Generator Assisted Mixture of Experts for Feature Acquisition in Batch

Authors: Vedang Asgaonkar, Aditya Jain , Abir De

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments with four datasets show that our approach outperforms these methods in terms of trade-off between accuracy and feature acquisition cost.
Researcher Affiliation Academia Indian Institute of Technology Bombay {vedang, adityajainjhs, abir}@cse.iitb.ac.in
Pseudocode Yes Algorithm 1: Training. Algorithm 2: Inference.
Open Source Code Yes Our code is in https://github.com/Vedang Asgaonkar/genex
Open Datasets Yes We experiment with four datasets for the classification task; DP (disease prediction), MNIST, CIFAR100 and Tiny Imagenet (TI). Details are provided in the extended version (Asgaonkar, Jain, and De 2023). (Extended version of current paper). ar Xiv preprint arxiv:2312.12574.
Dataset Splits Yes We split the entire dataset in 70% training, 10% validation and 20% test set.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, memory, or cloud instance types) used for running the experiments.
Software Dependencies No The paper mentions software components like "Python", "PyTorch", "β-VAE", "Wide Resnet", and "Efficient Net" but does not provide specific version numbers for any of them.
Experiment Setup Yes We set the number of buckets B = 8, 8, 4, 4 for DP, MNIST, CIFAR100 and Tiny Imagenet using cross validation. given a budget qmax for maximum number of oracle queries for each instance. where λ is a hyperparameter.