Digest

Reset

1180 papers

2026-09-30 · arXiv:2609.32481

JEPA Learns What the Mask Leaves Unrecoverable

Peng Xie, Amr Alanwar

The authors study how JEPA masking geometry determines what can be recovered from context. They model a mask as a linear measurement: atoms whose support lies in the hidden region fall in the null space in a wavelet basis. JEPA must predict only sufficiency, enabling shortcuts with a moving-average target encoder; removing the shortcut depends on coarse-scale content left unrecoverable and on reachable context. They test predictions in 151 pre-training runs.

JEPAAI summary
2026-09-30 · arXiv:2609.32512

What Do Latent Predictive Vehicle Representations Retain? Measuring State, Geometry, and Local Response

Enzo Nicol\'as Spotorno, Josafat Leal Filho, Ant\^onio Augusto Fr\"ohlich

The authors present a measurement protocol for action-conditioned latent vehicle predictors with a physical readout. It separately tests retention of physical quantities, organization in latent space, one-step forecasting, and local response to small command perturbations via three matched response paths. In a case study on IPG CarMaker data, representations retain planar outputs, improved future-command inputs help 1s forecasts, but local command-response can diverge in latent space and cause regret in nearby-command ranking.

Latent DynamicsAI summary
2026-09-30 · arXiv:2609.32517

LocalProp: Neuro-Localized Memory-Efficient Backpropagation

Diana-Nicoleta Grigore, Iuliana Georgescu, Radu Tudor Ionescu

LocalProp is a training procedure that locally updates model weights by restricting gradient propagation to a single active module (e.g., one transformer block) while detaching other parts. The pipeline follows “pre-training then fine-tuning” using a local variant of I-JEPA, followed by structured pruning and short local recovery, aiming to reduce peak GPU memory and control the accuracy–memory trade-off via the number of jointly optimized blocks.

JEPAAI summary
2026-09-30 · arXiv:2609.32591

Think Fast, Plan Selectively: Adaptive Deliberation for Efficient Data-Driven MPC

Yi Xian Goh, Sze Jue Yang, Hao Luan

Fast-TD-MPC accelerates data-driven MPC by adaptively switching between a fast amortized policy (System 1) and TD-MPC2’s MPPI planner (System 2) using a lightweight OOD detector in the latent space. The authors evaluate on 103 continuous control tasks across 4 domains, achieving up to ~4× faster inference while maintaining competitive task performance and comparable robustness under external disturbances.

DrivingAI summary
2026-09-30 · arXiv:2609.32679

The GUI Is Not the State: Diagnosing State Aliasing in GUI World Models

Dongsheng Liu, Chao Jin, Wenkui Yang et al.

GUI world models often predict future interfaces using only the current GUI observation and action, which can cause state aliasing: identical visible conditions correspond to different valid futures due to hidden transition-relevant state. The authors introduce StateAliasBench (strict-pair diagnostic) and a predictive-state recovery method that infers structured state from history and condition frozen GUI world models to restore state-sensitive prediction and improve AndroidWorld agent performance.

DrivingAI summary
2026-09-30 · arXiv:2609.32779

Copper-Policy: Focus on the Representation for Robust Robot Manipulation

Zexin Feng, Yixu Feng, Lingyu Xiao et al.

Copper-Policy learns a compact, task-conditioned world representation jointly with the policy using temporal joint-embedding prediction. It predicts future observation embeddings (not pixels) and decodes actions from current-frame details plus the learned compact representation; no future is generated at test time. The authors report strong control and efficient training, including a 2B model trained in 9.67 hours on 8×RTX 5090 and faster training than Fast-WAM on matched A100 GPUs.

DrivingAI summary
2026-09-30 · arXiv:2609.32908

Predicting the Next State Is Not Enough: JEPA Representations for Lean Theorem Proving

Aarnav Choudhary

The authors study whether a JEPA-style latent proof-transition objective from one-step Lean transitions can rank kernel-validated successor states for search. On a same-theorem ranking diagnostic, JEPA has higher Top-1 than matched InfoNCE (50.18% vs 31.55%). But in fixed-budget kernel-checked search, JEPA solves fewer theorems (282.3 vs proposer ordering 308) and requires more tactic checks, indicating one-step transition ranking is insufficient long-horizon search value under this setup.

