Task-Sufficient Contraction: Source Selection for Machine Information Interfaces
A theory paper shows when a task-fixed source reduction preserves the full downstream rate-regret problem.
The paper defines Task-Sufficient Contraction: a source reduction fixed by a declared task that still preserves the full downstream problem family. For a machine with fixed actions and loss, states are merged only when every action has the same regret, preserving the one-step rate-regret curve for finite action sets. For quadratic loss on affine actions, the canonical reduction is the projection onto directions where feasible actions differ. The authors relate the idea to Information Bottleneck, semantic rate-distortion, and goal-oriented quantization, including cases where contractions fail or must be corrected.
- Defines Task-Sufficient Contraction before choosing rate or distortion.
- Finite actions preserve the full one-step rate-regret curve.
- Quadratic loss reduces to a projection of differing action directions.
- Distinguishes exact, approximate, failed, and corrected contractions.
From the original article
1 min read at huggingface.co
A declared task can sometimes certify a reduced source before a downstream encoder, codebook, rate, distortion target, or optimizer is chosen. This paper studies when one such reduction preserves the complete downstream problem family, a property termed Task-Sufficient Contraction. The reduced source is fixed by the task before the later operating point is selected.
By Joss Armstrong · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.