AP-Perf: Incorporating Generic Performance Metrics in Differentiable Learning
This work addresses a practical issue for machine learning practitioners by enabling easier integration of custom metrics, though it appears incremental as it builds on existing adversarial prediction frameworks.
The authors tackled the problem of incorporating custom non-decomposable performance metrics into differentiable learning pipelines, such as neural networks, by developing a method based on adversarial prediction and marginal distribution techniques, resulting in demonstrated effectiveness on classification tasks with tabular and image datasets.
We propose a method that enables practitioners to conveniently incorporate custom non-decomposable performance metrics into differentiable learning pipelines, notably those based upon neural network architectures. Our approach is based on the recently developed adversarial prediction framework, a distributionally robust approach that optimizes a metric in the worst case given the statistical summary of the empirical distribution. We formulate a marginal distribution technique to reduce the complexity of optimizing the adversarial prediction formulation over a vast range of non-decomposable metrics. We demonstrate how easy it is to write and incorporate complex custom metrics using our provided tool. Finally, we show the effectiveness of our approach various classification tasks on tabular datasets from the UCI repository and benchmark datasets, as well as image classification tasks. The code for our proposed method is available at https://github.com/rizalzaf/AdversarialPrediction.jl.