JEPAAI summary
2026-09-30 · arXiv:2609.32921

Adaptive Latent Capacity for World Models

Idan Achituve, Lior Dikstein, Idit Diamant et al.

Adaptive LeWorldModel (ALeWM) is a JEPA-style world model that learns to concentrate predictive information into compact prefixes of a wide latent embedding. It uses a sequence-conditioned capacity network to sample a prefix length during training, then regularizes masked embeddings with MixSIGReg against a Gaussian-active/zero-suffix mixture target. The learned ordering supports recursive planning with lower average planning capacity than fixed-width LeWM while achieving higher mean success rates.

JEPAAI summary
2026-09-30 · arXiv:2609.32928

Phenomenon-Graph JEPA: Label-Efficient Representation Learning for Contactless Cardiorespiratory Sensing

Constantino \'Alvarez Casado, Nhi Nguyen, Mohammad Rakibur Rahman et al.

Phenomenon-Graph JEPA is a joint-embedding predictive architecture for contactless cardiorespiratory sensing that uses four processed 1D streams, predicts stopped target embeddings along typed edges, and uses within-episode forward prediction without negative pairs or synthetic augmentation in the base setup. On OMuSense-23, the authors report improved label-efficiency area over matched supervised training (3.91 pp) under one configuration, but no advantage of physiological edge typing versus wrong-edge/all-pairs controls.

JEPAAI summary
2026-09-30 · arXiv:2609.32966

Self-Confirming Superposition Traps in Reinforcement Learning

Dai Shi, Andi Han, Feng Chen et al.

The authors show that RL’s representation-fitting loop can sustain a lower-return policy: in a self-confirming superposition trap, globally optimal codes under the current policy share overlapping directions for features that rarely co-activate, yet interfere after an alternative action. Interventions that preserve access to neglected states or protect replay weights reduce interference and can improve control, including in DreamerV3–Crafter.

Latent DynamicsAI summary
2026-09-30 · arXiv:2609.33030

What Must a World Model Distinguish for Planning?

Rongzhe Wei, Hans Hao-Hsun Hsu, Peizhi Niu et al.

The authors formalize when a world model must preserve physical distinctions for planning, via mechanism/response/decision sufficiency and query-dependent resolution requirements. Experiments on collision, learned nonlinear dynamics, and robotic planning show that needed information depends on the query, candidate set, and planning stage. They also compare query placement: a modular design (query-aware proposal + query-independent action-conditioned prediction) reuses physical predictions across objectives and achieves lower held-out regret than action-conditioned prediction methods.

DrivingAI summary
2026-09-30 · arXiv:2609.33110

D-JEPA: Design-Recoverable JEPA Representation with Swappable Physics Decoders

Nitin Nagesh Kulkarni, Aashwin Anand Mishra, Yin Yu et al.

D-JEPA learns a geometry-only latent representation and uses swappable, lightweight physics-specific decoders conditioned on operating conditions to predict physical fields. An explicit design-recoverability objective encourages the geometry latent to preserve underlying design variables for linear recovery and design optimization. The authors report that on four 3D benchmarks it maintains/improves full-field accuracy and achieves near-perfect linear recoverability while enabling decoder transfer.

JEPAAI summary
2026-09-30 · arXiv:2609.33172

Dynamic Manipulation with World-Action Models via Counterfactual Planning

Sunwoo Park, Wonbin Lee, Seonghyun Jin et al.

World–Action models trained on static demonstrations can fail on moving targets due to “target-response collapse” during reactive replanning. The authors propose Dynamic Predictive Planning (DPP), which predicts interaction timing using WAM rollouts, forecasts the target’s interaction position, synthesizes a counterfactual observation in a familiar robot context, and connects the resulting canonical plan to the robot’s live execution. DPP improves dynamic manipulation in simulation and on a real robot without additional training on dynamic data.

