HyperSHAP: Shapley Values and Interactions for Hyperparameter Importance
Marcel Wever, Maximilian Muschalik, Fabian Fumagalli, Marius Lindauer · 2025
Read HyperSHAP: Shapley Values and Interactions for Hyperparameter Importance by Marcel Wever, Maximilian Muschalik, Fabian Fumagalli, Marius Lindauer (2025) 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-2502-01276.txt.
Description
Marcel Wever and co-authors wrote a paper from 2025 on explaining the tuning knobs of a learning algorithm with fair-share math. Tuning those knobs, called hyperparameters, decides much of a model's skill. Yet which knob matters, and how knobs interact, stays opaque, and black-box tuners give users little reason to trust them. The team borrows Shapley values from game theory, which split a team's payoff fairly among its players.
Their frame, named HyperSHAP, splits a model's score among the knobs that produced it. It also prices how pairs of knobs amplify or cancel each other. The split works knob by knob for one run, or pooled across many runs. Users can read it as an ablation guide, a measure of how much tuning headroom an algorithm holds, or a window on the tuner's own behavior. Tests across tuning benchmarks showed the readings are stable and usable.
Catalog record
| CREATOR | Marcel Wever, Maximilian Muschalik, Fabian Fumagalli, Marius Lindauer |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 71 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-01276