Cover Image

The Impact of Off-Policy Training Data on Probe Generalisation

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Kirch, Nathalie, Dower, Samuel, Skapars, Adrians, Yannakoudakis, Helen, Lubana, Ekdeep Singh, Krasheninnikov, Dmitrii
Format:	Preprint
Published:	2025
Subjects:	Artificial Intelligence Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Do Linear Probes Generalize Better in Persona Coordinates?
by: Mahadik, Prasad, et al.
Published: (2026)

How Do LLMs Persuade? Linear Probes Can Uncover Persuasion Dynamics in Multi-Turn Conversations
by: Jaipersaud, Brandon, et al.
Published: (2025)

GPT, But Backwards: Exactly Inverting Language Model Outputs
by: Skapars, Adrians, et al.
Published: (2025)

A Percolation Model of Emergence: Analyzing Transformers Trained on a Formal Language
by: Lubana, Ekdeep Singh, et al.
Published: (2024)

Was it Slander? Towards Exact Inversion of Generative Language Models
by: Skapars, Adrians, et al.
Published: (2024)

Projecting Assumptions: The Duality Between Sparse Autoencoders and Concept Geometry
by: Hindupur, Sai Sumedh R., et al.
Published: (2025)

Emergence of Hidden Capabilities: Exploring Learning Dynamics in Concept Space
by: Park, Core Francisco, et al.
Published: (2024)

Detecting High-Stakes Interactions with Activation Probes
by: McKenzie, Alex, et al.
Published: (2025)

Worst-Case Symbolic Constraints Analysis and Generalisation with Large Language Models
by: Koh, Daniel, et al.
Published: (2025)

Fresh in memory: Training-order recency is linearly encoded in language model activations
by: Krasheninnikov, Dmitrii, et al.
Published: (2025)

In-Context Learning Strategies Emerge Rationally
by: Wurgaft, Daniel, et al.
Published: (2025)

Implicit meta-learning may lead language models to trust more reliable sources
by: Krasheninnikov, Dmitrii, et al.
Published: (2023)

From Isolation to Entanglement: When Do Interpretability Methods Identify and Disentangle Known Concepts?
by: Mueller, Aaron, et al.
Published: (2025)

In-Context Learning Dynamics with Random Binary Sequences
by: Bigelow, Eric J., et al.
Published: (2023)

Towards an Understanding of Stepwise Inference in Transformers: A Synthetic Graph Navigation Model
by: Khona, Mikail, et al.
Published: (2024)

A Functional Perspective on Knowledge Distillation in Neural Networks
by: Mason-Williams, Israel, et al.
Published: (2025)

Transformers Use Causal World Models in Maze-Solving Tasks
by: Spies, Alex F., et al.
Published: (2024)

Belief Dynamics Reveal the Dual Nature of In-Context Learning and Activation Steering
by: Bigelow, Eric, et al.
Published: (2025)

Emergence of Hierarchical Emotion Organization in Large Language Models
by: Zhao, Bo, et al.
Published: (2025)

What Features in Prompts Jailbreak LLMs? Investigating the Mechanisms Behind Attacks
by: Kirch, Nathalie, et al.
Published: (2024)

Data Poisoning Attacks on Off-Policy Policy Evaluation Methods
by: Lobo, Elita, et al.
Published: (2024)

ICLR: In-Context Learning of Representations
by: Park, Core Francisco, et al.
Published: (2024)

Stories in Space: In-Context Learning Trajectories in Conceptual Belief Space
by: Bigelow, Eric, et al.
Published: (2026)

Trust the Batch, On- or Off-Policy: Adaptive Policy Optimization for RL Post-Training
by: Fakoor, Rasool, et al.
Published: (2026)

A Function-Centric Perspective on Flat and Sharp Minima
by: Mason-Williams, Israel, et al.
Published: (2025)

Abrupt Learning in Transformers: A Case Study on Matrix Completion
by: Gopalani, Pulkit, et al.
Published: (2024)

Do Sparse Autoencoders Capture Concept Manifolds?
by: Bhalla, Usha, et al.
Published: (2026)

VESPO: Variational Sequence-Level Soft Policy Optimization for Stable Off-Policy LLM Training
by: Shen, Guobin, et al.
Published: (2026)

Off-OAB: Off-Policy Policy Gradient Method with Optimal Action-Dependent Baseline
by: Meng, Wenjia, et al.
Published: (2024)

The Impact of On-Policy Parallelized Data Collection on Deep Reinforcement Learning Networks
by: Mayor, Walter, et al.
Published: (2025)

SOPE: Stabilizing Off-Policy Evaluation for Online RL with Prior Data
by: Romeo, Carlo, et al.
Published: (2026)

How Ensembles of Distilled Policies Improve Generalisation in Reinforcement Learning
by: Weltevrede, Max, et al.
Published: (2025)

Turning Sand to Gold: Recycling Data to Bridge On-Policy and Off-Policy Learning via Causal Bound
by: Fiskus, Tal, et al.
Published: (2025)

Clustering Context in Off-Policy Evaluation
by: Guzman-Olivares, Daniel, et al.
Published: (2025)

Zero-Shot Off-Policy Learning
by: Asadulaev, Arip, et al.
Published: (2026)

Concept-driven Off Policy Evaluation
by: Majumdar, Ritam, et al.
Published: (2024)

Off-Policy Actor-Critic for Adversarial Observation Robustness: Virtual Alternative Training via Symmetric Policy Evaluation
by: Nakanishi, Kosuke, et al.
Published: (2025)

Adversarial Latent-State Training for Robust Policies in Partially Observable Domains
by: Ahuja, Angad Singh
Published: (2026)

Large Language Model Post-Training: A Unified View of Off-Policy and On-Policy Learning
by: Zhao, Shiwan, et al.
Published: (2026)

Learning Action Embeddings for Off-Policy Evaluation
by: Cief, Matej, et al.
Published: (2023)