Copertina

What Makes and Breaks Safety Fine-tuning? A Mechanistic Study

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Jain, Samyak, Lubana, Ekdeep Singh, Oksuz, Kemal, Joy, Tom, Torr, Philip H. S., Sanyal, Amartya, Dokania, Puneet K.
Natura:	Preprint
Pubblicazione:	2024
Soggetti:	Machine Learning Computation and Language
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

MoCaE: Mixture of Calibrated Experts Significantly Improves Object Detection
di: Oksuz, Kemal, et al.
Pubblicazione: (2023)

Segment, Select, Correct: A Framework for Weakly-Supervised Referring Segmentation
di: Eiras, Francisco, et al.
Pubblicazione: (2023)

Fine-tuning can cripple your foundation model; preserving features may be the solution
di: Mukhoti, Jishnu, et al.
Pubblicazione: (2023)

Mechanistically analyzing the effects of fine-tuning on procedurally defined tasks
di: Jain, Samyak, et al.
Pubblicazione: (2023)

On Calibration of Object Detectors: Pitfalls, Evaluation and Baselines
di: Kuzucu, Selim, et al.
Pubblicazione: (2024)

Foundation Models for Trajectory Planning in Autonomous Driving: A Review of Progress and Open Challenges
di: Oksuz, Kemal, et al.
Pubblicazione: (2025)

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

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

Placing Objects in Context via Inpainting for Out-of-distribution Segmentation
di: de Jorge, Pau, et al.
Pubblicazione: (2024)

Random Representations Outperform Online Continually Learned Representations
di: Prabhu, Ameya, et al.
Pubblicazione: (2024)

Corrective Machine Unlearning
di: Goel, Shashwat, et al.
Pubblicazione: (2024)

Are language models aware of the road not taken? Token-level uncertainty and hidden state dynamics
di: Zur, Amir, et al.
Pubblicazione: (2025)

Towards Reliable Evaluation of Behavior Steering Interventions in LLMs
di: Pres, Itamar, et al.
Pubblicazione: (2024)

Uncovering Conceptual Blindspots in Generative Image Models Using Sparse Autoencoders
di: Bohacek, Matyas, et al.
Pubblicazione: (2025)

Analyzing (In)Abilities of SAEs via Formal Languages
di: Menon, Abhinav, et al.
Pubblicazione: (2024)

Compositional Capabilities of Autoregressive Transformers: A Study on Synthetic, Interpretable Tasks
di: Ramesh, Rahul, et al.
Pubblicazione: (2023)

Rethinking Safety in LLM Fine-tuning: An Optimization Perspective
di: Kim, Minseon, et al.
Pubblicazione: (2025)

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

Competition Dynamics Shape Algorithmic Phases of In-Context Learning
di: Park, Core Francisco, et al.
Pubblicazione: (2024)

Compositional Abilities Emerge Multiplicatively: Exploring Diffusion Models on a Synthetic Task
di: Okawa, Maya, et al.
Pubblicazione: (2023)

Representation Shattering in Transformers: A Synthetic Study with Knowledge Editing
di: Nishi, Kento, et al.
Pubblicazione: (2024)

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

From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit
di: Costa, Valérie, et al.
Pubblicazione: (2025)

Evaluating Sparse Autoencoders: From Shallow Design to Matching Pursuit
di: Costa, Valérie, et al.
Pubblicazione: (2025)

Mixture of Experts Made Intrinsically Interpretable
di: Yang, Xingyi, et al.
Pubblicazione: (2025)

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

An Iterative Algorithm for Differentially Private $k$-PCA with Adaptive Noise
di: Düngler, Johanna, et al.
Pubblicazione: (2025)

The Impact of Off-Policy Training Data on Probe Generalisation
di: Kirch, Nathalie, et al.
Pubblicazione: (2025)

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

Modeling Gene Expression Distributional Shifts for Unseen Genetic Perturbations
di: Ramakrishnan, Kalyan, et al.
Pubblicazione: (2025)

Mechanistic Fine-tuning for In-context Learning
di: Cho, Hakaze, et al.
Pubblicazione: (2025)

AnnoCaseLaw: A Richly-Annotated Dataset For Benchmarking Explainable Legal Judgment Prediction
di: Sesodia, Magnus, et al.
Pubblicazione: (2025)

Force Dipole Interactions in Tubular Fluid Membranes
di: Jain, Samyak, et al.
Pubblicazione: (2023)

Nuclear stability and the Fold Catastrophe
di: Jain, Samyak, et al.
Pubblicazione: (2023)

Tunneling half-lives in macroscopic-microscopic picture
di: Jain, Samyak, et al.
Pubblicazione: (2024)

Catastrophe theoretic approach to the Higgs Mechanism
di: Jain, Samyak, et al.
Pubblicazione: (2023)

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

Swing-by Dynamics in Concept Learning and Compositional Generalization
di: Yang, Yongyi, et al.
Pubblicazione: (2024)

Do as I do (Safely): Mitigating Task-Specific Fine-tuning Risks in Large Language Models
di: Eiras, Francisco, et al.
Pubblicazione: (2024)

Attention-Enhanced Hybrid Feature Aggregation Network for 3D Brain Tumor Segmentation
di: Yazıcı, Ziya Ata, et al.
Pubblicazione: (2024)