Measuring Bias of Web-filtered Text Datasets and Bias Propagation Through Training

• URL: http://arxiv.org/abs/2412.02857v1
• Date: Tue, 03 Dec 2024 21:43:58 GMT
• Title: Measuring Bias of Web-filtered Text Datasets and Bias Propagation Through Training
• Authors: Youssef Mansour, Reinhard Heckel,
• Abstract summary: We investigate biases in pretraining datasets for large language models (LLMs) through dataset classification experiments. We analyze popular open-source pretraining datasets for LLMs derived from CommonCrawl including C4, RefinedWeb, DolmaCC, RedPajama-V2, FineWeb, and DCLM-Baseline. neural networks can classify surprisingly well which dataset a single text sequence belongs to, significantly better than a human can.
• Score: 22.53813258871828
• License:
• Abstract: We investigate biases in pretraining datasets for large language models (LLMs) through dataset classification experiments. Building on prior work demonstrating the existence of biases in popular computer vision datasets, we analyze popular open-source pretraining datasets for LLMs derived from CommonCrawl including C4, RefinedWeb, DolmaCC, RedPajama-V2, FineWeb, and DCLM-Baseline. Despite those datasets being obtained with similar filtering and deduplication steps, neural networks can classify surprisingly well which dataset a single text sequence belongs to, significantly better than a human can. This indicates that popular pretraining datasets have their own unique biases or fingerprints. Those biases remain even when the text is rewritten with LLMs. Moreover, these biases propagate through training: Random sequences generated by models trained on those datasets can be classified well by a classifier trained on the original datasets.

Related papers

• Zero-shot Meta-learning for Tabular Prediction Tasks with Adversarially Pre-trained Transformer [2.1677183904102257]
  We present an Adversarially Pre-trained Transformer (APT) that is able to perform zero-shot meta-learning on tabular prediction tasks without pre-training on any real-world dataset. APT is pre-trained with adversarial synthetic data agents, who deliberately challenge the model with different synthetic datasets. We show that our framework matches state-of-the-art performance on small classification tasks without filtering on dataset characteristics.
  arXiv  Detail & Related papers  (2025-02-06T23:58:11Z)
• Generating Realistic Tabular Data with Large Language Models [49.03536886067729]
  Large language models (LLM) have been used for diverse tasks, but do not capture the correct correlation between the features and the target variable. We propose a LLM-based method with three important improvements to correctly capture the ground-truth feature-class correlation in the real data. Our experiments show that our method significantly outperforms 10 SOTA baselines on 20 datasets in downstream tasks.
  arXiv  Detail & Related papers  (2024-10-29T04:14:32Z)
• Data Filtering Networks [67.827994353269]
  We study the problem of learning a data filtering network (DFN) for this second step of filtering a large uncurated dataset. Our key finding is that the quality of a network for filtering is distinct from its performance on downstream tasks. Based on our insights, we construct new data filtering networks that induce state-of-the-art image-text datasets.
  arXiv  Detail & Related papers  (2023-09-29T17:37:29Z)
• Large Language Model as Attributed Training Data Generator: A Tale of Diversity and Bias [92.41919689753051]
  Large language models (LLMs) have been recently leveraged as training data generators for various natural language processing (NLP) tasks. We investigate training data generation with diversely attributed prompts, which have the potential to yield diverse and attributed generated data. We show that attributed prompts outperform simple class-conditional prompts in terms of the resulting model's performance.
  arXiv  Detail & Related papers  (2023-06-28T03:31:31Z)
• Quality Not Quantity: On the Interaction between Dataset Design and Robustness of CLIP [43.7219097444333]
  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.
  arXiv  Detail & Related papers  (2022-08-10T18:24:23Z)
• Generating Data to Mitigate Spurious Correlations in Natural Language Inference Datasets [27.562256973255728]
  Natural language processing models often exploit spurious correlations between task-independent features and labels in datasets to perform well only within the distributions they are trained on. We propose to tackle this problem by generating a debiased version of a dataset, which can then be used to train a debiased, off-the-shelf model. Our approach consists of 1) a method for training data generators to generate high-quality, label-consistent data samples; and 2) a filtering mechanism for removing data points that contribute to spurious correlations.
  arXiv  Detail & Related papers  (2022-03-24T09:08:05Z)
• Learning a Self-Expressive Network for Subspace Clustering [15.096251922264281]
  We propose a novel framework for subspace clustering, termed Self-Expressive Network (SENet), which employs a properly designed neural network to learn a self-expressive representation of the data. Our SENet can not only learn the self-expressive coefficients with desired properties on the training data, but also handle out-of-sample data. In particular, SENet yields highly competitive performance on MNIST, Fashion MNIST and Extended MNIST and state-of-the-art performance on CIFAR-10.
  arXiv  Detail & Related papers  (2021-10-08T18:06:06Z)
• Dataset Bias Mitigation Through Analysis of CNN Training Scores [0.0]
  We propose a novel, domain-independent approach, called score-based resampling (SBR), to locate the under-represented samples of the original training dataset. In our method, once trained, we use the same CNN model to infer on its own training samples, obtain prediction scores, and based on the distance between predicted and ground-truth, we identify samples that are far away from their ground-truth. The obtained results confirmed the validity of our proposed method regrading identifying under-represented samples among original dataset to decrease categorical bias of classifying certain groups.
  arXiv  Detail & Related papers  (2021-06-28T16:07:49Z)
• FIND: Human-in-the-Loop Debugging Deep Text Classifiers [55.135620983922564]
  We propose FIND -- a framework which enables humans to debug deep learning text classifiers by disabling irrelevant hidden features. Experiments show that by using FIND, humans can improve CNN text classifiers which were trained under different types of imperfect datasets.
  arXiv  Detail & Related papers  (2020-10-10T12:52:53Z)
• 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.