On Benchmark Hacking in ML Contests: Modeling, Insights and Design
For contest hosts and ML practitioners, this provides a theoretical framework to understand and mitigate benchmark hacking in competitive settings.
The paper models benchmark hacking in ML contests as a strategic game where contestants allocate effort between creative and mechanistic types. It finds that low-type contestants always hack, and more skewed reward structures reduce hacking, supported by empirical evidence.
Benchmark hacking refers to tuning a machine learning model to score highly on certain evaluation criteria without improving true generalization or faithfully solving the intended problem. We study this phenomenon in a generic machine learning contest, where each contestant chooses two types of effort: creative effort that improves model capability as desired by the contest host, and mechanistic effort that only improves the model's fitness to the particular task in contest without contributing to true generalization. We establish the existence of a symmetric monotone pure strategy equilibrium in this competition game. It also provides a natural definition of benchmark hacking in this strategic context by comparing a player's equilibrium effort allocation to that of a single-agent baseline scenario. Under our definition, contestants with types below certain threshold (low types) always engage in benchmark hacking, whereas those above the threshold do not. Furthermore, we show that more skewed reward structures (favoring top-ranked contestants) can elicit more desirable contest outcomes. We also provide empirical evidence to support our theoretical predictions.