Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

CS-Shapley: Class-wise Shapley Values for Data Valuation in Classification

About

Data valuation, or the valuation of individual datum contributions, has seen growing interest in machine learning due to its demonstrable efficacy for tasks such as noisy label detection. In particular, due to the desirable axiomatic properties, several Shapley value approximation methods have been proposed. In these methods, the value function is typically defined as the predictive accuracy over the entire development set. However, this limits the ability to differentiate between training instances that are helpful or harmful to their own classes. Intuitively, instances that harm their own classes may be noisy or mislabeled and should receive a lower valuation than helpful instances. In this work, we propose CS-Shapley, a Shapley value with a new value function that discriminates between training instances' in-class and out-of-class contributions. Our theoretical analysis shows the proposed value function is (essentially) the unique function that satisfies two desirable properties for evaluating data values in classification. Further, our experiments on two benchmark evaluation tasks (data removal and noisy label detection) and four classifiers demonstrate the effectiveness of CS-Shapley over existing methods. Lastly, we evaluate the "transferability" of data values estimated from one classifier to others, and our results suggest Shapley-based data valuation is transferable for application across different models.

Stephanie Schoch, Haifeng Xu, Yangfeng Ji• 2022

Related benchmarks

TaskDatasetResultRank
Noisy label detectionCovertype
AUC0.712
20
Noisy label detectionClick
AUC0.855
18
Noisy label detectionphoneme
AUC0.579
18
Label Noise IdentificationMNIST (train)
AUC0.877
15
Noisy label detectionCPU
AUC0.808
13
Noisy label detectionCIFAR10
AUC0.45
11
High-value data removalCIFAR10 binarized (test)--
11
High-value data removalClick (test)
Weighted Accuracy Drop0.4
8
Noisy label detectionDiabetes
AUC0.441
8
Noisy label detectionFMNIST
AUC57
8
Showing 10 of 23 rows

Other info

Code

Follow for update