Forgetting before Learning: Utilizing Parametric Arithmetic for Knowledge Updating in Large Language Models
Open accessAbstract
Recent advancements in Large Language Models (LLMs) have showcased their remarkable capabilities in text understanding and generation.However, even stronger LLMs are susceptible to acquiring erroneous or obsolete information from the training corpus.Direct secondary fine-tuning with data containing new knowledge may be ineffective in updating knowledge due to the conflict between old and new knowledge.In this paper, we propose a new paradigm for fine-tuning called F-Learning (Forgetting before Learning), which employs parametric arithmetic to facilitate the forgetting of old knowledge and learning of new knowledge.Experimental results on two publicly available datasets demonstrate that our proposed F-Learning can obviously improve the knowledge updating performance of both full fine-tuning and LoRA fine-tuning, simultaneously outperforming the existing baselines in most cases.Moreover, we have also discovered that forgetting old knowledge by subtracting the parameters of LoRA can yield a similar effect to subtracting the parameters of full fine-tuning, and occasionally even surpass it significantly.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The 'forgetting before learning' paradigm of knowledge renewal is effective for fine-tuning with data containing new knowledge.
“Referring to [24], this paper embraces a paradigm of knowledge renewal, namely, forgetting before learning, and implements it in defect detection.”
From Cloud-Edge Collaborative Defect Detection Based on Efficient Yolo Networks and Incremental Learning (2024) · cites this paper as [24]
Cites
Still being checked.
Cited by
Sorted by citations
Model Merging in LLMs, MLLMs, and Beyond: Methods, Theories, Applications, and Opportunities
Yang, E., Shen, L., Guo, G., Wang, X., Cao, X., Zhang, J., & Tao, D. (2026) · ACM Computing Surveys · doi:10.1145/3787849 · Cited by 102 · see citing works
Knowledge Editing for Large Language Models: A Survey
Wang, S., Zhu, Y., Liu, H., Zheng, Z., Chen, C., & Li, J. (2025) · ACM Computing Surveys · doi:10.1145/3698590 · Cited by 79 · see citing works
Defending Large Language Models Against Jailbreak Attacks via Layer-specific Editing
Zhao, W., Li, Z., Li, Y., Ye, Z., & Sun, J. (2024) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2024.findings-emnlp.293 · Cited by 37 · see citing works
Digital forgetting in large language models: a survey of unlearning methods
Blanco-Justicia, A., Jebreel, N., Manzanares-Salor, B., Sánchez, D., Domingo-Ferrer, J., Collell, G., & Eeik Tan, K. (2025) · Artificial Intelligence Review · doi:10.1007/s10462-024-11078-6 · Cited by 31 · see citing works
Cloud-Edge Collaborative Defect Detection Based on Efficient Yolo Networks and Incremental Learning
Lei, Z., Zhang, Y., Wang, J., & Zhou, M. (2024) · Sensors · doi:10.3390/s24185921 · Cited by 16 · see citing works
Liang, J., Huang, W., Wan, G., Yang, Q., & Ye, M. (2025) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52734.2025.02437 · Cited by 7 · see citing works
Multi-Stage LLM Fine-Tuning with a Continual Learning Setting
Guan, C., Huang, C., Li, H., Li, Y., Cheng, N., Liu, Z., Chen, Y., Xu, J., & Liu, J. (2025) · Findings of the Association for Computational Linguistics: NAACL · doi:10.18653/v1/2025.findings-naacl.303 · Cited by 7 · see citing works
TACO: Enhancing Multimodal In-context Learning via Task Mapping-Guided Sequence Configuration
Li, Y., Yang, J., Yun, T., Feng, P., Huang, J., & Tang, R. (2025) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2025.emnlp-main.39 · Cited by 5 · see citing works
Preference Guided Meta-Learning for Cross Domain Time Series Forecasting
Li, X., Teng, F., Li, T., & Duan, Q. (2026) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2026.3658637 · Cited by 2 · see citing works
CoME: An Unlearning-based Approach to Conflict-free Model Editing
Jung, D., Seo, J., Lee, J., Park, C., & Lim, H. (2025) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2025.naacl-long.325 · Cited by 2 · see citing works
LoRASculpt: Harmonious Low-Rank Adaptation for Multimodal Large Language Models
Liang, J., Huang, W., Guo, X., Wan, G., Du, B., & Ye, M. (2026) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2026.3707616 · Cited by 1 · see citing works
CREME: Robustness Enhancement of Code LLMs via Layer-Aware Model Editing
Liu, S.-S., Hu, X., Huang, K., Yang, X., Lo, D., & Xia, X. (2026) · ACM/IEEE International Conference on Software Engineering: Companion Proceedings (ICSE-Companion) · doi:10.1145/3744916.3773111 · Cited by 1 · see citing works
Self-citationChen, D., Liu, Z., Fang, F., Leong, C. T., Ni, S., Argha, A., Alinejad‐Rokny, H., Yang, M., & Li, C. (2025) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2025.emnlp-main.645
Self-citationEnhancing Parameter-Efficient Code Representations with Retrieval and Structural Priors
Zheng, S., Li, Y., & Ma, X. (2026) · Applied Sciences · doi:10.3390/app16021106
MolEdit: Knowledge Editing for Multimodal Molecule Language Models
Lei, Z., Soga, P., Zhu, Y., He, Y., Dong, Y., & Li, J. (2026) · ACM International Conference on Web Search and Data Mining (WSDM) · doi:10.1145/3773966.3777928
Machine Unlearning in Large Language Models: A Survey of Challenges and MethodsPreprint
Tu, X., Zhu, T., Liu, Z., Xiong, P., & Zhou, W. (2026) · Preprints.org · doi:10.20944/preprints202603.0114.v1
KEditVis: A Visual Analytics System for Knowledge Editing of Large Language Models
Chen, Z., Zhan, H., Huang, Y., Wu, X., Deng, D., Weng, D., & Wu, Y. (2026) · IEEE Transactions on Visualization and Computer Graphics · doi:10.1109/tvcg.2026.3694436
Probability-Lift Fine-Tuning for Knowledge Editing: Target-Token Prior as a Signal of Update DemandPreprint
Jeong, H., Kim, G., & Kang, S. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7656178
- Authors
- Ni, S., Chen, D., Li, C., Hu, X., Xu, R., & Yang, M.
- Affiliations
- Chinese Academy of Sciences (CN)
- Harbin Institute of Technology (CN)
- Venue
- Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
- Published
- 2024
- Pages
- 5716-5731
- ISSN
- 0736-587X (ISSN-L)
- OpenAlex
- W4402671654
- Cited by
- 6
- References
- 0
- Citations by year
- 2026: 5 · 2025: 1
- FWCI
- 1.77
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- dx.doi.org
- Licence
- cc-by
- Locations
- Proceedings of the conference - Association for Computational Linguistics. Meeting (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Natural Science Foundation of China (62376262)
- Natural Science Foundation of Guangdong Province
- National Key Research and Development Program of China (2022YFF0902100)
- Shenzhen Science and Technology Innovation Program (KQTD20190929172835662)
- Basic and Applied Basic Research Foundation of Guangdong Province
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Keywords
- knowledge updating, large language models, low-rank adaptation, full fine-tuning, knowledge conflicts
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
