Researchers studied the impact of semantic pairs on self-supervised representation learning in AI, finding that semantic pairs improve generalization and induce invariances beyond standard transformations. This has implications for selecting and designing frameworks for AI agents that can effectively exploit semantic pair information.

Researchers propose a method for training AI models with noisy labels by pre-training a feature extractor using self-supervised learning, improving model robustness and accuracy in noisy environments.

A self-supervised framework, Distributed JEPA, is proposed for energy forecasting. It uses a distributed Joint Embedding Predictive Architecture to predict latent representations of masked temporal segments and integrates temporal observations and contextual information. JEPA is evaluated on energy consumption and generation datasets and shows comparable or better performance than a Transformer baseline.