On the Generalization of Multi-modal Contrastive Learning

Authors: Qi Zhang, Yifei Wang, Yisen Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this paper, we propose the first theoretical analysis on the generalization ability of MMCL... Based on this perspective, we compare MMCL and SSCL on real-world data and show that text-induced positive pairs have better semantic consistency and diversity... which validates our understanding of the superiority of multi-modal positive pairs. ...we propose four different techniques and they both bring improvements (as much as 6.2%) on Image Net.
Researcher Affiliation Academia 1National Key Lab of General Artificial Intelligence, School of Intelligence Science and Technology, Peking University 2School of Mathematical Sciences, Peking University 3Institute for Artificial Intelligence, Peking University.
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Code is available at https://github. com/PKU-ML/CLIP-Help-Sim CLR.
Open Datasets Yes we pretrain the same backbone Vi T-B (Dosovitskiy et al., 2021) on the same dataset, YFCC15M (Thomee et al., 2016; Radford et al., 2021), and evaluate the learned representations on Image Net (Deng et al., 2009a).
Dataset Splits Yes For efficiency, we randomly draw 1,000 samples from 10 random classes of the Image Net validation set.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory specifications) used for running its experiments.
Software Dependencies No The paper mentions software components like SimCLR, CLIP, ResNet-50, and specific optimizers (LARS, SGD), but it does not provide specific version numbers for these software dependencies or programming languages used.
Experiment Setup Yes We train the encoder for 100 epochs on Image Net with 512 batch size and use the LARS optimizer with a cosine annealed learning rate schedule. ...we train a linear classifier following the frozen backbones and optimize the Cross Entropy loss with the SGD optimizer.