Disentangling Representations of Text by Masking Transformers

• URL: http://arxiv.org/abs/2104.07155v1
• Date: Wed, 14 Apr 2021 22:45:34 GMT
• Title: Disentangling Representations of Text by Masking Transformers
• Authors: Xiongyi Zhang, Jan-Willem van de Meent, Byron C. Wallace
• Abstract summary: We learn binary masks over transformer weights or hidden units to uncover subsets of features that correlate with a specific factor of variation. We evaluate this method with respect to its ability to disentangle representations of sentiment from genre in movie reviews, "toxicity" from dialect in Tweets, and syntax from semantics.
• Score: 27.6903196190087
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Representations from large pretrained models such as BERT encode a range of features into monolithic vectors, affording strong predictive accuracy across a multitude of downstream tasks. In this paper we explore whether it is possible to learn disentangled representations by identifying existing subnetworks within pretrained models that encode distinct, complementary aspect representations. Concretely, we learn binary masks over transformer weights or hidden units to uncover subsets of features that correlate with a specific factor of variation; this eliminates the need to train a disentangled model from scratch for a particular task. We evaluate this method with respect to its ability to disentangle representations of sentiment from genre in movie reviews, "toxicity" from dialect in Tweets, and syntax from semantics. By combining masking with magnitude pruning we find that we can identify sparse subnetworks within BERT that strongly encode particular aspects (e.g., toxicity) while only weakly encoding others (e.g., race). Moreover, despite only learning masks, we find that disentanglement-via-masking performs as well as -- and often better than -- previously proposed methods based on variational autoencoders and adversarial training.

Related papers

• MaskInversion: Localized Embeddings via Optimization of Explainability Maps [49.50785637749757]
  MaskInversion generates a context-aware embedding for a query image region specified by a mask at test time. It can be used for a broad range of tasks, including open-vocabulary class retrieval, referring expression comprehension, as well as for localized captioning and image generation.
  arXiv  Detail & Related papers  (2024-07-29T14:21:07Z)
• Open-Vocabulary Segmentation with Unpaired Mask-Text Supervision [87.15580604023555]
  Unpair-Seg is a novel weakly-supervised open-vocabulary segmentation framework. It learns from unpaired image-mask and image-text pairs, which can be independently and efficiently collected. It achieves 14.6% and 19.5% mIoU on the ADE-847 and PASCAL Context-459 datasets.
  arXiv  Detail & Related papers  (2024-02-14T06:01:44Z)
• Robust Infidelity: When Faithfulness Measures on Masked Language Models Are Misleading [5.124348720450654]
  We show that iterative masking produces large variation in faithfulness scores between otherwise comparable Transformer encoder text classifiers. We explore task-specific considerations that undermine principled comparison of interpretability using iterative masking.
  arXiv  Detail & Related papers  (2023-08-13T15:44:39Z)
• Block the Label and Noise: An N-Gram Masked Speller for Chinese Spell Checking [0.0]
  This paper proposes an n-gram masking layer that masks current and/or surrounding tokens to avoid label leakage and error disturbance. Experiments on SIGHAN datasets have demonstrated that the pluggable n-gram masking mechanism can improve the performance of prevalent CSC models.
  arXiv  Detail & Related papers  (2023-05-05T06:43:56Z)
• Word Order Matters when you Increase Masking [70.29624135819884]
  We study the effect of removing position encodings on the pre-training objective itself, to test whether models can reconstruct position information from co-occurrences alone. We find that the necessity of position information increases with the amount of masking, and that masked language models without position encodings are not able to reconstruct this information on the task.
  arXiv  Detail & Related papers  (2022-11-08T18:14:04Z)
• What You See is What You Classify: Black Box Attributions [61.998683569022006]
  We train a deep network, the Explainer, to predict attributions for a pre-trained black-box classifier, the Explanandum. Unlike most existing approaches, ours is capable of directly generating very distinct class-specific masks. We show that our attributions are superior to established methods both visually and quantitatively.
  arXiv  Detail & Related papers  (2022-05-23T12:30:04Z)
• Few-shot Semantic Image Synthesis Using StyleGAN Prior [8.528384027684192]
  We present a training strategy that performs pseudo labeling of semantic masks using the StyleGAN prior. Our key idea is to construct a simple mapping between the StyleGAN feature and each semantic class from a few examples of semantic masks. Although the pseudo semantic masks might be too coarse for previous approaches that require pixel-aligned masks, our framework can synthesize high-quality images from not only dense semantic masks but also sparse inputs such as landmarks and scribbles.
  arXiv  Detail & Related papers  (2021-03-27T11:04:22Z)
• Masking as an Efficient Alternative to Finetuning for Pretrained Language Models [49.64561153284428]
  We learn selective binary masks for pretrained weights in lieu of modifying them through finetuning. In intrinsic evaluations, we show that representations computed by masked language models encode information necessary for solving downstream tasks.
  arXiv  Detail & Related papers  (2020-04-26T15:03:47Z)
• UniLMv2: Pseudo-Masked Language Models for Unified Language Model Pre-Training [152.63467944568094]
  We propose to pre-train a unified language model for both autoencoding and partially autoregressive language modeling tasks. Our experiments show that the unified language models pre-trained using PMLM achieve new state-of-the-art results on a wide range of natural language understanding and generation tasks.
  arXiv  Detail & Related papers  (2020-02-28T15:28:49Z)

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.