Papers by Pengwei Zhan
Mitigating the Inconsistency Between Word Saliency and Model Confidence with Pathological Contrastive Training (2022.findings-acl)
Copied to clipboard
| Challenge: | Neural networks are used for various NLP tasks, but their complexity makes them difficult to interpret. |
| Approach: | They propose a framework to mitigate the model pathology and obtain more interpretable models by using contrastive learning and saliency-based samples augmentation to calibrate the sentences representation. |
| Outcome: | The proposed framework can mitigate the model pathology and generate more interpretable models while keeping the model performance. |
Multimodal Fusion with Co-Attention Networks for Fake News Detection (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to detect fake news with textual and visual contents are ineffective because they concatenate unimodal features without considering inter-modality relations. |
| Approach: | They propose to fuse textual and visual features for fake news detection using multimodal co-attention networks to learn inter-dependencies between multimodal features. |
| Outcome: | Extensive experiments on two realworld datasets show that the proposed network outperforms state-of-the-art methods and learns inter-dependencies among multimodal features. |
Similarizing the Influence of Words with Contrastive Learning to Defend Word-level Adversarial Text Attack (2023.findings-acl)
Copied to clipboard
| Challenge: | Neural language models are vulnerable to word-level adversarial text attacks . previous word-based search methods assume important words influence prediction . |
| Approach: | They propose a method for similarizing the influence of words with contrast learning that encourages model to learn sentence representations in which words of varying importance have a more uniform influence on prediction. |
| Outcome: | The proposed method is compatible with various training methods and improves model robustness against various adversarial attacks. |
Unveiling the Lexical Sensitivity of LLMs: Combinatorial Optimization for Prompt Enhancement (2024.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) demonstrate exceptional instruct-following ability to complete downstream tasks. |
| Approach: | They propose a black-box combinatorial optimization framework that iteratively improves lexical choices in prompts by a search strategy related to word influence. |
| Outcome: | The proposed framework recovers the model's ability to instruct-follow and solve downstream tasks even when the variations are imperceptible to humans. |
PARSE: An Efficient Search Method for Black-box Adversarial Text Attacks (2022.coling-1)
Copied to clipboard
| Challenge: | Neural networks are vulnerable to adversarial examples, i.e., under a black-box scenario. |
| Approach: | They propose a word-level search algorithm that searches for subareas under dynamic search space following the subarea importance. |
| Outcome: | The proposed algorithm can achieve comparable success rates to complex search methods while saving numerous queries and time. |
Rethinking Word-level Adversarial Attack: The Trade-off between Efficiency, Effectiveness, and Imperceptibility (2024.lrec-main)
Copied to clipboard
| Challenge: | Neural language models have demonstrated impressive performance but remain vulnerable to word-level adversarial attacks. |
| Approach: | They propose two standardized search spaces to address the problem of word-level adversarial attacks. |
| Outcome: | The proposed search spaces improve performance and trade-offs in different scenarios. |
Contrastive Learning with Adversarial Examples for Alleviating Pathology of Language Model (2023.acl-long)
Copied to clipboard
| Challenge: | Existing interpretation methods fail to obtain faithful attributions on these models, thereby failing to reveal potential flaws and biases. |
| Approach: | They propose a Contrastive learning regularization method which calibrates the sentence representation of out-of-distribution examples and utilizes adversarial examples to introduce direction information in regularization. |
| Outcome: | The proposed method alleviates the model pathology while impacting generalization ability on in-distribution examples and thus helps interpretation methods obtain more faithful results. |