Model-Aware Rate–Distortion Limits for Task–Oriented Source Coding
Andriy Enttsel, Vincent Corlay · 2026
Read Model-Aware Rate–Distortion Limits for Task–Oriented Source Coding by Andriy Enttsel, Vincent Corlay (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-12866.txt.
Description
When machines, not people, consume images, compression should preserve what the task needs rather than what the eye likes. Andriy Enttsel and Vincent Corlay revisit the fundamental limits of task-oriented source coding through indirect rate-distortion theory. Earlier bounds often assumed tasks were perfectly identifiable and ignored the quirks of the model actually deployed.
The authors spell out when existing bounds can be reached and where they break in realistic settings. They then propose model-aware bounds that account for a task model's flaws and architectural limits. Tests on standard classification benchmarks show current learned schemes operating far from those limits, with transmitter-side complexity as a key bottleneck. The gap points to where future coding designs should focus.
Catalog record
| CREATOR | Andriy Enttsel, Vincent Corlay |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 37 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-12866