尽管人们已在缓解持续学习中的灾难性遗忘方面付出巨大努力，但其内在机制仍未被充分理解。在本研究中，我们证明了“伪遗忘”的存在：模型在先前任务上的性能下降并非源于能力丧失，而是由于指令未能激活相应的模型能力。我们表明，通过两种简单干预可以恢复模型在先前任务上的表现：（1）提供部分外部正确推理依据；（2）在原始指令后附加语义上无意义的后缀，以引导生成正确的推理依据。通过对支配推理依据生成的内部机制进行实证分析，我们揭示出表现出伪遗忘的模型在推理依据生成过程中对指令的依赖性降低，从而导致其固有能力未得到最优激活。基于这一洞见，我们提出了基于推理依据引导难度的回放（Rationale-Guidance Difficulty based Replay，RGD-R）框架，该框架根据模型正确利用其固有能力的能力动态分配回放数据。实验结果表明，RGD-R能够有效缓解伪遗忘，同时保持模型的可塑性。
英文摘要：
Although substantial efforts have been made to mitigate catastrophic forgetting in continual learning, the intrinsic mechanisms are not well understood. In this work, we demonstrate the existence of "pseudo forgetting": the performance degradation on previous tasks is not attributed to a loss of capabilities, but rather to the failure of the instructions to activate the appropriate model abilities. We show that the model's performance on previous tasks can be restored through two simple interventions: (1) providing partial external correct rationale, and (2) appending semantically meaningless suffixes to the original instructions, to guide the generation of correct rationales. Through empirical analysis of the internal mechanisms governing rationale generation, we reveal that models exhibiting pseudo forgetting show reduced instruction dependence during rationale generation, leading to suboptimal activation of their inherent capabilities. Based on this insight, we propose Rationale-Guidance Difficulty based Replay (RGD-R) framework that dynamically allocates replay data based on the model's ability to correctly leverage the intrinsic capabilities. Experimental results demonstrate that RGD-R effectively mitigates pseudo forgetting while maintaining model plasticity.
University College London（伦敦大学学院）；Alan Turing Institute（艾伦·图灵研究所）
Neural networks are known to suffer from catastrophic forgetting when trained on sequential datasets. While there have been numerous attempts to solve this problem in large-scale supervised classification, little has been done to overcome catastrophic forgetting in few-shot classification problems. We demonstrate that the popular gradient-based model-agnostic meta-learning algorithm (MAML) indeed suffers from catastrophic forgetting and introduce a Bayesian online meta-learning framework that tackles this problem. Our framework utilises Bayesian online learning and meta-learning along with Lap…
North Carolina State University（北卡罗来纳州立大学）；Salesforce Research（赛富时研究院）
Addressing catastrophic forgetting is one of the key challenges in continual learning where machine learning systems are trained with sequential or streaming tasks. Despite recent remarkable progress in state-of-the-art deep learning, deep neural networks (DNNs) are still plagued with the catastrophic forgetting problem. This paper presents a conceptually simple yet general and effective framework for handling catastrophic forgetting in continual learning with DNNs. The proposed method consists of two components: a neural structure optimization component and a parameter learning and/or fine-tu…
Unveiling Factual Recall Behaviors of Large Language Models through Knowledge Neurons
Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所）；University of Chinese Academy of Sciences（中国科学院大学）；Beijing Wenge Technology Co., Ltd（北京文格科技有限公司）
Unveiling the Magic of Code Reasoning through Hypothesis Decomposition and Amendment
University of Science and Technology of China（中国科学技术大学）；Institute of Artificial Intelligence, Hefei Comprehensive National Science Center（合肥综合性国家科学中心人工智能研究院）
Enhancing Continual Learning for Software Vulnerability Prediction: Addressing Catastrophic Forgetting via Hybrid-Confidence-Aware Selective Replay for Temporal LLM Fine-Tuning
School of Computer Science, University of Nottingham, Nottingham, United Kingdom（诺丁汉大学计算机科学学院）；School of Informatics and Computing, Northern Arizona University, Flagstaff, United States of America（北亚利桑那大学信息学院）；School of Computer Science, University of Nottingham, Ningbo, China（诺丁汉大学宁波校区）
Eindhoven University of Technology（埃因霍温理工大学）；University of Twente（特温特大学）
Using task-specific components within a neural network in continual learning (CL) is a compelling strategy to address the stability-plasticity dilemma in fixed-capacity models without access to past data. Current methods focus only on selecting a sub-network for a new task that reduces forgetting of past tasks. However, this selection could limit the forward transfer of relevant past knowledge that helps in future learning. Our study reveals that satisfying both objectives jointly is more challenging when a unified classifier is used for all classes of seen tasks-class-Incremental Learning (cl…
Unveiling Linguistic Regions in Large Language Models
School of Computer Science, Fudan University（复旦大学计算机科学技术学院）；Institute of Modern Languages and Linguistics, Fudan University（复旦大学现代语言与语言学研究所）；Shanghai Collaborative Innovation Center of Intelligent Visual Computing（上海智能视觉计算协同创新中心）
RAcQUEt: Unveiling the Dangers of Overlooked Referential Ambiguity in Visual LLMs
Amsterdam UMC, Department of Medical Informatics（阿姆斯特丹大学医学中心，医学信息学系）；Center for Information and Language Processing, LMU Munich（信息与语言处理中心，慕尼黑大学）；Munich Center for Machine Learning (MCML), Munich（慕尼黑机器学习中心（MCML））；Institute for Logic, Language and Computation (ILLC), University of Amsterdam（逻辑、语言与计算研究所（ILLC），阿姆斯特丹大学）