DrivingAI summary
2026-09-30 · arXiv:2609.33205

SMORE: Stability-Promoting Mesh-Agnostic Model Reduction for Time-Dependent PDEs

Yangyuan Li, Weichao Li, Shaowu Pan

SMORE is a mesh-agnostic reduced-order model for time-dependent PDEs. It uses an INR autodecoder to map sparse spatial measurements to a latent state, then evolves the latent state with structured latent dynamics (linear or linear-quadratic). A Lyapunov-guided stability regularization is added to promote stable long-horizon rollouts, and the authors provide theoretical guarantees under structural assumptions.

Latent DynamicsAI summary
2026-09-30 · arXiv:2609.33336

Beyond Conservatism: Recoverability-Conditioned Exploration for Model-Based Imitation Learning

Xuanlin Chen, Ziyue Wang, Xunlan Zhou et al.

RECON improves model-based imitation learning by separating conservative policy learning from active real-environment data collection. It trains a main (conservative) policy for task execution and an explorer optimized using epistemic uncertainty conditioned on recoverability estimated from multi-step imagination with the main policy, prioritizing uncertain yet recoverable dynamics. Experiments across locomotion, navigation, and manipulation show consistent gains in interaction efficiency, imitation performance, and robustness.

DrivingAI summary
2026-09-30 · arXiv:2609.33347

MultiEcho: An Experimental Science of Learned Worlds

Meng Zhu, Airui Zhang

MultiEcho treats frozen world models as experimental systems by estimating response laws from controlled counterfactual interventions. Using “three-reference” forward fits and reverse readouts, the authors predict complete intervention responses and recover intervention parameters across nine simulated physical systems and seven frozen model configurations, then test applicability and physical correspondence via temporal/visual/material intervention variants.

Latent DynamicsAI summary
2026-09-30 · arXiv:2609.33350

KoopCell: Koopman-Based Generative Model for Learning Single-Cell Dynamics from Distribution Snapshots

Wanfeng Lu, Yutong Zhang, Keyi Zhou et al.

KoopCell is a Koopman-based generative framework for learning continuous single-cell population dynamics from temporally sparse, unpaired distribution snapshots. It jointly learns latent coordinates and linear latent dynamics guided by Koopman–Mori–Zwanzig theory, with convergence guarantees from a weak continuity equation. KoopCell-M adds non-Markovian memory via Markovian embedding to model branching in latent dynamics.

JEPAAI summary
2026-09-30 · arXiv:2609.33487

What masking geometry works best for EEG foundation models?

Pierre Guetschel, Bruno Aristimunha, Yassine El Ouahidi et al.

The authors run a controlled sweep over spatio-temporal EEG masking geometry for 58 masked-prediction foundation models trained with MAE and JEPA, then evaluate all models on 12 OpenEEGBench datasets using linear probing. Both frameworks agree on an optimal mask configuration and shared failure modes; performance is robust outside the best region, while JEPA has an additional failure mode (“bias-inflation collapse”).

JEPAAI summary
2026-09-30 · arXiv:2609.33497

Hamiltonian JEPA: Action-Conditioned World Models with an Inherited Control State

Tamim Zoabi, Ameen Ali, Lior Wolf

H-JEPA is an action-conditioned, reconstruction-free world model that separates a wide perceptual code from a fixed orthonormal control state slice. The control state inherits the code’s covariance, then evolves with phase-conditioned dissipative port-Hamiltonian dynamics. Port-inverse consistency (PIC) ties the action readout to the input port transpose, shown to equal a rollout error projection reweighting prediction error. The authors report matching or exceeding reconstruction-free baselines on four pixel-based control benchmarks in ≤10 epochs.

2026-09-30 · arXiv:2609.33563

MA-JEPA: Joint-Embedding World Models for Multi-Agent Reinforcement Learning

Brandon Gary Kaplowitz, Osaze James Obahor, Christian Schroeder de Witt

MA-JEPA introduces a stochastic joint-embedding predictive world model for multi-agent RL with centralized training and decentralized execution. It replaces observation reconstruction with prediction of target observation embeddings, using a categorical latent state and a causal Transformer. A training-only joint predictor conditions on all agents’ local states/actions to predict each agent’s next embedding, which is passed through the same local posterior as in execution for actor-critic learning from latent imagination.

JEPAAI summary