VideoMAE

MCG-NJU (Nanjing University)

Structures found in this family (4)

Linear Direction, Linear Separability, Aristotelian Representation Hypothesis, Circle

By model (2)

VideoMAE-base

Linear Direction, Linear Separability, Aristotelian Representation Hypothesis

VideoMAE-v2

Linear Direction, Aristotelian Representation Hypothesis, Circle

Observations (6)

A physics-plausibility CAV steers a video world model

A directional-coherence score separates V-JEPA from cosine-similar rivals

Inverting video diffusion transformers makes physical plausibility linearly decodable

After correcting CKA and mutual k-NN alignment for a proven width/depth-dependent null-baseline confound (O(d/n) for global spectral metrics vs O(k/n) for local-neighborhood metrics), the Platonic Representation Hypothesis's global cross-modal convergence largely vanishes while its local-neighborhood convergence survives, across real image-text and video-language pretrained model pairs

Video world models encode motion direction as a circular population code

Video foundation models vary in physics decodability, V-JEPA strongest

Papers

Causal Physics Steering in Video World Models via Concept Activation Vectors (2026), Latent Video Prediction Learns Better World Models (2026), The Invisible Hand of Physics: When Video Diffusion Models Know More Than They Show (2026), Revisiting the Platonic Representation Hypothesis: An Aristotelian View (2026), Interpreting Physics in Video World Models (2026), Do Video Foundation Models Understand Intuitive Physics? A Layerwise Probing Analysis (2026)