aiqbench analysis
Contrastive predictive objectives that sample negatives across trajectories risk encoding slow, trajectory-specific noise as the primary signal, which degrades downstream performance regardless of training scale. The study demonstrates that restricting negative sampling to within a single trajectory eliminates this shortcut, forcing the model to learn true dynamical variables. This finding offers a concrete design principle for improving self-supervised learning in physical systems where experimental observations are inherently noisy.
Contrast encodes inductive bias: separating slow noise from dynamics in predictive representation learning
arXiv:2606.07770v1 Announce Type: new Abstract: Self-supervised methods that learn representations and predict dynamics fully in the latent space, such as JEPA, have been shown to confuse slowly varying noise with the dynamical signals they aim to capture. Specifically, when noise features remain approximately constant within each trajectory, contrastive predictive objectives preferentially encode these features instead of the true latent variables governing the system. The learned representation then becomes dominated by trajectory-specific noise, so downstream performance degrades with noise strength and does not improve even as the number and duration of training trajectories increase. We argue that this failure is a property of the objective itself, shared by a long line of contrastive predictive objectives that sample negatives across trajectories. To illustrate this generality, we study the failure mode and its remedy in two settings: a standard SimCLR-style JEPA on a synthetic moving-dot dataset, and DySIB, a recently introduced method designed for extracting physically interpretable representations of dynamics, on movies of a rigid-body pendulum. When negatives are instead sampled within a single trajectory, the slow noise can no longer distinguish frames within that trajectory, removing the predictive shortcut. Training one encoder simultaneously on many such trajectories then forces it to encode the variables relevant for the dynamics, with longer trajectories yielding better representations even for strong slow noise. Our results point toward principles for designing contrastive predictive objectives in dynamical representation learning, especially for physical systems with noisy experimental observations.
Abstract:Self-supervised methods that learn representations and predict dynamics fully in the latent space, such as JEPA, have been shown to confuse slowly varying noise with the dynamical signals they aim to capture. Specifically, when noise features remain approximately constant within each trajectory, contrastive predictive objectives preferentially encode these features instead of the true latent variables governing the system. The learned representation then becomes dominated by trajectory-specific noise, so downstream performance degrades with noise strength and does not improve even as the number and duration of training trajectories increase. We argue that this failure is a property of the objective itself, shared by a long line of contrastive predictive objectives that sample negatives across trajectories. To illustrate this generality, we study the failure mode and its remedy in two settings: a standard SimCLR-style JEPA on a synthetic moving-dot dataset, and DySIB, a recently introduced method designed for extracting physically interpretable representations of dynamics, on movies of a rigid-body pendulum. When negatives are instead sampled within a single trajectory, the slow noise can no longer distinguish frames within that trajectory, removing the predictive shortcut. Training one encoder simultaneously on many such trajectories then forces it to encode the variables relevant for the dynamics, with longer trajectories yielding better representations even for strong slow noise. Our results point toward principles for designing contrastive predictive objectives in dynamical representation learning, especially for physical systems with noisy experimental observations.
Machine Learning (cs.LG)
arXiv:2606.07770 [cs.LG]
(or arXiv:2606.07770v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2606.07770 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 328 words
Related coverage
Disentanglement with Holographic Reduced Representations
arXiv · Jun 9, 2026
Inductive Biases in Field-Level Cosmological Inference from Galaxy Catalogs
arXiv · Sep 10, 2026
Does DiffusionGemma do latent reasoning?
AI Alignment Forum · Aug 16, 2026
The Loss Is Not Enough: Sampling Conditions and Inductive Bias in Contrastive Representation Learning
arXiv · Jun 5, 2026