Mol-JEPA: A Multimodal JEPA Foundation Model for Molecules, One Year in the Making
TerribleAntelope9348 · reddit · 2026-09-04
A Reddit user shared Mol-JEPA, a research project they've been working on for about a year, with a summary website of key results.
- Core idea: extends the JEPA (joint-embedding predictive architecture) paradigm to chemistry, building a multimodal molecular foundation model
- The author notes performance still has room for improvement and invites community feedback and ideas
- Early-stage work, self-published as [R]
More from Research
- NVIDIA Parabricks HaplotypeCaller lands in nf-core/sarek for GPU-accelerated germline calling — AllThingsApx · 2026-09-04
- An image editing model can directly edit video latents via two zero-training projection layers — qixing_huang · 2026-09-04
- Small Probe-Based Judges Can Replace Large Models for Rubric-Based RL Rewards — Fengyu Xie · 2026-09-04
- New blog unpacks removing the square root in Adam: what second-order methods reveal — burny_tech · 2026-09-04
- Terence Tao predicts AI will be a trustworthy co-author in math research by 2026 — burny_tech · 2026-09-04
- True Positive Weekly #176: chatbot trust, recursive self-improvement, vLLM speculative decoding — burkov · 2026-09-04