Image de couverture de livre

Geometric Deviation as an Unsupervised Pre-Generation Reliability Signal: Probing LLM Representations for Answerability

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteur principal:	Du, Yucheng
Format:	Preprint
Publié:	2026
Sujets:	Computation and Language Machine Learning I.2.7
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

WSM: Decay-Free Learning Rate Schedule via Checkpoint Merging for LLM Pre-training
par: Tian, Changxin, et autres
Publié: (2025)

The Geometry of Harmful Intent: Training-Free Anomaly Detection via Angular Deviation in LLM Residual Streams
par: Llorente-Saguer, Isaac
Publié: (2026)

$S^3$ -- Semantic Signal Separation
par: Kardos, Márton, et autres
Publié: (2024)

Evaluating the Efficacy of Hybrid Deep Learning Models in Distinguishing AI-Generated Text
par: Oketunji, Abiodun Finbarrs
Publié: (2023)

PowLU: An Activation Function for Stable Pre-Training of LLMs
par: Jiang, Peijie, et autres
Publié: (2026)

Influence-driven Curriculum Learning for Pre-training on Limited Data
par: Schoenegger, Loris, et autres
Publié: (2025)

Large Language Model (LLM) Bias Index -- LLMBI
par: Oketunji, Abiodun Finbarrs, et autres
Publié: (2023)

Resonant Context Anchoring: Decoupling Attention Routing and Signal Gain at Inference Time
par: Zhao, Mingkuan, et autres
Publié: (2026)

Weakly Supervised Distillation of Hallucination Signals into Transformer Representations
par: Salehmohamed, Shoaib Sadiq, et autres
Publié: (2026)

Harmful Intent as a Geometrically Recoverable Feature of LLM Residual Streams
par: Llorente-Saguer, Isaac
Publié: (2026)

QUAD: Quantization and Parameter-Efficient Tuning of LLM with Activation Decomposition
par: Hu, Yuxuan, et autres
Publié: (2025)

Generalizable LLM Learning of Graph Synthetic Data with Post-training Alignment
par: Zhang, Yizhuo, et autres
Publié: (2025)

Pressure-Testing Deception Probes in LLMs: Scaling, Robustness, and the Geometry of Deceptive Representations
par: Kumar, Sachin
Publié: (2026)

Less Is More: Cognitive Load and the Single-Prompt Ceiling in LLM Mathematical Reasoning
par: Cazares, Manuel Israel
Publié: (2026)

HYPEROFA: Expanding LLM Vocabulary to New Languages via Hypernetwork-Based Embedding Initialization
par: Özeren, Enes, et autres
Publié: (2025)

Pre-trained Models Perform the Best When Token Distributions Follow Zipf's Law
par: He, Yanjin, et autres
Publié: (2025)

Integrating Expert Labels into LLM-based Emission Goal Detection: Example Selection vs Automatic Prompt Design
par: Wrzalik, Marco, et autres
Publié: (2024)

PersonalLLM: Tailoring LLMs to Individual Preferences
par: Zollo, Thomas P., et autres
Publié: (2024)

LLM Vocabulary Compression for Low-Compute Environments
par: Vennam, Sreeram, et autres
Publié: (2024)

Harnessing Negative Signals: Reinforcement Distillation from Teacher Data for LLM Reasoning
par: Xu, Shuyao, et autres
Publié: (2025)

Representation-Aware Unlearning via Activation Signatures: From Suppression to Entity-Signature Erasure
par: Mahmood, Syed Naveed, et autres
Publié: (2026)

Kronecker Embeddings: Byte-Level Structured Token Representations for Parameter-Efficient Language Models
par: Shravan, Rohan
Publié: (2026)

Trading Complexity for Expressivity Through Structured Generalized Linear Token Mixing
par: Fagnou, Erwan, et autres
Publié: (2026)

Quantization-Robust LLM Unlearning via Low-Rank Adaptation
par: Abitante, João Vitor Boer, et autres
Publié: (2026)

HyDRA: Hybrid Dynamic Routing Architecture for Heterogeneous LLM Pools
par: Garg, Aashna, et autres
Publié: (2026)

Decoding-Time Debiasing via Process Reward Models: From Controlled Fill-in to Open-Ended Generation
par: Khan, Muneeb Ur Raheem
Publié: (2026)

GenAI Content Detection Task 3: Cross-Domain Machine-Generated Text Detection Challenge
par: Dugan, Liam, et autres
Publié: (2025)

Are LLM Uncertainty and Correctness Encoded by the Same Features? A Functional Dissociation via Sparse Autoencoders
par: Patel, Het, et autres
Publié: (2026)

Engineering A Large Language Model From Scratch
par: Oketunji, Abiodun Finbarrs
Publié: (2024)

Anka: A Domain-Specific Language for Reliable LLM Code Generation
par: Mazrouei, Saif Khalfan Saif Al
Publié: (2025)

Thread Detection and Response Generation using Transformers with Prompt Optimisation
par: T, Kevin Joshua, et autres
Publié: (2024)

The Metacognitive Probe: Five Behavioural Calibration Diagnostics for LLMs
par: Oliveira, Rafael C. T.
Publié: (2026)

Ouroboros: Dynamic Weight Generation for Recursive Transformers via Input-Conditioned LoRA Modulation
par: Jaber, Jaber, et autres
Publié: (2026)

Reliable Part-of-Speech Tagging of Historical Corpora through Set-Valued Prediction
par: Heid, Stefan, et autres
Publié: (2020)

Do Personality Traits Interfere? Geometric Limitations of Steering in Large Language Models
par: Bhandari, Pranav, et autres
Publié: (2026)

The Architecture of Errors: From Universal Impossibility to Patch-Local LLM Reliability
par: Arbuzov, Mikhail L., et autres
Publié: (2026)

CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features
par: Cho, Seonglae, et autres
Publié: (2025)

Towards Latent Diffusion Suitable For Text
par: Midavaine, Nesta, et autres
Publié: (2026)

SpectralLoRA: Is Low-Frequency Structure Sufficient for LoRA Adaptation? A Spectral Analysis of Weight Updates
par: Singh, Rajveer
Publié: (2026)

FIM-LoRA: Task-Informative Rank Allocation for LoRA via Calibration-Time Gradient-Variance Estimation
par: Sathyavageeswaran, Ramakrishnan
Publié: (2026)