Characterization of Overfitting in Robust Multiclass Classification

Authors: Jingyuan Xu, Weiwei Liu

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical In this paper, we derive both upper and lower bounds of h U(k, n, m), and demonstrate that our upper bounds and lower bounds are matching within logarithmic factors when n and the distribution of test dataset features DX are fixed. Next, we give a brief overview of proof techniques used to obtain the main results. We first note that throughout this paper we use the notion of corrupted hypothesis [28], which transforms the formulation of robust accuracy to a non-robust one thus greatly simplifying the proofs. The definition of corrupted hypothesis is presented in the beginning of Section 3. We establish the upper bounds via minimum description length argument, following closely a proof of an analogous result by [5] for non-robust setting. ... To obtain the lower bounds, we propose computationally efficient algorithms for two regions of k respectively. The algorithms are modified from [6]...
Researcher Affiliation Academia Jingyuan Xu Weiwei Liu School of Computer Science, Wuhan University National Engineering Research Center for Multimedia Software, Wuhan University Institute of Artificial Intelligence, Wuhan University Hubei Key Laboratory of Multimedia and Network Communication Engineering, Wuhan University {jingyuanxu777,liuweiwei863}@gmail.com
Pseudocode Yes Algorithm 1 Asmall (k = 1); Algorithm 2 Asmall (k > 1); Algorithm 3 Abig(C)
Open Source Code No The paper is a theoretical work focusing on deriving mathematical bounds and does not provide any statements regarding the release of open-source code for the described methodologies.
Open Datasets No The paper is theoretical and focuses on mathematical derivations. It does not mention or utilize specific datasets, public or otherwise, for training or evaluation.
Dataset Splits No The paper is a theoretical work and does not describe any experimental setup involving training, validation, or test dataset splits.
Hardware Specification No The paper is theoretical and does not describe any experiments or computations that would require specific hardware specifications.
Software Dependencies No The paper is purely theoretical and does not mention any software dependencies or their specific version numbers.
Experiment Setup No The paper is theoretical and focuses on mathematical derivations; it does not include details on experimental setup, hyperparameters, or system-level training settings.