Cover Image

Escaping Optimization Stagnation: Taking Steps Beyond Task Arithmetic via Difference Vectors

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Wang, Jinping, Gao, Zhiqiang, Zhang, Dinggen, Xie, Zhiwu
Format:	Preprint
Published:	2025
Subjects:	Machine Learning Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Space Alignment Matters: The Missing Piece for Inducing Neural Collapse in Long-Tailed Learning
by: Wang, Jinping, et al.
Published: (2025)

Rethinking Loss Reweighting for Imbalance Learning as an Inverse Problem: A Neural Collapse Point of View
by: Wang, Jinping, et al.
Published: (2026)

Provable In-Context Vector Arithmetic via Retrieving Task Concepts
by: Bu, Dake, et al.
Published: (2025)

Bus-Conditioned Zero-Shot Trajectory Generation via Task Arithmetic
by: Liu, Shuai, et al.
Published: (2026)

Fix the Loss, Not the Radius: Rethinking the Adversarial Perturbation of Sharpness-Aware Minimization
by: Wang, Jinping, et al.
Published: (2026)

Backdoor Vectors: a Task Arithmetic View on Backdoor Attacks and Defenses
by: Pawlak, Stanisław, et al.
Published: (2025)

Take a Step and Reconsider: Sequence Decoding for Self-Improved Neural Combinatorial Optimization
by: Pirnay, Jonathan, et al.
Published: (2024)

Stagnation in Evolutionary Algorithms: Convergence $\neq$ Optimality
by: Zhou, Xiaojun
Published: (2025)

Boosting Maximum Entropy Reinforcement Learning via One-Step Flow Matching
by: Li, Zeqiao, et al.
Published: (2026)

FoldToken: Learning Protein Language via Vector Quantization and Beyond
by: Gao, Zhangyang, et al.
Published: (2024)

Generative AI Models for Different Steps in Architectural Design: A Literature Review
by: Li, Chengyuan, et al.
Published: (2024)

Variational Task Vector Composition
by: Zhang, Boyuan, et al.
Published: (2025)

Principled Understanding of Generalization for Generative Transformer Models in Arithmetic Reasoning Tasks
by: Xu, Xingcheng, et al.
Published: (2024)

Goldilocks RL: Tuning Task Difficulty to Escape Sparse Rewards for Reasoning
by: Mahrooghi, Ilia, et al.
Published: (2026)

On Task Vectors and Gradients
by: Zhou, Luca, et al.
Published: (2025)

Language Models Do Hard Arithmetic Tasks Easily and Hardly Do Easy Arithmetic Tasks
by: Gambardella, Andrew, et al.
Published: (2024)

Escaping the Verifier: Learning to Reason via Demonstrations
by: Cai, Locke, et al.
Published: (2025)

Error-Driven Prompt Optimization for Arithmetic Reasoning
by: Pándy, Árpád, et al.
Published: (2025)

Diffusion World Model: Future Modeling Beyond Step-by-Step Rollout for Offline Reinforcement Learning
by: Ding, Zihan, et al.
Published: (2024)

Take a Step Back: Evoking Reasoning via Abstraction in Large Language Models
by: Zheng, Huaixiu Steven, et al.
Published: (2023)

Beyond Sharp Minima: Robust LLM Unlearning via Feedback-Guided Multi-Point Optimization
by: Wu, Wenhan, et al.
Published: (2025)

Beyond Importance Sampling: Rejection-Gated Policy Optimization
by: Sun, Ziwu, et al.
Published: (2026)

Distilling Linearized Behavior into Non-Linear Fine-Tuning for Effective Task Arithmetic
by: Sommariva, Thomas, et al.
Published: (2026)

Efficient Multi-Task Reinforcement Learning via Task-Specific Action Correction
by: Feng, Jinyuan, et al.
Published: (2024)

Task Arithmetic in Trust Region: A Training-Free Model Merging Approach to Navigate Knowledge Conflicts
by: Sun, Wenju, et al.
Published: (2025)

CATA: Continual Machine Unlearning via Conflict-Averse Task Arithmetic
by: Lin, Shen, et al.
Published: (2026)

VP-VAE: Rethinking Vector Quantization via Adaptive Vector Perturbation
by: Zhai, Linwei, et al.
Published: (2026)

Transporting Task Vectors across Different Architectures without Training
by: Rinaldi, Filippo, et al.
Published: (2026)

VFA: Relieving Vector Operations in Flash Attention with Global Maximum Pre-computation
by: Sun, Yupeng, et al.
Published: (2026)

OccamLLM: Fast and Exact Language Model Arithmetic in a Single Step
by: Dugan, Owen, et al.
Published: (2024)

CPL: Critical Plan Step Learning Boosts LLM Generalization in Reasoning Tasks
by: Wang, Tianlong, et al.
Published: (2024)

Do Latent-CoT Models Think Step-by-Step? A Mechanistic Study on Sequential Reasoning Tasks
by: Liang, Jia, et al.
Published: (2026)

Deciphering Scientific Reasoning Steps from Outcome Data for Molecule Optimization
by: Liu, Zequn, et al.
Published: (2026)

Hey Pentti, We Did (More of) It!: A Vector-Symbolic Lisp With Residue Arithmetic
by: Hanley, Connor, et al.
Published: (2025)

Beyond Losses Reweighting: Empowering Multi-Task Learning via the Generalization Perspective
by: Phan, Hoang, et al.
Published: (2022)

Step-size Optimization for Continual Learning
by: Degris, Thomas, et al.
Published: (2024)

Learning to Generate Formally Verifiable Step-by-Step Logic Reasoning via Structured Formal Intermediaries
by: Chen, Luoxin, et al.
Published: (2026)

Multi-task Federated Learning with Encoder-Decoder Structure: Enabling Collaborative Learning Across Different Tasks
by: Zhou, Jingxuan, et al.
Published: (2025)

Dimer-Enhanced Optimization: A First-Order Approach to Escaping Saddle Points in Neural Network Training
by: Hu, Yue, et al.
Published: (2025)

Beyond Alignment: Expanding Reasoning Capacity via Manifold-Reshaping Policy Optimization
by: Wang, Dayu, et al.
Published: (2026)