GRAM accepted to NeurIPS 2026: recursive reasoning becomes stochastic latent trajectories
SungjinAhn_ · x · 2026-09-25
GRAM (Generative Recursive reAsoning Models) from KAIST, Mila, NYU and Yoshua Bengio is accepted to NeurIPS 2026.
- Existing recursive reasoners (HRM, TRM, Looped Transformers) are deterministic: the same input always converges to one answer, collapsing all plausible reasoning paths into a single attractor.
- GRAM models recursion as a stochastic latent trajectory, keeping multiple hypotheses and alternative strategies, with inference-time scaling in both depth and width (parallel trajectory sampling).
- The same formulation supports conditional reasoning p(y|x) and unconditional generation p(x). Trained with amortized variational inference, it beats deterministic baselines on structured reasoning and multi-solution constraint satisfaction; revised paper and code coming soon.
More from Research
- SkillsBench accepted at NeurIPS 2026: 87-task benchmark tests whether agent skills actually help — HanchungLee · 2026-09-25
- FineVision, the 17M-image open VLM dataset from 200+ sources, accepted to NeurIPS — andimarafioti · 2026-09-25
- High-BW TFLN Modulator Demo, but the Dual-Band Grating Coupler Steals the Show — jwt0625 · 2026-09-25
- MindGames, a benchmark for detecting lying AI agents, accepted at NeurIPS after 30K-game competition — LChoshen · 2026-09-25
- Alignment Is Likely Spiky Too: Models May Be Aligned in Some Domains, Misaligned in Others — nabla_theta · 2026-09-25
- NeurIPS paper accepted with 5/5/4 scores, yet final review cites AI-use concerns — Rich-Mycologist-75 · 2026-09-25