Humans learn all their life long. They accumulate knowledge from a sequence of learning experiences and remember the essential concepts without forgetting what they have learned previously. Artificial neural networks struggle to learn similarly. They often rely on data rigorously preprocessed to learn solutions to specific problems such as classification or regression. In particular, they forget their past learning experiences if trained on new ones. Therefore, artificial neural networks are often inept to deal with real-life settings such as an autonomous-robot that has to learn on-line to ad…
Arizona State University（亚利桑那州立大学）；University of California, Davis（加利福尼亚大学戴维斯分校）
By learning a sequence of tasks continually, an agent in continual learning (CL) can improve the learning performance of both a new task and `old' tasks by leveraging the forward knowledge transfer and the backward knowledge transfer, respectively. However, most existing CL methods focus on addressing catastrophic forgetting in neural networks by minimizing the modification of the learnt model for old tasks. This inevitably limits the backward knowledge transfer from the new task to the old tasks, because judicious model updates could possibly improve the learning performance of the old tasks …
Better Aligned with Survey Respondents or Training Data? Unveiling Political Leanings of LLMs on U.S. Supreme Court Cases
Technical University of Munich（慕尼黑技术大学）；Allen Institute for AI（艾伦人工智能研究所）；University of Washington（华盛顿大学）；LMU Munich & Munich Center for Machine Learning (MCML)（慕尼黑大学及慕尼黑机器学习中心）
CEL: A Continual Learning Model for Disease Outbreak Prediction by Leveraging Domain Adaptation via Elastic Weight Consolidation
Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences（中国科学院深圳先进技术研究院）；University of Chinese Academy of Sciences（中国科学院大学）；University of Macau（澳门大学）
Find Them All: Unveiling MLLMs for Versatile Person Re-identification
School of Computer Science and Technology, East China Normal University（东华大学计算机科学与技术学院）；Institute of Image Communication and Information Processing, Shanghai Jiao Tong University（上海交通大学图像通信与信息处理研究所）；Macao Polytechnic University（澳门 polytechnic 大学）；Shanghai AI Laboratory（上海人工智能实验室）
University of Cambridge（剑桥大学）；Carnegie Mellon University（卡内基梅隆大学）；Microsoft Research（微软研究院）
Approaches to continual learning aim to successfully learn a set of related tasks that arrive in an online manner. Recently, several frameworks have been developed which enable deep learning to be deployed in this learning scenario. A key modelling decision is to what extent the architecture should be shared across tasks. On the one hand, separately modelling each task avoids catastrophic forgetting but it does not support transfer learning and leads to large models. On the other hand, rigidly specifying a shared component and a task-specific part enables task transfer and limits the model siz…
Technische Universität München（慕尼黑工业大学）；TUM（慕尼黑工业大学）；Siemens AG（西门子股份公司）；LMU（慕尼黑大学）
Topic models such as LDA, DocNADE, iDocNADEe have been popular in document analysis. However, the traditional topic models have several limitations including: (1) Bag-of-words (BoW) assumption, where they ignore word ordering, (2) Data sparsity, where the application of topic models is challenging due to limited word co-occurrences, leading to incoherent topics and (3) No Continuous Learning framework for topic learning in lifelong fashion, exploiting historical knowledge (or latent topics) and minimizing catastrophic forgetting. This thesis focuses on addressing the above challenges within ne…
Learning to Modulate Random Weights: Neuromodulation-inspired Neural Networks For Efficient Continual Learning
School of Computing and Augmented Intelligence（计算与增强智能学院）；School of Sustainability（可持续发展学院）；School of Complex Adaptive Systems（复杂自适应系统学院）；School of Life Sciences（生命科学学院）；Arizona State University（亚利桑那州立大学）
Mind Scramble: Unveiling Large Language Model Psychology Via Typoglycemia
University of Science and Technology of China (USTC)（中国科学技术大学）；Squirrel AI（松鼠AI）；National University of Singapore (NUS)（新加坡国立大学）；NanYang Technological University (NTU)（南洋理工大学）；The Hong Kong University of Science and Technology (HKUST)（香港科技大学）
Athinoula A. Martinos Center for Biomedical Imaging（阿西努拉·A·马蒂诺斯生物医学影像中心）；Massachusetts Institute of Technology（麻省理工学院）；Stanford University（斯坦福大学）
Model brittleness is a primary concern when deploying deep learning models in medical settings owing to inter-institution variations, like patient demographics and intra-institution variation, such as multiple scanner types. While simply training on the combined datasets is fraught with data privacy limitations, fine-tuning the model on subsequent institutions after training it on the original institution results in a decrease in performance on the original dataset, a phenomenon called catastrophic forgetting. In this paper, we investigate trade-off between model refinement and retention of pr…
DOTA: Distributional Test-Time Adaptation of Vision-Language Models
State Key Laboratory of Networking and Switching Technology（网络与交换技术国家重点实验室）；Beijing University of Posts and Telecommunications（北京邮电大学）；College of Intelligence and Computing（智能与计算学院）；Tianjin University（天津大学）；School of Computer Science and Technology（计算机科学与技术学院）；Harbin Institute of Technology Shenzhen（哈尔滨工业大学深圳学院）；Institute for Infocomm Research, A*STAR（信息通信研究机构，A*STAR）；Show Lab, National University of Singapore（新加坡国立大学Show实验室）；National University of Singapore（新加坡国立大学）
Guided Prompt Evolution for Vision-Language Models Adaptation
Tsinghua Shenzhen International Graduate School, Tsinghua University（清华大学深圳国际研究生院，清华大学）；Sun Yat-sen University（中山大学）；Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳））
