Unmasking and Improving Data Credibility: A Study with Datasets for Training Harmless Language Models

• URL: http://arxiv.org/abs/2311.11202v2
• Date: Sun, 24 Mar 2024 22:02:47 GMT
• Title: Unmasking and Improving Data Credibility: A Study with Datasets for Training Harmless Language Models
• Authors: Zhaowei Zhu, Jialu Wang, Hao Cheng, Yang Liu,
• Abstract summary: This study focuses on the credibility of real-world datasets, including the popular benchmarks Jigsaw Civil Comments, Anthropic Harmless & Red Team, PKU BeaverTails & SafeRLHF. Given the cost and difficulty of cleaning these datasets by humans, we introduce a systematic framework for evaluating the credibility of datasets. We find and fix an average of 6.16% label errors in 11 datasets constructed from the above benchmarks.
• Score: 25.893228797735908
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: Language models have shown promise in various tasks but can be affected by undesired data during training, fine-tuning, or alignment. For example, if some unsafe conversations are wrongly annotated as safe ones, the model fine-tuned on these samples may be harmful. Therefore, the correctness of annotations, i.e., the credibility of the dataset, is important. This study focuses on the credibility of real-world datasets, including the popular benchmarks Jigsaw Civil Comments, Anthropic Harmless & Red Team, PKU BeaverTails & SafeRLHF, that can be used for training a harmless language model. Given the cost and difficulty of cleaning these datasets by humans, we introduce a systematic framework for evaluating the credibility of datasets, identifying label errors, and evaluating the influence of noisy labels in the curated language data, specifically focusing on unsafe comments and conversation classification. With the framework, we find and fix an average of 6.16% label errors in 11 datasets constructed from the above benchmarks. The data credibility and downstream learning performance can be remarkably improved by directly fixing label errors, indicating the significance of cleaning existing real-world datasets. We provide an open-source tool, Docta, for data cleaning at https://github.com/Docta-ai/docta.

Related papers

• Unveiling Imitation Learning: Exploring the Impact of Data Falsity to Large Language Model [6.097530398802087]
  This paper explores the correlation between the degree of noise and its impact on language models through instruction tuning. Specifically, we found multiple intriguing findings of the correlation between the factuality of the dataset and instruction tuning.
  arXiv  Detail & Related papers  (2024-04-15T12:20:09Z)
• Maximizing Data Efficiency for Cross-Lingual TTS Adaptation by Self-Supervised Representation Mixing and Embedding Initialization [57.38123229553157]
  This paper presents an effective transfer learning framework for language adaptation in text-to-speech systems. We focus on achieving language adaptation using minimal labeled and unlabeled data. Experimental results show that our framework is able to synthesize intelligible speech in unseen languages with only 4 utterances of labeled data and 15 minutes of unlabeled data.
  arXiv  Detail & Related papers  (2024-01-23T21:55:34Z)
• Soft Curriculum for Learning Conditional GANs with Noisy-Labeled and Uncurated Unlabeled Data [70.25049762295193]
  We introduce a novel conditional image generation framework that accepts noisy-labeled and uncurated data during training. We propose soft curriculum learning, which assigns instance-wise weights for adversarial training while assigning new labels for unlabeled data. Our experiments show that our approach outperforms existing semi-supervised and label-noise robust methods in terms of both quantitative and qualitative performance.
  arXiv  Detail & Related papers  (2023-07-17T08:31:59Z)
• Bring Your Own Data! Self-Supervised Evaluation for Large Language Models [52.15056231665816]
  We propose a framework for self-supervised evaluation of Large Language Models (LLMs) We demonstrate self-supervised evaluation strategies for measuring closed-book knowledge, toxicity, and long-range context dependence. We find strong correlations between self-supervised and human-supervised evaluations.
  arXiv  Detail & Related papers  (2023-06-23T17:59:09Z)
• Are labels informative in semi-supervised learning? -- Estimating and leveraging the missing-data mechanism [4.675583319625962]
  Semi-supervised learning is a powerful technique for leveraging unlabeled data to improve machine learning models. It can be affected by the presence of informative'' labels, which occur when some classes are more likely to be labeled than others. We propose a novel approach to address this issue by estimating the missing-data mechanism and using inverse propensity weighting to debias any SSL algorithm.
  arXiv  Detail & Related papers  (2023-02-15T09:18:46Z)
• Annotation Error Detection: Analyzing the Past and Present for a More Coherent Future [63.99570204416711]
  We reimplement 18 methods for detecting potential annotation errors and evaluate them on 9 English datasets. We define a uniform evaluation setup including a new formalization of the annotation error detection task. We release our datasets and implementations in an easy-to-use and open source software package.
  arXiv  Detail & Related papers  (2022-06-05T22:31:45Z)
• Towards A Reliable Ground-Truth For Biased Language Detection [3.2202224129197745]
  Existing methods to detect bias mostly rely on annotated data to train machine learning models. We evaluate data collection options and compare labels obtained from two popular crowdsourcing platforms. We conclude that detailed annotator training increases data quality, improving the performance of existing bias detection systems.
  arXiv  Detail & Related papers  (2021-12-14T14:13:05Z)
• ORDisCo: Effective and Efficient Usage of Incremental Unlabeled Data for Semi-supervised Continual Learning [52.831894583501395]
  Continual learning assumes the incoming data are fully labeled, which might not be applicable in real applications. We propose deep Online Replay with Discriminator Consistency (ORDisCo) to interdependently learn a classifier with a conditional generative adversarial network (GAN) We show ORDisCo achieves significant performance improvement on various semi-supervised learning benchmark datasets for SSCL.
  arXiv  Detail & Related papers  (2021-01-02T09:04:14Z)
• Data Cleansing with Contrastive Learning for Vocal Note Event Annotations [1.859931123372708]
  We propose a novel data cleansing model for time-varying, structured labels. Our model is trained in a contrastive learning manner by automatically creating local deformations of likely correct labels. We demonstrate that the accuracy of a transcription model improves greatly when trained using our proposed strategy.
  arXiv  Detail & Related papers  (2020-08-05T12:24:37Z)
• Learning with Out-of-Distribution Data for Audio Classification [60.48251022280506]
  We show that detecting and relabelling certain OOD instances, rather than discarding them, can have a positive effect on learning. The proposed method is shown to improve the performance of convolutional neural networks by a significant margin.
  arXiv  Detail & Related papers  (2020-02-11T21:08:06Z)

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.