Understanding Influence Functions and Datamodels via Harmonic Analysis

Authors: Nikunj Saunshi, Arushi Gupta, Mark Braverman, Sanjeev Arora

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments are conducted on the CIFAR-10 data to test the estimation procedure and the quality of the linear fit in Figures 1 and 2. We use the FFCV library Leclerc et al. (2022) to train models on CIFAR-10; each model takes 30s to train on our GPUs.
Researcher Affiliation Academia Nikunj Saunshi, Arushi Gupta, Mark Braverman, Sanjeev Arora Department of Computer Science, Princeton University {nsaunshi, arushig, mbraverm, arora}@cs.princeton.edu
Pseudocode Yes Algorithm 1 Efficient algorithm for residual estimation
Open Source Code No The paper references the FFCV library (Leclerc et al., 2022) which is open-source, but does not state that the authors' *own* code for this paper's methodology or experiments is being released or is publicly available.
Open Datasets Yes Experiments are conducted on the CIFAR-10 data to test the estimation procedure and the quality of the linear fit in Figures 1 and 2.
Dataset Splits No The paper mentions training on subsets of CIFAR-10 and refers to 'test examples' but does not specify the train/validation/test splits or their sizes for reproducibility, beyond stating models were trained on sets of size 5000 from a 10k image subset.
Hardware Specification No The paper vaguely mentions "train on our GPUs" but does not provide specific GPU models, CPU models, or other detailed hardware specifications.
Software Dependencies No The paper mentions using "the FFCV library Leclerc et al. (2022)" but does not specify a version number for it or any other software dependencies.
Experiment Setup Yes We use the default Res Net based architecture in FFCV with a batch size of 512, an initial learning rate of 0.5, 24 epochs, weight decay of 5e-4 and SGD with momentum as the optimizer.