Does Distributionally Robust Supervised Learning Give Robust Classifiers?

Authors: Weihua Hu, Gang Niu, Issei Sato, Masashi Sugiyama

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Motivated by our analysis, we propose simple DRSL that overcomes this pessimism and empirically demonstrate its effectiveness. Finally, we demonstrate the effectiveness of our DRSL through experiments (Section 6).
Researcher Affiliation Academia 1University of Tokyo, Japan 2RIKEN, Tokyo, Japan.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any concrete access information for source code.
Open Datasets Yes We obtained six classification datasets from the UCI repository (Blake & Merz, 1998), two of which are for multi-class classification. We also obtained MNIST (Le Cun et al., 1998) and 20newsgroups (Lang, 1995).
Dataset Splits Yes The regularization hyper-parameter λ was selected from {1.0, 0.1, 0.01, 0.001, 0.0001} via 5-fold cross validation.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions general software components like 'linear models with softmax output' and 'cross-entropy loss', but does not specify any software names with version numbers.
Experiment Setup Yes For all the methods, we used linear models with softmax output for the prediction function gθ(x). The cross-entropy loss with ℓ2 regularization was adopted. The regularization hyper-parameter λ was selected from {1.0, 0.1, 0.01, 0.001, 0.0001} via 5-fold cross validation. We used the two f-divergences (the KL and PE divergences) and set δ = 0.5 for AERM and structural AERM.