AI
One Proposal for Every Margin: Zero-Shot Amortized Sequential Importance Sampling for Binary Matrices
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 30/09/2026, 04:50:48 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 04:18:24 UTC.
Briefings
A new framework called MarginFlow uses generative flow networks to propose samples for counting and sampling binary matrices with fixed row and column sums. By treating the proposal distribution as a learnable policy, MarginFlow amortizes the design process across different matrix margins. The model was trained on 1904 margins and tested zero-shot on 1190 held-out margins, achieving a median effective sample fraction of 99.8% on most cases.
Key points
- MarginFlow uses generative flow networks as the ideal proposal distribution for Sequential Importance Sampling (SIS) to count and sample uniformly from binary matrices with fixed margins.
- The framework amortizes the proposal design across margins by treating partial matrices as reduced-margin instances served by a single set transformer.
Context and limitations
Evaluation was limited to 1190 held-out margins from 3x3 to 870x6 matrices, with results based on a single training pool of 1904 margins.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/09/2026, 04:50:48 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeihfolaag2r2drwbsxdrs765d6cz4yxunininn6yuyqadopvmb327a