Quality Not Quantity: On the Interaction between Dataset Design and
Robustness of CLIP
- URL: http://arxiv.org/abs/2208.05516v1
- Date: Wed, 10 Aug 2022 18:24:23 GMT
- Title: Quality Not Quantity: On the Interaction between Dataset Design and
Robustness of CLIP
- Authors: Thao Nguyen, Gabriel Ilharco, Mitchell Wortsman, Sewoong Oh, Ludwig
Schmidt
- Abstract summary: We introduce a testbed of six publicly available data sources to investigate how pre-training distributions induce robustness in CLIP.
We find that the performance of the pre-training data varies substantially across distribution shifts.
We find that combining multiple sources does not necessarily yield better models, but rather dilutes the robustness of the best individual data source.
- Score: 43.7219097444333
- License: http://creativecommons.org/licenses/by/4.0/
- Abstract: Web-crawled datasets have enabled remarkable generalization capabilities in
recent image-text models such as CLIP (Contrastive Language-Image pre-training)
or Flamingo, but little is known about the dataset creation processes. In this
work, we introduce a testbed of six publicly available data sources - YFCC,
LAION, Conceptual Captions, WIT, RedCaps, Shutterstock - to investigate how
pre-training distributions induce robustness in CLIP. We find that the
performance of the pre-training data varies substantially across distribution
shifts, with no single data source dominating. Moreover, we systematically
study the interactions between these data sources and find that combining
multiple sources does not necessarily yield better models, but rather dilutes
the robustness of the best individual data source. We complement our empirical
findings with theoretical insights from a simple setting, where combining the
training data also results in diluted robustness. In addition, our theoretical
model provides a candidate explanation for the success of the CLIP-based data
filtering technique recently employed in the LAION dataset. Overall our results
demonstrate that simply gathering a large amount of data from the web is not
the most effective way to build a pre-training dataset for robust
generalization, necessitating further study into dataset design.
Related papers
- Not All LLM-Generated Data Are Equal: Rethinking Data Weighting in Text Classification [7.357494019212501]
We propose efficient weighted-loss approaches to align synthetic data with real-world distribution.
We empirically assessed the effectiveness of our method on multiple text classification tasks.
arXiv Detail & Related papers (2024-10-28T20:53:49Z) - A CLIP-Powered Framework for Robust and Generalizable Data Selection [51.46695086779598]
Real-world datasets often contain redundant and noisy data, imposing a negative impact on training efficiency and model performance.
Data selection has shown promise in identifying the most representative samples from the entire dataset.
We propose a novel CLIP-powered data selection framework that leverages multimodal information for more robust and generalizable sample selection.
arXiv Detail & Related papers (2024-10-15T03:00:58Z) - Visual Data Diagnosis and Debiasing with Concept Graphs [50.84781894621378]
Deep learning models often pick up inherent biases in the data during the training process, leading to unreliable predictions.
We present CONBIAS, a novel framework for diagnosing and mitigating Concept co-occurrence Biases in visual datasets.
We show that by employing a novel clique-based concept balancing strategy, we can mitigate these imbalances, leading to enhanced performance on downstream tasks.
arXiv Detail & Related papers (2024-09-26T16:59:01Z) - Data Proportion Detection for Optimized Data Management for Large Language Models [32.62631669919273]
We introduce a new topic, textitdata proportion detection, which enables the automatic estimation of pre-training data proportions.
We provide rigorous theoretical proofs, practical algorithms, and preliminary experimental results for data proportion detection.
arXiv Detail & Related papers (2024-09-26T04:30:32Z) - FLIGAN: Enhancing Federated Learning with Incomplete Data using GAN [1.5749416770494706]
Federated Learning (FL) provides a privacy-preserving mechanism for distributed training of machine learning models on networked devices.
We propose FLIGAN, a novel approach to address the issue of data incompleteness in FL.
Our methodology adheres to FL's privacy requirements by generating synthetic data in a federated manner without sharing the actual data in the process.
arXiv Detail & Related papers (2024-03-25T16:49:38Z) - Federated Learning with Projected Trajectory Regularization [65.6266768678291]
Federated learning enables joint training of machine learning models from distributed clients without sharing their local data.
One key challenge in federated learning is to handle non-identically distributed data across the clients.
We propose a novel federated learning framework with projected trajectory regularization (FedPTR) for tackling the data issue.
arXiv Detail & Related papers (2023-12-22T02:12:08Z) - Omni-supervised Facial Expression Recognition via Distilled Data [120.11782405714234]
We propose omni-supervised learning to exploit reliable samples in a large amount of unlabeled data for network training.
We experimentally verify that the new dataset can significantly improve the ability of the learned FER model.
To tackle this, we propose to apply a dataset distillation strategy to compress the created dataset into several informative class-wise images.
arXiv Detail & Related papers (2020-05-18T09:36:51Z) - DeGAN : Data-Enriching GAN for Retrieving Representative Samples from a
Trained Classifier [58.979104709647295]
We bridge the gap between the abundance of available data and lack of relevant data, for the future learning tasks of a trained network.
We use the available data, that may be an imbalanced subset of the original training dataset, or a related domain dataset, to retrieve representative samples.
We demonstrate that data from a related domain can be leveraged to achieve state-of-the-art performance.
arXiv Detail & Related papers (2019-12-27T02:05:45Z)
This list is automatically generated from the titles and abstracts of the papers in this site.
This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.