proomt

Search

Search posts, papers, and topics

All posts

Hugging Face Daily PapersShuaijun Liu, Chengyu Wu, Qifu Wen1 min readpaperadvanced

D-JEPA: A Decision-Aligned Latent World Model

Read on arxiv.org

Summary

D-JEPA is a latent world model designed to bridge the gap between predicted outcomes and actual decision success in robotics. It learns decision-relevant relationships from executed actions, improving action selection by aligning latent space geometry with real-world results.

  • Latent world models can have a "decision-local prediction gap" where predicted closeness to a goal doesn't guarantee better real-world outcomes.
  • D-JEPA addresses this by learning decision-relevant relations among candidate futures from executed outcomes.
  • It employs a bounded, permutation-equivariant operator to refine predictive geometry where action choices are most consequential.
  • The model integrates with JEPA-compatible representations, enabling deployment through native latent-distance planning.

This paper is crucial for engineers and researchers in robotics and reinforcement learning, as it provides a method to make latent world models more reliable for real-world control and decision-making.

8/10
machine learningreinforcement learningroboticsworld modelscontrol systems

Related reading

  1. JEPA-Anything: Learning Predictive Models across Different Worlds

    JEPA-Anything extends joint‑embedding predictive architectures with orthogonal predictive factorization, letting a single model learn complementary latent factors that can be recombined for prediction across disparate domains. The paper shows consistent performance gains on ten dynamics tasks, molecular simulations, and clinical event forecasting, plus experimental validation of a biologically‑de…

    Hugging Face Daily Papersarxiv.org1 minpaper

  2. Tactile-JEPA: Topology-Aware Self-Supervised Representation Learning for Distributed Tactile Sensors

    Tactile-JEPA is a self‑supervised pre‑training technique that leverages the irregular graph layout of distributed tactile skins to learn richer representations. Experiments show it reduces force and orientation estimation errors and improves downstream policy learning across diverse sensor types.

    Hugging Face Daily Papersarxiv.org1 minpaper

  4. Laya the open source version of Jev

    Laya is an open‑source, bidirectional‑encoder model family for ultra‑fast, calibrated decision‑making (choice, score, boolean) over structured schemas. It runs 6‑8× faster than the closed‑source Jev, supports 100+ languages via three checkpoints, and includes a lightweight router that selects the appropriate checkpoint before inference. Benchmarks show higher accuracy, far better calibration (ECE…

    Hacker News front pageconvaiinnovations.com8 minreleaseHN1330314lobste.rs2

  5. Introducing System One Models and Jev

    TypeSafe AI announced its first “System One” model, Jev, a non‑text‑generating LLM that outputs type‑safe structured decisions with calibrated probabilities. It claims 40‑200× lower latency (70‑500 ms) and 100‑500× lower cost versus frontier LLMs, no hallucinations, and parallel sampling. The post includes a side‑by‑side demo, a custom “workflow” benchmark comparing Jev to GPT‑5.6/6 and other mod…

    Hacker News front pagetypesafe.ai9 minHN1824480lobste.rs26