Eraser: Jailbreaking Defense in Large Language Models via Unlearning Harmful Knowledge

Jailbreaking attacks can enable Large Language Models (LLMs) to bypass the safeguard and generate harmful content. Existing jailbreaking defense methods have failed to address the fundamental issue that harmful knowledge resides within the model, leading to potential jailbreak risks for LLMs. In this paper, we propose a novel defense method called Eraser, which mainly includes three goals: unlearning harmful knowledge, retaining general knowledge, and maintaining safety alignment. The intuition is that if an LLM forgets the specific knowledge required to answer a harmful question, it will no longer have the ability to answer harmful questions. The training of Erase does not actually require the model's own harmful knowledge, and it can benefit from unlearning general answers related to harmful queries, which means it does not need assistance from the red team. The experimental results show that Eraser can significantly reduce the jailbreaking success rate for various attacks without compromising the general capabilities of the model.

On the use of Silver Standard Data for Zero-shot Classification Tasks in Information Extraction

The superior performance of supervised classification methods in the information extraction (IE) area heavily relies on a large amount of gold standard data. Recent zero-shot classification methods converted the task to other NLP tasks (e.g., textual entailment) and used off-the-shelf models of these NLP tasks to directly perform inference on the test data without using a large amount of IE annotation data. A potentially valuable by-product of these methods is the large-scale silver standard data, i.e., pseudo-labeled data by the off-the-shelf models of other NLP tasks. However, there is no further investigation into the use of these data. In this paper, we propose a new framework, Clean-LaVe, which aims to utilize silver standard data to enhance the zero-shot performance. Clean-LaVe includes four phases: (1) Obtaining silver data; (2) Identifying relatively clean data from silver data; (3) Finetuning the off-the-shelf model using clean data; (4) Inference on the test data. The experimental results show that Clean-LaVe can outperform the baseline by 5% and 6% on TACRED and Wiki80 dataset in the zero-shot relation classification task, and by 3%-7% on Smile (Korean and Polish) in the zero-shot cross-lingual relation classification task, and by 8% on ACE05-E+ in the zero-shot event argument classification task. The code is share in https://github.com/wjw136/Clean_LaVe.git.

Learning with Silver Standard Data for Zero-shot Relation Extraction

The superior performance of supervised relation extraction (RE) methods heavily relies on a large amount of gold standard data. Recent zero-shot relation extraction methods converted the RE task to other NLP tasks and used off-the-shelf models of these NLP tasks to directly perform inference on the test data without using a large amount of RE annotation data. A potentially valuable by-product of these methods is the large-scale silver standard data. However, there is no further investigation on the use of potentially valuable silver standard data. In this paper, we propose to first detect a small amount of clean data from silver standard data and then use the selected clean data to finetune the pretrained model. We then use the finetuned model to infer relation types. We also propose a class-aware clean data detection module to consider class information when selecting clean data. The experimental results show that our method can outperform the baseline by 12% and 11% on TACRED and Wiki80 dataset in the zero-shot RE task. By using extra silver standard data of different distributions, the performance can be further improved.

Joint Learning for Pulmonary Nodule Segmentation, Attributes and Malignancy Prediction

Refer to the literature of lung nodule classification, many studies adopt Convolutional Neural Networks (CNN) to directly predict the malignancy of lung nodules with original thoracic Computed Tomography (CT) and nodule location. However, these studies cannot tell how the CNN works in terms of predicting the malignancy of the given nodule, e.g., it's hard to conclude that whether the region within the nodule or the contextual information matters according to the output of the CNN. In this paper, we propose an interpretable and multi-task learning CNN -- Joint learning for \textbf{P}ulmonary \textbf{N}odule \textbf{S}egmentation \textbf{A}ttributes and \textbf{M}alignancy \textbf{P}rediction (PN-SAMP). It is able to not only accurately predict the malignancy of lung nodules, but also provide semantic high-level attributes as well as the areas of detected nodules. Moreover, the combination of nodule segmentation, attributes and malignancy prediction is helpful to improve the performance of each single task. In addition, inspired by the fact that radiologists often change window widths and window centers to help to make decision on uncertain nodules, PN-SAMP mixes multiple WW/WC together to gain information for the raw CT input images. To verify the effectiveness of the proposed method, the evaluation is implemented on the public LIDC-IDRI dataset, which is one of the largest dataset for lung nodule malignancy prediction. Experiments indicate that the proposed PN-SAMP achieves significant improvement with respect to lung nodule classification, and promising performance on lung nodule segmentation and attribute learning, compared with the-state-of-the-art methods.