Imagen de Portada

Differentiable Faithfulness Alignment for Cross-Model Circuit Transfer

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Shao, Shun, Wang, Binxu, Cohen, Shay B., Korhonen, Anna, Belinkov, Yonatan
Formato:	Preprint
Publicado:	2026
Materias:	Computation and Language
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

Have Faith in Faithfulness: Going Beyond Circuit Overlap When Finding Model Mechanisms
por: Hanna, Michael, et al.
Publicado: (2024)

Iterative Multilingual Spectral Attribute Erasure
por: Shao, Shun, et al.
Publicado: (2025)

Measuring Chain of Thought Faithfulness by Unlearning Reasoning Steps
por: Tutek, Martin, et al.
Publicado: (2025)

Old Habits Die Hard: How Conversational History Geometrically Traps LLMs
por: Simhi, Adi, et al.
Publicado: (2026)

Spectral Editing of Activations for Large Language Model Alignment
por: Qiu, Yifu, et al.
Publicado: (2024)

BlackboxNLP-2025 MIB Shared Task: Improving Circuit Faithfulness via Better Edge Selection
por: Nikankin, Yaniv, et al.
Publicado: (2025)

ContraSim -- Analyzing Neural Representations Based on Contrastive Learning
por: Rahamim, Adir, et al.
Publicado: (2023)

Back Attention: Understanding and Enhancing Multi-Hop Reasoning in Large Language Models
por: Yu, Zeping, et al.
Publicado: (2025)

Bootstrapping Action-Grounded Visual Dynamics in Unified Vision-Language Models
por: Qiu, Yifu, et al.
Publicado: (2025)

Same Task, Different Circuits: Disentangling Modality-Specific Mechanisms in VLMs
por: Nikankin, Yaniv, et al.
Publicado: (2025)

REVS: Unlearning Sensitive Information in Language Models via Rank Editing in the Vocabulary Space
por: Ashuach, Tomer, et al.
Publicado: (2024)

Think While You Write: Hypothesis Verification Promotes Faithful Knowledge-to-Text Generation
por: Qiu, Yifu, et al.
Publicado: (2023)

Position-aware Automatic Circuit Discovery
por: Haklay, Tal, et al.
Publicado: (2025)

Leveraging Prototypical Representations for Mitigating Social Bias without Demographic Information
por: Iskander, Shadi, et al.
Publicado: (2024)

Concept-Best-Matching: Evaluating Compositionality in Emergent Communication
por: Carmeli, Boaz, et al.
Publicado: (2024)

ReFACT: Updating Text-to-Image Models by Editing the Text Encoder
por: Arad, Dana, et al.
Publicado: (2023)

Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models
por: Marks, Samuel, et al.
Publicado: (2024)

Are formal and functional linguistic mechanisms dissociated in language models?
por: Hanna, Michael, et al.
Publicado: (2025)

SAEs Are Good for Steering -- If You Select the Right Features
por: Arad, Dana, et al.
Publicado: (2025)

Planted in Pretraining, Swayed by Finetuning: A Case Study on the Origins of Cognitive Biases in LLMs
por: Itzhak, Itay, et al.
Publicado: (2025)

Structured RAG for Answering Aggregative Questions
por: Koshorek, Omri, et al.
Publicado: (2025)

DEPTH: Discourse Education through Pre-Training Hierarchically
por: Bamberger, Zachary, et al.
Publicado: (2024)

Self-Improving World Modelling with Latent Actions
por: Qiu, Yifu, et al.
Publicado: (2026)

Follow the Flow: On Information Flow Across Textual Tokens in Text-to-Image Models
por: Kaplan, Guy, et al.
Publicado: (2025)

Findings of the BlackboxNLP 2025 Shared Task: Localizing Circuits and Causal Variables in Language Models
por: Arad, Dana, et al.
Publicado: (2025)

Backward Lens: Projecting Language Model Gradients into the Vocabulary Space
por: Katz, Shahar, et al.
Publicado: (2024)

Can Large Language Model Summarizers Adapt to Diverse Scientific Communication Goals?
por: Fonseca, Marcio, et al.
Publicado: (2024)

Can Large Language Models Follow Concept Annotation Guidelines? A Case Study on Scientific and Financial Domains
por: Fonseca, Marcio, et al.
Publicado: (2023)

Arithmetic Without Algorithms: Language Models Solve Math With a Bag of Heuristics
por: Nikankin, Yaniv, et al.
Publicado: (2024)

Fast Forwarding Low-Rank Training
por: Rahamim, Adir, et al.
Publicado: (2024)

Will it Merge? On The Causes of Model Mergeability
por: Rahamim, Adir, et al.
Publicado: (2026)

DeLeaker: Dynamic Inference-Time Reweighting For Semantic Leakage Mitigation in Text-to-Image Models
por: Ventura, Mor, et al.
Publicado: (2025)

Distinguishing Ignorance from Error in LLM Hallucinations
por: Simhi, Adi, et al.
Publicado: (2024)

Constructing Benchmarks and Interventions for Combating Hallucinations in LLMs
por: Simhi, Adi, et al.
Publicado: (2024)

Context-aware Prompt Tuning: Advancing In-Context Learning with Adversarial Methods
por: Blau, Tsachi, et al.
Publicado: (2024)

Modeling News Interactions and Influence for Financial Market Prediction
por: Wang, Mengyu, et al.
Publicado: (2024)

Silent Tokens, Loud Effects: Padding in LLMs
por: Himelstein, Rom, et al.
Publicado: (2025)

From Feelings to Metrics: Understanding and Formalizing How Users Vibe-Test LLMs
por: Itzhak, Itay, et al.
Publicado: (2026)

Reasoning Models Know What's Important, and Encode It in Their Activations
por: Nikankin, Yaniv, et al.
Publicado: (2026)

Beyond Transfer Accuracy: Faithful Circuits for Controlled Low-Resource Adaptation
por: Nur'aini, Khumaisa, et al.
Publicado: (2026)