Market-Conditioned JEPA learns asset signals from realized market shocks
On 11 years of minute-level Chinese A-share data, it beat per-asset JEPA on returns, exposure, risk and liquidity.
Stories per day · 30 days
1 RSS
Relatedrepresentation-learning4self-supervised-learning2world-models2android1electronic-health-records1feature-selection1financial-time-series1gui-agents1interpretability1latent-worlds1market-factors1masked-language-modeling1mimic-iv1mobile1multimodal1object-slots1planning1platonic-representation1
27 stories, newest first · one card per story
Stories tagged #jepa
Paper turns the Platonic Representation Hypothesis into a JEPA theorem
A 109M-parameter text JEPA beat language models up to 72B on alignment with DINOv2 and CLIP image encoders.
GUI-JEPA scores mobile GUI actions in latent space without decoding
On AndroidControl it is 2.8–4.0x faster than a fine-tuned Qwen2.5-VL-7B agent while using 6% of the parameters.
JEPA pretraining underperforms masked models on MIMIC-IV EHRs
Three joint-embedding formulations trained without collapse but still lost to standard masked language modelling downstream.
Slot-JEPA adds object slots and masks to latent world models
Frozen slot features stay transferable on seven benchmarks and improve planning on six planar and four robot tasks.
Paper characterizes JEPA feature selection as predictive closure
Unpredictable features can still be encoded when they help predict other features.
Perspective maps a four-stage agenda for JEPA latent-world interpretability
Authors say existing work only covers grounding, while structure, causal checks, and intervention remain largely unexplored.
RSI-Anything: Unlocking Recursive Self-Improvement with World Models
RSI-Anything uses predictive world models to recursively improve policies across code, GUI, lab, and chemistry tasks.
RSI-Anything: Unlocking Recursive Self-Improvement with World Models
RSI-Anything uses world models to select actions and updates, improving code repair, GUI tasks, and lab planning.
LeWAM: A JEPA World Action Model with Diffusion-Steering-Based MPC
LeWAM is a decoder-free JEPA world action model that predicts dynamics and policies while planning in policy-head noise space.
RoboJEPA: Scaling Robotic Latent World Models
RoboJEPA, an 8B JEPA world model on 12 embodiments, reports compute scaling laws for robotic planning.
Researchers stretch LeCun's JEPA AI into a universal world model that works from physics to biology
Researchers unveil JEPA-Anything, a multi-predictor world model spanning physics through biology.
9
Beyond Domain-Specific World Models: JEPA-Anything Uses 1 Recipe for 7 Fields
JEPA-Anything factorizes latent targets into four predictors and beats JEPA baselines across seven domains.
H-JEPA: End-to-End Learning of Hierarchical World Models for Visual Planning
H-JEPA trains hierarchical action-conditioned world models, raising Visual AntMaze planning success from 18% to 73%.
JEPA-TTT: Persistent Test-Time Training of Latent World Models for Planning under Dynamics Shifts
JEPA-TTT adapts a pretrained latent world model at test time when environment dynamics shift.
FlexiWorld: Learning and Planning via Flexible Action Chunks Across Multiple Time Scales
FlexiWorld uses variable-length action chunks and mixed-span goals, reaching 89.29 percent mean success.
PDE-JEPA: Predictive Representation Learning of Latent Dynamics Modeling for Parametric PDEs
PDE-JEPA uses masked-latent predictive pretraining with geometry alignment for parametric PDE dynamics, beating SOTA by 33.4% in-distribution and 51.4% extrapolating.
What masking geometry works best for EEG foundation models?
An EEG masking ablation matches REVE-level decoding performance at a fraction of the pre-training compute.
DroneWAM: Efficient World Action Model for Drone Visual Navigation
DroneWAM predicts drone navigation futures in latent space, cutting rollout depth while improving accuracy.
TT-VidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining
TT-VidT pairs a DINOv3 spatial ViT with a temporal layer, leading motion benchmarks at lower FLOPs.
Adaptive Latent Capacity for World Models
ALeWM learns compact latent prefixes for JEPA world models and beats fixed-width LeWM.
Underwater C3-JEPA: An Object-Centric Cross-View World Model for ROV Salvage
Researchers introduce Underwater C3-JEPA, a control-conditioned world model for near-field ROV salvage.
D-JEPA: A Decision-Aligned Latent World Model
D-JEPA aligns latent world-model distances with executed outcomes, improving robot and driving action selection.
Tactile-JEPA: Topology-Aware Self-Supervised Representation Learning for Distributed Tactile Sensors
Tactile-JEPA pretrains topology-aware encoders for sparse robot tactile skins using graph-guided masking.