Task Arithmetic

Follow this topic Ask about this topic

Momentum

3 papers in the last four weeks, against 2 the four weeks before. 0.0% of all new papers.

Jul 13Week of Sep 28

Latest papers 18

All topics

Task Vector Descent: Learning from Non-IID Batches
Oct 4, 2026Anton Baumann, Jonas Hübotter, Zeynep Akata +1Continual Learning for LLMs Task Arithmetic
ETH Zurich · Technical University of Munich
Platonic Task Arithmetic
Oct 1, 2026Junghwan Park, Woojin ChoVLM Adaptation Task Arithmetic
TelePIX
No Pain, More Gain: Iterative Merging for Effective Multi-Teacher On-Policy Distillation
Sep 28, 2026Seonghyeon Kim, Chaeyun Jang, Noah Lee +2Task Arithmetic Language Model Distillation
KAIST AI · Kakao
Reading the Gate, Not the Interference: Output-Side Interference Measurement Does Not Track Merge Collapse
Aug 12, 2026Chencheng ZhuTask Vector Merging Task Arithmetic
UNSW Sydney
When Do Task Vectors Interfere? Mapping the Validity Boundaries of Weight-Space Composition
Aug 10, 2026Chencheng Zhu, Xiaoyang Li, Taotao CaiPrompt Sensitivity Weight-Space Analysis
UNSW Sydney · University of Southern Queensland
Suppression Sticks, Locality Is Fragile: A Closed-Loop Target-and-Control Audit of Task-Vector Negation in VLA Policies
Aug 5, 2026Shaoguang Wang, Weiyu Guo, Rushi Dai +3Robotic Policy Evaluation Task Arithmetic
The Hong Kong University of Science and Technology (Guangzhou) · AI2 Robotics · The Hong Kong University of Science and Technology
First-Order Predictable but Pairwise Fragile: Local Task Adaptation in Trained Transformers
Jul 18, 2026Irina Piontkovskaia, Sergey NikolenkoFine-Tuning Neural Network Optimization
1DAIMLD, Moscow, Russia · 2St. Petersburg Department of the Steklov Institute of Mathematics, St. Petersburg, Russia · 3St. Petersburg State University
Overthinking: Amplifying Reasoning Weights to Extract Learned Secrets
Jul 9, 2026Jack Hopkins, Dipika Khullar, Fabien RogerOverthinking in Language Models Privacy Leakage in Language Models
Anthropic Fellows Program · MATS · Anthropic.
Revocable Learned State via Process Sidecars
Jun 29, 2026John SweeneyLLM Safety Alignment Task Arithmetic
Sideplane AI
From Parameters to Feature Space: Task Arithmetic for Backdoor Mitigation in Model Merging
Jun 10, 2026Zhenqian Zhu, Yamin Hu, Yiya Diao +3Task Arithmetic Model Merging
Institute of Cyberspace Security, School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen, 518055, China. · Shenzhen Key Laboratory of Media Security, Shenzhen University, Shenzhen, 518060, China.
Task-Vector Arithmetic for Emotional Expressivity Control in Language-Model-Based Text-to-Speech
Jun 3, 2026Daniel Oliveira de Brito, Arnaldo Candido JuniorTTS Synthesis Task Arithmetic
Instituto de Biociências, Letras e Ciências Exatas Universidade Estadual Paulista “Júlio de Mesquita Filho” (UNESP) São José do Rio Preto, SP, Brazil
Distilling Linearized Behavior into Non-Linear Fine-Tuning for Effective Task Arithmetic
May 18, 2026Thomas Sommariva, Francesca Morandi, Simone Calderara +1Fine-Tuning Task Arithmetic
ImageLab, Department of Engineering “Enzo Ferrari”, University of Modena and Reggio Emilia, Italy · University of Pisa, Italy
CATA: Continual Machine Unlearning via Conflict-Averse Task Arithmetic
May 18, 2026Shen Lin, Junhao Dong, Rongjie Chen +3VLM Unlearning Vision-Language Models
1Fujian Normal University · 2Nanyang Technological University · 3Xidian University
Ecologically-Constrained Task Arithmetic for Multi-Taxa Bioacoustic Classifiers Without Shared Data
May 5, 2026Ragib Amin Nihal, Benjamin Yen, Runwu Shi +2Task Arithmetic Bioacoustics
Systems and Control Engineering, Institute of Science Tokyo, Japan · RIKEN BDR, Japan
Model Merging: Foundations and Algorithms
May 2, 2026Donato CrisostomiMulti-Task Learning Task Vector Merging
SAPIENZA UNIVERSITY OF ROME · GLADIA research lab · Department of Computer Science +1
Decouple before Integration: Test-time Synthesis of SFT and RLVR Task Vectors
May 1, 2026Chaohao Yuan, Chenghao Xiao, Yu Rong +2Task Arithmetic Language Model Merging
Department of Systems Engineering and Engineering Management, Chinese University of Hong Kong, Hong Kong, China · DAMO Academy, Alibaba Group, Hangzhou, China · Hupan Lab, Hangzhou, China +1
Understanding and Enforcing Weight Disentanglement in Task Arithmetic
Apr 18, 2026Shangge Liu, Yuehan Yin, Lei Wang +5Task Vectors Weight-Space Analysis
State Key Laboratory for Novel Software Technology, Nanjing University, China · University of Wollongong, Australia · Nanyang Technological University, Singapore
Task Vector Bases: A Unified and Scalable Framework for Compressed Task Arithmetic
Feb 3, 2025Siqi Zeng, Yifei He, Meitong Liu +5Model Compression Task Vectors
University of Illinois Urbana-Champaign · University of Pennsylvania · Okinawa Institute of Science and Technology

Oct 4, 2026, cs.LGJ/K move · Enter open · S save

Task Vector Descent: Learning from Non-IID Batches

Anton Baumann, Jonas Hübotter, Zeynep Akata, Andreas Krause

ETH Zurich · Technical University of Munich

Continual Learning for LLMs Task Arithmetic Continued Pretraining Online Continual Learning

Open paper AskPDF arXiv

A central challenge in continual learning is to acquire new knowledge without forgetting what the model has already learned. This challenge appears in language model training when training data comes from various domain-, user-, or task-specific distributions that are encountered unevenly over time. In such settings, successive minibatches are temporally clustered by distribution instead of being sampled i.i.d. from the overall data mixture. Training on temporally clustered data induces a stability-plasticity tradeoff. Adapting the model to the active distribution can improve the model on the active distribution but may lead to a performance degradation on data it previously trained on. We find that this tradeoff intensifies with longer exposure to the same distribution. We therefore ask if the parameter displacement produced by such a sequence (the task vector) should be fully retained or applied partially. We compare applying the full displacement (

λ=1

) with partial integration, which scales the task vector by

λ

before applying it to the continuing model and scales the optimizer state by the same coefficient. Across continual pretraining, pretraining from random initialization, supervised post-training, and reinforcement post-training, we find that intermediate values of

λ

often improve average continuing-model performance relative to full integration, particularly after longer same-distribution sequences. In continual-pretraining experiments with both controlled streams and naturally defined adaptation sequences, task-vector scaling outperforms full integration at the matched learning rate, showing that its benefits are not reproduced by learning-rate scaling alone.

Momentum

Related topics

Latest papers 18