Less Is Better: Recovering Intended-Feature Subspace to Robustify NLU Models (2022.coling-1)
Copied to clipboard
| Challenge: | Existing approaches to debiase datasets rely on knowledge of bias attributes . current approaches focus on how to leverage kinds of supervision effectively . |
| Approach: | They propose to extend the supervision on bias by extending it into feature space. |
| Outcome: | Empirical results show that a low-dimensional subspace with intended features can represent biased datasets. |
Similar Papers
Towards Debiasing NLU Models from Unknown Biases (2020.emnlp-main)
Copied to clipboard
| Challenge: | Recent proposed debiasing methods rely on the assumption that the types of bias should be known a-priori, which limits their application to many NLU tasks and datasets. |
| Approach: | They propose a framework that prevents models from mainly utilizing biases without knowing them in advance. |
| Outcome: | The proposed framework allows existing methods to retain performance improvement on challenge datasets without specifically targeting biases. |
Mind the Trade-off: Debiasing NLU Models without Degrading the In-distribution Performance (2020.acl-main)
Copied to clipboard
| Challenge: | Recent studies show that pre-trained language models rely heavily on idiosyncratic biases of datasets. |
| Approach: | They propose a method which discourages models from exploiting biases while enabling them to receive enough incentive to learn from all the training examples. |
| Outcome: | The proposed method improves on out-of-distribution datasets while maintaining original in-district accuracy. |
End-to-End Bias Mitigation by Modelling Biases in Corpora (2020.acl-main)
Copied to clipboard
| Challenge: | Recent studies have shown that strong natural language understanding models are prone to relying on unwanted dataset biases without learning the underlying task. |
| Approach: | They propose two learning strategies to train neural models that are more robust to dataset biases and transfer better to out-of-domain datasets. |
| Outcome: | The proposed methods improve robustness in all settings and transfer better to out-of-domain datasets. |
When and Why Does Bias Mitigation Work? (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Neural models exploit shallow surface features to perform language understanding tasks, rather than learning the deeper language understanding and reasoning skills that practitioners desire. |
| Approach: | They propose to use model debiasing techniques to pressure models away from spurious features and to use them to learn useful representations instead. |
| Outcome: | The proposed methods increase models' reliance on hidden biases instead of learning robust features that help them solve a task. |
IBADR: an Iterative Bias-Aware Dataset Refinement Framework for Debiasing NLU models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Using manual data analysis, dataset refinement approaches are often unable to cover all the potential biased features. |
| Approach: | They propose an iterative bias-aware dataset refinement framework which debiases NLU models without predefining biased features. |
| Outcome: | The proposed framework outperforms existing methods and is compatible with model-centric methods. |
Subspace Defense: Discarding Adversarial Perturbations by Learning a Subspace for Clean Signals (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing models that extract discrete inputs into fixed-length representations are vulnerable to adversarial attacks that place perturbations on clean inputs to fool DNNs. |
| Approach: | They propose to inspect the subspaces of sample features through spectral analysis to better understand adversarial attacks. |
| Outcome: | The proposed strategy enables the model to inherently suppress adversaries, which boosts model robustness and motivates new directions of effective adversarial defense. |
Modular and On-demand Bias Mitigation with Attribute-Removal Subnetworks (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing studies show that pre-trained language models can be used to mitigate societal biases and stereotypes. |
| Approach: | They propose a modular bias mitigation approach that integrates debiasing modules into the core model on-demand at inference time. |
| Outcome: | The proposed approach improves on-par with baseline finetuning on gender, race, and age protected attributes on three classification tasks with gender, age, and race as protected attributes. |
Model Unlearning via Sparse Autoencoder Subspace Guided Projections (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing unlearning strategies lack interpretability or fail to provide robust defense against adversarial prompts. |
| Approach: | They propose a framework that leverages SAE features to drive targeted updates in the model’s parameter space. |
| Outcome: | The proposed framework reduces harmful knowledge accuracy by 3.22% compared to baselines and improves adversarial robustness under jailbreak prompts. |
Improving the robustness of NLI models with minimax training (2023.acl-long)
Copied to clipboard
| Challenge: | Experimental results show that our method consistently outperforms other robustness enhancement techniques on out-of-distribution adversarial test sets, while maintaining high in-distance accuracy. |
| Approach: | They propose a minimax objective between a learner model being trained for the task and an auxiliary model aiming to maximize the learner's loss by up-weighting underrepresented "hard" examples with patterns that contradict the shortcuts learned from the prevailing "easy" examples. |
| Outcome: | The proposed method outperforms other robustness enhancement techniques on out-of-distribution adversarial test sets while maintaining high in-distance accuracy. |
OSCaR: Orthogonal Subspace Correction and Rectification of Biases in Word Embeddings (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to mitigate stereotypical biases by linear projection are too aggressive . existing methods remove bias, but they also erase valuable information from word embeddings . |
| Approach: | They propose a bias-mitigating method that disentangles biased associations between concepts instead of removing concepts wholesale. |
| Outcome: | The proposed method disentangles biased associations between concepts rather than eliminating concepts wholesale. |