Papers by Tegawendé Bissyandé
Evaluating Parameter-Efficient Finetuning Approaches for Pre-trained Models on the Financial Domain (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Large-scale language models with millions, billions, or trillions of trainable parameters are becoming increasingly popular. |
| Approach: | They compare performance of financial BERT-like models to their fully fine-tuned counterparts by using parameter-efficient tuning methods. |
| Outcome: | The proposed approaches match full fine-tuning performance on common NLP tasks, but are less studied in finance. |
Revisiting Code Similarity Evaluation with Abstract Syntax Tree Edit Distance (2024.acl-short)
Copied to clipboard
| Challenge: | Abstract Syntax Tree (AST) editing distance is a new evaluation metric for code generation tasks. |
| Approach: | They propose, optimize, and publish an enhanced version of Tree Similarity of Edit Distance (TSED) based on AST editing distance and prompt-based GPT similarity scores. |
| Outcome: | The proposed metric is an enhanced version of Tree Similarity of Edit Distance (TSED) it is compared to BLEU score, execution match, and Jaccard similarity across languages. |
CodeAgent: Autonomous Communicative Agents for Code Review (2024.emnlp-main)
Copied to clipboard
Xunzhu Tang, Kisub Kim, Yewei Song, Cedric Lothritz, Bei Li, Saad Ezzini, Haoye Tian, Jacques Klein, Tegawendé Bissyandé
| Challenge: | Existing methods for code review rely on single input-output generative models and thus lack the collaborative nature of code review. |
| Approach: | They propose a multi-agent Large Language Model (LLM) system for code review automation that incorporates a supervisory agent to ensure that all the agents’ contributions address the initial review question. |
| Outcome: | The proposed system detects inconsistencies between code changes and commit messages, identify vulnerabilities, validates code style adherence, and suggests code revisions. |