Tag
This paper establishes a formal correspondence between the SIGReg objective used in Joint-Embedding Predictive Architectures (JEPAs) and the Active Inference framework, showing that under certain conditions the training objective becomes an exact variational free energy. It organizes non-contrastive regularisers into an entropy-estimator hierarchy and proves theoretical results including exact information bottleneck and preserved surprise bounds for SIGReg.
A solo-built multi-agent cognitive architecture uses hyperbolic geometry on a Poincaré ball manifold, variational free energy for belief updating, and wave interference for memory retrieval, allowing personality to emerge from memory interactions rather than scripting.
Introduces MODE-RAG, a multi-agent system using Variational Free Energy and Monte Carlo Tree Search to dynamically gate interventions for mitigating hallucinations in Multimodal Retrieval-Augmented Generation systems, along with the ModeVent evaluation dataset.