For self-supervised learning, Rationality implies generalization, provably

Authors: Yamini Bansal, Gal Kaplun, Boaz Barak

ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We also conduct an extensive empirical study of the generalization gap and the quantities used in our assumptions for a variety of self-supervision based algorithms, including Sim CLR, AMDIM and Big Bi GAN, on the CIFAR-10 and Image Net datasets.
Researcher Affiliation Academia Yamini Bansal Harvard University Gal Kaplun Harvard University Boaz Barak Harvard University
Pseudocode No The paper describes algorithms conceptually and mathematically but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes We provide our code and data in an anonymous repository on: http://github.com/ICLR2021-rep-gen/.
Open Datasets Yes We perform an empirical study of the RRM bound for a wide variety of self-supervised training methods on the Image Net (Deng et al., 2009) and CIFAR-10 (Krizhevsky et al., 2009) training datasets.
Dataset Splits No The paper frequently mentions 'train' and 'test' data but does not explicitly describe a 'validation dataset split' with specific percentages, counts, or methodology for its experiments.
Hardware Specification No The paper does not specify the hardware used for running the experiments (e.g., GPU models, CPU types, or cloud instance details).
Software Dependencies No The paper describes optimization algorithms and hyperparameters but does not list specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes Table H.3 Summary of training methods with their hyper-parameters for CIFAR-10; Table H.4 Summary of training methods with their hyper-parameters for Image Net.