Imagen de Portada

SAE as a Crystal Ball: Interpretable Features Predict Cross-domain Transferability of LLMs without Training

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Zhang, Qi, Wang, Yifei, Wang, Xiaohan, Chai, Jiajun, Yin, Guojun, Lin, Wei, Wang, Yisen
Formato:	Preprint
Publicado:	2026
Materias:	Artificial Intelligence
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

SSL4RL: Revisiting Self-supervised Learning as Intrinsic Reward for Visual-Language Reasoning
por: Guo, Xiaojun, et al.
Publicado: (2025)

ZipRL: Adaptive Multi-Turn Context Compression with Hindsight Response Replay
por: Hu, Zhexin, et al.
Publicado: (2026)

Beyond Interpretability: The Gains of Feature Monosemanticity on Model Robustness
por: Zhang, Qi, et al.
Publicado: (2024)

CDRRM: Contrast-Driven Rubric Generation for Reliable and Interpretable Reward Modeling
por: Liu, Dengcan, et al.
Publicado: (2026)

Identifying and Understanding Cross-Class Features in Adversarial Training
por: Wei, Zeming, et al.
Publicado: (2025)

AMR-SD: Asymmetric Meta-Reflective Self-Distillation for Token-Level Credit Assignment
por: Wei, Zhenlin, et al.
Publicado: (2026)

An Augmentation Overlap Theory of Contrastive Learning
por: Zhang, Qi, et al.
Publicado: (2025)

MTIR-SQL: Multi-turn Tool-Integrated Reasoning Reinforcement Learning for Text-to-SQL
por: Xu, Zekun, et al.
Publicado: (2025)

RLAE: Reinforcement Learning-Assisted Ensemble for LLMs
por: Fu, Yuqian, et al.
Publicado: (2025)

ToolForge: A Data Synthesis Pipeline for Multi-Hop Search without Real-World APIs
por: Chen, Hao, et al.
Publicado: (2025)

AutoSearch: Adaptive Search Depth for Efficient Agentic RAG via Reinforcement Learning
por: Sun, Jingbo, et al.
Publicado: (2026)

Contextual Rollout Bandits for Reinforcement Learning with Verifiable Rewards
por: Lu, Xiaodong, et al.
Publicado: (2026)

RLFactory: A Plug-and-Play Reinforcement Learning Post-Training Framework for LLM Multi-Turn Tool-Use
por: Chai, Jiajun, et al.
Publicado: (2025)

LANPO: Bootstrapping Language and Numerical Feedback for Reinforcement Learning in LLMs
por: Li, Ang, et al.
Publicado: (2025)

G1: Teaching LLMs to Reason on Graphs with Reinforcement Learning
por: Guo, Xiaojun, et al.
Publicado: (2025)

Training Multi-Image Vision Agents via End2End Reinforcement Learning
por: Dong, Chengqi, et al.
Publicado: (2025)

Are Smarter LLMs Safer? Exploring Safety-Reasoning Trade-offs in Prompting and Fine-Tuning
por: Li, Ang, et al.
Publicado: (2025)

Promoting Efficient Reasoning with Verifiable Stepwise Reward
por: Yue, Chuhuai, et al.
Publicado: (2025)

Non-negative Contrastive Learning
por: Wang, Yifei, et al.
Publicado: (2024)

Dissecting the Failure of Invariant Learning on Graphs
por: Wang, Qixun, et al.
Publicado: (2024)

Can In-context Learning Really Generalize to Out-of-distribution Tasks?
por: Wang, Qixun, et al.
Publicado: (2024)

Joint Training of Multi-Token Prediction in Reinforcement Learning via Optimal Coefficient Calibration
por: Wang, Zili, et al.
Publicado: (2026)

Do Generated Data Always Help Contrastive Learning?
por: Wang, Yifei, et al.
Publicado: (2024)

When More is Less: Understanding Chain-of-Thought Length in LLMs
por: Wu, Yuyang, et al.
Publicado: (2025)

Enhancing Cross-domain Link Prediction via Evolution Process Modeling
por: Huang, Xuanwen, et al.
Publicado: (2024)

Finding and Reactivating Post-Trained LLMs' Hidden Safety Mechanisms
por: Li, Mingjie, et al.
Publicado: (2026)

LouvreSAE: Sparse Autoencoders for Interpretable and Controllable Style Transfer
por: Panda, Raina, et al.
Publicado: (2025)

How to Craft Backdoors with Unlabeled Data Alone?
por: Wang, Yifei, et al.
Publicado: (2024)

Compositional Literary Primitives in Instruction-Tuned LLMs: Cross-Architectural SAE Features for Self, Style, and Affect
por: Presa, Joao Paulo Cavalcante, et al.
Publicado: (2026)

Committed SAE-Feature Traces for Audited-Session Substitution Detection in Hosted LLMs
por: Liu, Ziyang
Publicado: (2026)

Rethinking Invariance in In-context Learning
por: Fang, Lizhe, et al.
Publicado: (2025)

LocalSearchBench: Benchmarking Agentic Search in Real-World Local Life Services
por: He, Hang, et al.
Publicado: (2025)

Jailbreak and Guard Aligned Language Models with Only Few In-Context Demonstrations
por: Wei, Zeming, et al.
Publicado: (2023)

On the Adversarial Transferability of Generalized "Skip Connections"
por: Wang, Yisen, et al.
Publicado: (2024)

Generalist++: A Meta-learning Framework for Mitigating Trade-off in Adversarial Training
por: Wang, Yisen, et al.
Publicado: (2025)

SRFT: A Single-Stage Method with Supervised and Reinforcement Fine-Tuning for Reasoning
por: Fu, Yuqian, et al.
Publicado: (2025)

Language Ranker: A Lightweight Ranking framework for LLM Decoding
por: Zhang, Chenheng, et al.
Publicado: (2025)

FaithfulSAE: Towards Capturing Faithful Features with Sparse Autoencoders without External Dataset Dependencies
por: Cho, Seonglae, et al.
Publicado: (2025)

Autoregressive Models Rival Diffusion Models at ANY-ORDER Generation
por: Du, Tianqi, et al.
Publicado: (2026)

Cross-Document Cross-Lingual NLI via RST-Enhanced Graph Fusion and Interpretability Prediction
por: Yuan, Mengying, et al.
Publicado: (2025)