Minh-Phuong Nguyen

Dang Huu-Tien, Minh-Phuong Nguyen, Naoya Inoue

Label noise in datasets could significantly damage the performance and robustness of deep neural networks (DNNs) trained on these datasets. As the size of modern DNNs grows, there is a growing demand for automated tools for detecting such errors. In this paper, we propose post-hoc, model-agnostic noise detection and rectification methods utilizing the penultimate feature from a DNN. Our idea is based on the observation that the similarity between the penultimate feature of a mislabeled data point and its true class data points is higher than that for data points from other classes, making the probability of label occurrence within a tight, similar cluster informative for detecting and rectifying errors. Through theoretical and empirical analyses, we demonstrate that our approach achieves high detection performance across diverse, realistic noise scenarios and can automatically rectify these errors to improve dataset quality. Our implementation is available at https://anonymous.4open.science/r/noise-detection-and-rectification-AD8E.

Dang Huu-Tien, Hoang Thanh-Tung, Anh Bui et al.

Here, we show that current state-of-the-art LLM unlearning methods inherently reduce models' robustness, causing them to misbehave even when a single non-adversarial forget-token is present in the retain-query. Toward understanding underlying causes, we propose a novel theoretical framework that reframes the unlearning process as backdoor attacks and defenses: forget-tokens act as backdoor triggers that, when activated in retain-queries, cause disruptions in unlearned models' behaviors, similar to successful backdoor attacks. The sense that, LLM unlearning methods themselves poison the model, make it more vulnerable to forget-tokens, and hide rather than erase target knowledge, describes their true mechanism. To mitigate the vulnerability caused by the forgetting process, we reinterpret the retaining process as a backdoor defense and propose Random Noise Augmentation (RNA), a lightweight, model and method-agnostic approach with theoretical guarantees for improving the robustness of models. Extensive experiments demonstrate that RNA significantly improves the robustness of unlearned models while preserving forget and retain performances. This backdoor attack-defense framework offers insights into the mechanism of unlearning that can shed light on future research directions for improving unlearning robustness.

Ha-Thanh Nguyen, Minh-Phuong Nguyen, Thi-Hai-Yen Vuong et al.

In this paper, we introduce our approaches using Transformer-based models for different problems of the COLIEE 2021 automatic legal text processing competition. Automated processing of legal documents is a challenging task because of the characteristics of legal documents as well as the limitation of the amount of data. With our detailed experiments, we found that Transformer-based pretrained language models can perform well with automated legal text processing problems with appropriate approaches. We describe in detail the processing steps for each task such as problem formulation, data processing and augmentation, pretraining, finetuning. In addition, we introduce to the community two pretrained models that take advantage of parallel translations in legal domain, NFSP and NMSP. In which, NFSP achieves the state-of-the-art result in Task 5 of the competition. Although the paper focuses on technical reporting, the novelty of its approaches can also be an useful reference in automated legal document processing using Transformer-based models.

Ha-Thanh Nguyen, Vu Tran, Tran-Binh Dang et al.

Attention is all we need as long as we have enough data. Even so, it is sometimes not easy to determine how much data is enough while the models are becoming larger and larger. In this paper, we propose HYDRA heads, lightweight pretrained linguistic self-attention heads to inject knowledge into transformer models without pretraining them again. Our approach is a balanced paradigm between leaving the models to learn unsupervised and forcing them to conform to linguistic knowledge rigidly as suggested in previous studies. Our experiment proves that the approach is not only the boost performance of the model but also lightweight and architecture friendly. We empirically verify our framework on benchmark datasets to show the contribution of linguistic knowledge to a transformer model. This is a promising result for a new approach to transferring knowledge from linguistic resources into transformer-based models.