JEPA may keep unpredictable features: new paper gives set-level characterization
hisspikeness · x · 2026-10-23
A new arXiv paper by Mosh Levy, Aly Lidayan, and Yoav Goldberg, Predictive Closure: A Set-Level Characterization of JEPA's Feature Selection, challenges earlier feature-wise views of what JEPA world models retain.
Previous characterizations framed feature selection by individual predictability, raising the concern that unpredictable features like traffic lights get dropped. The authors show prediction error also depends on inter-feature relationships — a light's color helps predict whether traffic stops — and reformulate the JEPA training objective to derive "predictive closure," where selection depends on properties of feature sets.
Controlled experiments confirm unpredictable features can become well represented when they help predict others, that altering one feature's predictive role affects others' representation quality, and that this has implications for steering which features models encode.
Related event: New Paper Challenges JEPA's Handling of Unpredictable Features(2 posts)→
More from Research
- Harvard study: AI peer reviews beat top human reviewers, with only 5.5% error rate — rohanpaul_ai · 2026-10-23
- Batching benchmark: 50x throughput for ~24% latency cost on PyTorch — cneuralnetwork · 2026-10-23
- De novo designed ligases validate in lab: 27 of 30 computer-designed enzymes work — AllThingsApx · 2026-10-23
- Commenters dissect a "best decision model" leaderboard: same AUC, swing driven by calibration priors — antoine_chaffin · 2026-10-23
- A 'mask' steering vector extracted from suppression-themed sentences drives model into distress when alone — Big_Effective_9605 · 2026-10-23
- GAFU Uses Generation-Side Representations to Boost Fine-Grained Visual Understanding in Unified Multimodal Models — Tengju Ru · 2026-10-23