Information Retention via Learning Supplemental Features

Authors: Zhipeng Xie, Yahe Li

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

Reproducibility Variable Result LLM Response
Research Type Experimental 3 EXPERIMENTS This section is devoted to a thorough empirical study of the proposed Info R-LSF method
Researcher Affiliation Academia Zhipeng Xie*, Yahe Li* (* Equal Contribution) School of Computer Science, Fudan University, Shanghai 200434, China xiezp@fudan.edu.cn, yaheli21@m.fudan.edu.cn
Pseudocode Yes C ALGORITHM TABLE OF INFOR-LSF
Open Source Code Yes 1Code available at https://github.com/liyahe/Info R-LSF
Open Datasets Yes We use two image classification datasets, CIFAR10 (Krizhevsky et al., 2009) and CIFAR100(Krizhevsky et al., 2009). We use two sentiment analysis datasets, namely IMDB (Maas et al., 2011) and YELP (Zhang et al., 2015).
Dataset Splits Yes Table 7: Datasets used in experiments.
Hardware Specification No The paper provides computational complexity analysis in Section A.4 and Table 9 with running times, but it does not specify any particular hardware components such as GPU models, CPU types, or memory used for these experiments or for training.
Software Dependencies No The paper mentions optimizers (SGD, AdamW) and backbone networks (ResNet-18, BERTBASE) and refers to a PyTorch implementation in a footnote, but it does not specify exact version numbers for any software libraries or dependencies (e.g., 'PyTorch 1.x' or 'Python 3.x').
Experiment Setup Yes For CIFAR10, ... train the model for 200 epochs with batch size 256, initial learning rate 0.1 and weight decay 5e-4. We apply SGD optimizer with momentum 0.9 and a step scheduler that decays the learning rate by 0.1 every 160 epochs.