Learning to Help in Multi-Class Settings

Authors: Yu Wu, Yansong Li, Zeyu Dong, Nitya Sathyavageeswaran, Anand Sarwate

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show that our proposed methods offer an efficient and practical solution for multi-class classification in resource-constrained environments.
Researcher Affiliation Academia Rutgers University, University of Illinois Chicago, Stony Brook University
Pseudocode Yes Algorithm 1 Optimization With Our Surrogate Loss Function
Open Source Code No The paper does not provide an explicit statement about open-sourcing code or a link to a code repository. No mention of code in supplementary materials.
Open Datasets Yes In this section, we test the proposed surrogate loss function in equation 7 and algorithms for different settings on CIFAR-10 (Krizhevsky & Hinton, 2009), SVHN (Netzer et al., 2011) , and CIFAR-100 (Krizhevsky & Hinton, 2009) datasets.
Dataset Splits Yes CIFAR-10 consists of 32 × 32 color images drawn from 10 classes and is split into 50000 training and 10000 testing images.
Hardware Specification Yes The experiments are conducted in RTX 3090.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes To reduce the computation, we use Stochastic Gradient Descent (SGD) for presentation. Specifically, we choose ce from an interval between [0, 0.5] with fixed inaccuracy costs c1 = 1 and c1 = 1.25. In our experiments, the base network structure for the client classifier and the rejector is Le Net-5, and the server classifier is either Alex Net or Vi T.