theoretical paradox of how latent-space prediction differentiates between stochastic latent signals and unpredictable observation nuisance. They show that SSL implicitly instantiates a latent-variable model where predictive mutual information maximization retains forecast-relevant dynamics, while latent distribution matching constrains the representation geometry. Consequently, the framework provably recovers the ground-truth stochastic latent state up to an affine transformation, a result empirically validated on dynamic Gaussian systems.
Predictive SSL can provably distinguish intrinsic process stochasticity from random observation noise—resolving a foundational puzzle behind why non-generative latent prediction captures genuine world dynamics.
Self-supervised learning (SSL) by predicting in latent space, without generating the input data itself, learns highly abstract, useful representations. Intuitively, this success is often attributed to its ability to discard nuisance information that is irrelevant to prediction. However, this poses a conundrum: both stochastic variation in a prediction-relevant latent signal and true nuisance make observations partly unpredictable; how could they be distinguished? Surprisingly, we prove that common SSL methods can achieve exactly this, by implicitly instantiating a latent-variable model with stochastic dynamics and observation-private nuisance. We trace their ability to recover the stochastic signal to two complementary principles: Predictive mutual information maximization ensures that representations retain the information needed for prediction, while latent distribution matching constrains how this information is encoded, thereby making the retained signal identifiable. We confirm this identifiability result in simulations for Gaussian predictors, which recover the true signal up to an affine transformation even in dynamic, nuisance-laden environments.