Lifting Data-Tracing Machine Unlearning to Knowledge-Tracing for Foundation Models

2025/06/12 by Yuwen Tan, Boqing Gong, Tan, Yuwen +1
Engineering · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Mineral Processing and Grinding #Reservoir Engineering and Simulation Methods #Tunneling and Rock Mechanics

paper · pdf · doi:10.48550/arxiv.2506.11253

openalex publication_date 2025/06/12 · openalex created_date 2025/10/11 · openalex updated_date 2026/07/28

Abstract

Machine unlearning removes certain training data points and their influence from AI models (e.g., when a data owner revokes their consent to allow models to learn from the data). In this position paper, we propose to lift data-tracing machine unlearning to knowledge-tracing for foundation models (FMs). We support this position based on practical needs and insights from cognitive studies. Practically, tracing data cannot meet the diverse unlearning requests for FMs, which may be from regulators, enterprise users, product teams, etc., who have no access to FMs' massive training data. Instead, it is convenient for these parties to issue an unlearning request about the knowledge or capability FMs (should not) possess. Cognitively, knowledge-tracing unlearning aligns with how the human brain forgets more closely than tracing individual training data points does. We further discuss the nontrivial challenges in the knowledge-tracing machine unlearning paradigm. Finally, we provide a concrete case study about a vision-language FM to illustrate how an unlearner might instantiate the knowledge-tracing machine unlearning paradigm. Code is available at: https://1yuwen.github.io/Knowledge-Tracing-MU-Page.

Citations

• Benchmarking Vision Language Model Unlearning via Fictitious Facial Identity Dataset
• CLIPErase: Efficient Unlearning of Visual-Textual Associations in CLIP
• GPT-4o System Card
• Scalability of memorization-based machine unlearning
• A Closer Look at Machine Unlearning for Large Language Models
• Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning
• Position: LLM Unlearning Benchmarks are Weak Measures of Progress
• Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
• Towards Robust and Parameter-Efficient Knowledge Unlearning for LLMs
• LLaVA-OneVision: Easy Visual Task Transfer
• Reliable and Efficient Concept Erasure of Text-to-Image Diffusion Models
• Direct Unlearning Optimization for Robust and Safe Text-to-Image Models
• MUSE: Machine Unlearning Six-Way Evaluation for Language Models
• To Forget or Not? Towards Practical Knowledge Unlearning for Large Language Models
• MC-MKE: A Fine-Grained Multimodal Knowledge Editing Benchmark Emphasizing Modality Consistency
• RWKU: Benchmarking Real-World Knowledge Unlearning for Large Language Models
• Are we making progress in unlearning? Findings from the first NeurIPS unlearning competition
• What makes unlearning hard and what to do about it
• Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models
• Single Image Unlearning: Efficient Machine Unlearning in Multimodal Large Language Models
• Offset Unlearning for Large Language Models
• Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning
• VLKEB: A Large Vision-Language Model Knowledge Editing Benchmark
• The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning
• Machine Unlearning of Pre-trained Large Language Models
• MIKE: A New Benchmark for Fine-grained Multimodal Entity Knowledge Editing
• Towards Safer Large Language Models through Machine Unlearning
• Rethinking Machine Unlearning for Large Language Models
• TOFU: A Task of Fictitious Unlearning for LLMs
• MultiDelete for Multimodal Machine Unlearning
• Unlearn What You Want to Forget: Efficient Unlearning for LLMs
• DEPN: Detecting and Editing Privacy Neurons in Pretrained Language Models
• Knowledge Editing for Large Language Models: A Survey
• SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation
• Large Language Model Unlearning
• In-Context Unlearning: Language Models as Few Shot Unlearners
• Who's Harry Potter? Approximate Unlearning in LLMs
• Knowledge Sanitization of Large Language Models
• Separate the Wheat from the Chaff: Model Deficiency Unlearning via Parameter-Efficient Module Operation
• A Comprehensive Survey of Forgetting in Deep Learning Beyond Continual Learning
• A Comprehensive Survey of Forgetting in Deep Learning Beyond Continual Learning
• Self-regulating Prompts: Foundational Model Adaptation without Forgetting
• Composing Parameter-Efficient Modules with Arithmetic Operations
• MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models
• KGA: A General Machine Unlearning Framework Based on Knowledge Gap Alignment
• Visual Instruction Tuning
• Model Sparsity Can Simplify Machine Unlearning
• Boundary Unlearning
• GPT-4 Technical Report
• Erasing Concepts from Diffusion Models
• LLaMA: Open and Efficient Foundation Language Models
• Towards Unbounded Machine Unlearning
• Editing Models with Task Arithmetic
• Knowledge Unlearning for Mitigating Privacy Risks in Language Models
• Memory-Based Model Editing at Scale
• Continual Learning and Private Unlearning
• On the Necessity of Auditable Algorithmic Definitions for Machine Unlearning
• Fast Model Editing at Scale
• Unrolling SGD: Understanding Factors Influencing Machine Unlearning
• On the Opportunities and Risks of Foundation Models
• Not All Memories are Created Equal: Learning to Forget by Expiring
• Remember What You Want to Forget: Algorithms for Machine Unlearning
• Learning Transferable Visual Models From Natural Language Supervision
• Language Models are Few-Shot Learners
• Machine Unlearning
• Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep\n Networks
• Certified Data Removal from Machine Learning Models
• TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for\n Reading Comprehension
• A Large-Scale Car Dataset for Fine-Grained Categorization and Verification
• Microsoft COCO Captions: Data Collection and Evaluation Server
• The form of the forgetting curve and the fate of memories

Cited by

• Apollo: A Posteriori Label-Only Membership Inference Attack Towards Machine Unlearning

Related

• Exploring the Landscape of Machine Unlearning: A Comprehensive Survey and Taxonomy
• KGA: A General Machine Unlearning Framework Based on Knowledge Gap Alignment
• Efficient Knowledge Deletion from Trained Models through Layer-wise Partial Machine Unlearning
• How to unlearn a learned Machine Learning model ?
• Efficient Machine Unlearning via Influence Approximation