I-JEPA and V-JEPA proved feature-space prediction works: 81.9% on Kinetics-400
TheTuringPost · x · 2026-11-21
From TheTuringPost's thread: 2023–2024 proved predicting in feature space works. I-JEPA (Jan 2023) predicts representations of other image blocks from one block, with no hand-crafted augmentations; V-JEPA (Feb 2024) applied the same idea to 2M videos, with a frozen ViT-H/16 reaching 81.9% on Kinetics-400.
More from Research
- WorldTrace beats sliding window recall by 32.6% at same cache size, tops all 12 LoopBench conditions — cindy_x_wu · 2026-11-21
- Proposed proof of Generalized Riemann Hypothesis released; mathematician stresses it is entirely human, no AI — paulabartabajo_ · 2026-11-21
- Arcfront turns real hospital and factory work records into RL environments and frontier model evals — rohanpaul_ai · 2026-11-21
- Goodfire probes Inkling 2 Mini: looped model layers read from "future" layers — fatihdin4en · 2026-11-21
- Terence Tao offers a bet to anyone who really believes RH will be proved next year — jaivin · 2026-11-21
- Terence Tao: AI is winning at math but we've learned almost nothing from it — bansalg_ · 2026-11-21