| Challenge: | Feature attribution methods are used to help users interpret complex models. |
| Approach: | They propose a feature attribution method that integrates feature attributed features into the objective function to allow machine learning practitioners to incorporate priors in model building. |
| Outcome: | The proposed method reduces undesired model biases without a tradeoff on the original task and improves classifier performance in scarce data setting. |
Similar Papers
From Shortcuts to Balance: Attribution Analysis of Speech-Text Feature Utilization in Distinguishing Original from Machine-Translated Texts (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent work demonstrated strong performance in distinguishing machine-translated text from human-authored or human-transcribed content using pretrained language models. |
| Approach: | They find that bimodal integration reduces reliance on NEs while moderating overemphasis attribution patterns in speech features. |
| Outcome: | The proposed models show that they are more balanced while relying less on NEs. |
Locally Distributed Activation Vectors for Guided Feature Attribution (2022.coling-1)
Copied to clipboard
| Challenge: | Existing methods to explain predictions of deep neural networks are unstable and do not always provide faithful explanations to the target model. |
| Approach: | They propose a method to learn explanations-specific representations while constructing deep network models for text classification. |
| Outcome: | The proposed method improves model interpretability while preserving predictive performance. |
Combining Feature and Instance Attribution to Detect Artifacts (2022.findings-acl)
Copied to clipboard
| Challenge: | In this paper, we evaluate use of different attribution methods for aiding identification of training data artifacts. |
| Approach: | They propose hybrid methods that combine saliency maps and instance attribution methods to aid in identifying training data artifacts. |
| Outcome: | The proposed methods can be used to efficiently uncover artifacts in training data when a challenging validation set is available. |
Necessity and Sufficiency for Explaining Text Classifiers: A Case Study in Hate Speech Detection (2022.naacl-main)
Copied to clipboard
| Challenge: | XAI features usually provide a single importance score for each token, but feature attribution methods provide two complementary and theoretically-grounded scores for each utterance. |
| Approach: | They propose a feature attribution method that generates explicit perturbations of the input text, allowing the importance scores themselves to be explainable. |
| Outcome: | The proposed method explain the predictions of hate speech detection models on a set of curated examples from a test suite. |
Explaining Pre-Trained Language Models with Attribution Scores: An Analysis in Low-Resource Settings (2024.lrec-main)
Copied to clipboard
| Challenge: | Currently, prompt-based models are gaining popularity due to their easier adaptability in low-resource settings. |
| Approach: | They analyze attribution scores extracted from prompt-based models w.r.t. plausibility and faithfulness and compare them with attribution score extracted from fine-tuned models and large language models. |
| Outcome: | The proposed model outperforms attention and Integrated Gradients in plausibility and faithfulness, while fine-tuning models are harder to explain in low-resource settings. |
Investigating Ensemble Methods for Model Robustness Improvement of Text Classifiers (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to reduce model's reliance on bias features ignore the learnability of these features. |
| Approach: | They propose to reduce models' reliance on bias features by first training models with fixed low-capacity models which ignore the learnability of the bias features. |
| Outcome: | The proposed models can perform better on out-of-distribution datasets than baseline models with a more sophisticated model design. |
Through the Looking Glass: Learning to Attribute Synthetic Text Generated by Language Models (2021.eacl-main)
Copied to clipboard
| Challenge: | Recent advances in natural language processing have enabled synthetic text generation that is often comparable to the organic text. |
| Approach: | They propose and test several ML-based methods to attribute authorship of synthetic text to language models (LMs) they propose to use a fine-tuned version of XLNet to achieve excellent accuracy . |
| Outcome: | The proposed method achieves excellent accuracy (91% to near perfect 98%) across a range of experiments where the synthetic text may be generated using pre-trained LMs, fine-tuned LM, or by varying text generation parameters. |
Upstream Mitigation Is Not All You Need: Testing the Bias Transfer Hypothesis in Pre-Trained Language Models (2022.acl-long)
Copied to clipboard
| Challenge: | Large language models and other massively pre-trained "foundation" models can easily adapt to a wide variety of downstream tasks in a process called finetuning. |
| Approach: | They propose to use the bias transfer hypothesis to reduce social biases internalized by large language models during pre-training into harmful task-specific behavior after fine-tuning. |
| Outcome: | The bias transfer hypothesis is the theory that social biases internalized by large language models during pre-training transfer into harmful task-specific behavior after fine-tuning. |
Exploring Distantly-Labeled Rationales in Neural Network Models (2021.acl-long)
Copied to clipboard
| Challenge: | Existing methods focus on distantly-labeled rationales, ignoring the potential important non-rationale words and not distinguishing the importance of different rationale words. |
| Approach: | They propose two novel auxiliary loss functions to make better use of distantly-labeled rationales, which encourage models to maintain their focus on important words beyond labeled rationals (PINs) and alleviate redundant training on non-helpful rationale (NoIRs). |
| Outcome: | The proposed methods outperform existing methods on two representative classification tasks while maintaining the ability to spread focus to other unlabeled important words. |
Topics to Avoid: Demoting Latent Confounds in Text Classification (D19-1)
Copied to clipboard
| Challenge: | Despite impressive performance on many text classification tasks, deep neural networks tend to learn frequent superficial patterns that are specific to the training data and do not always generalize well. |
| Approach: | They propose a method that represents latent topical confounds and a model which “unlearns” confounding features by predicting both the label of the input text and the confound. |
| Outcome: | The proposed model generalizes better and learns features indicative of the writing style rather than the content. |