Papers by Thang Pham
Double Trouble: How to not Explain a Text Classifier’s Decisions Using Counterfactuals Synthesized by Masked Language Models? (2022.aacl-main)
Copied to clipboard
| Challenge: | Input Marginalization (IM) is a method that takes the prediction difference between before-and-after an input feature (here, a token) is removed as its attribution. |
| Approach: | They propose to use a BERT-based method to replace a token with a feature to give more plausible counterfactuals. |
| Outcome: | The proposed method is effective, but the Deletion-BERT metric is biased towards IM, and the results are not convincing. |
PiC: A Phrase-in-Context Dataset for Phrase Understanding and Semantic Search (2023.eacl-main)
Copied to clipboard
| Challenge: | Existing benchmarks for phrase-similarity compare phrases alone (without context) and phrases with context (with or without context). |
| Approach: | They propose to use a dataset of 28K noun phrases accompanied by their contextual Wikipedia pages to train machine phrase embeddings. |
| Outcome: | The proposed dataset improves ranking-models’ accuracy and pushes span selection models near human accuracy, which is 95% Exact Match (EM) on semantic search given a query phrase and a passage. |
PEEB: Part-based Image Classifiers with an Explainable and Editable Language Bottleneck (2024.findings-naacl)
Copied to clipboard
| Challenge: | CLIP-based classifiers rely on the prompt containing a class name that is known to the text encoder and perform poorly on new classes or the classes whose names rarely appear on the Internet. |
| Approach: | They propose to use a set of text descriptors to express a class name into a textual descriptable and match the embeddings of the detected parts to their textual ones to compute a logit score. |
| Outcome: | The proposed classifier outperforms CLIP-based classifiers on zero-shot and supervised learning settings by 88.80% and 92.20% accuracy on CUB-200 and Stanford Dogs-120. |
Out of Order: How important is the sequential order of words in a sentence in Natural Language Understanding tasks? (2021.findings-acl)
Copied to clipboard
| Challenge: | In July 2019, RoBERTa was the first to surpass a human baseline on GLUE . since then, 13 more methods have outperformed humans on the GLu leaderboard . |
| Approach: | They found that 75% to 90% of correct predictions of BERT-based classifiers remain constant after input words are randomly shuffled. |
| Outcome: | The proposed model outperforms humans on GLUE and SQuAD 2.0. |