AI

AutoAssumption: Searching over Assumptions Discovers ML-Systems Algorithms

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 13/11/2026, 07:48:24 UTC.

Briefings

AutoAssumption formalizes 89 ML algorithms by treating them as proof graphs with explicit assumptions, enabling automated search over assumptions to generate and audit code. The system improved on published methods in KV-cache eviction and weight pruning, achieving the highest RULER-4K score among tested methods and the lowest perplexity on four LLaMA models. The harness scores within a point of OpenEvolve, an open-source implementation of AlphaEvolve, with lower standard deviation across repeats.

Key points

• AutoAssumption formalizes 89 ML algorithms into proof graphs with explicit assumptions, allowing automated search over assumptions to generate and audit code.
• The system improved on published methods in KV-cache eviction and weight pruning, achieving the highest RULER-4K score among tested methods and the lowest perplexity on four LLaMA models.

Context

This research addresses the challenge of ML algorithms rarely stating their assumptions, making it difficult to attribute gains to specific edits. By formalizing algorithms as proof graphs, AutoAssumption turns design into a structured search space auditable by LLMs and proof checkers.

Limitations

Evidence is limited to a single preprint (arXiv:2611.14844v1) from November 2026, with no experimental results beyond the initial harness runs. The study does not report long-term stability, deployment metrics, or comparative analysis against commercial systems.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 13/11/2026, 08:04:33 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeig7i44kksrn4ibmfdtdcschatjerzuiddsitosbakgmpo2kkiogj4