Modality Collapse as Mismatched Decoding: Information-Theoretic Limits of Multimodal LLMs

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autore principale:	Billa, Jayadev
Natura:	Preprint
Pubblicazione:	2026
Soggetti:	Computation and Language Artificial Intelligence Machine Learning
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

The Geometric Anatomy of Capability Acquisition in Transformers
di: Billa, Jayadev
Pubblicazione: (2026)

The Cascade Equivalence Hypothesis: When Do Speech LLMs Behave Like ASR$\rightarrow$LLM Pipelines?
di: Billa, Jayadev
Pubblicazione: (2026)

Predicting Where Steering Vectors Succeed
di: Billa, Jayadev
Pubblicazione: (2026)

Understanding Multimodal LLMs Under Distribution Shifts: An Information-Theoretic Approach
di: Oh, Changdae, et al.
Pubblicazione: (2025)

When Audio-LLMs Don't Listen: A Cross-Linguistic Study of Modality Arbitration
di: Billa, Jayadev
Pubblicazione: (2026)

GTBench: Uncovering the Strategic Reasoning Limitations of LLMs via Game-Theoretic Evaluations
di: Duan, Jinhao, et al.
Pubblicazione: (2024)

The Price of Format: Diversity Collapse in LLMs
di: Yun, Longfei, et al.
Pubblicazione: (2025)

A Theoretical Perspective for Speculative Decoding Algorithm
di: Yin, Ming, et al.
Pubblicazione: (2024)

Distillation Contrastive Decoding: Improving LLMs Reasoning with Contrastive Decoding and Distillation
di: Phan, Phuc, et al.
Pubblicazione: (2024)

Theoretical Benefit and Limitation of Diffusion Language Model
di: Feng, Guhao, et al.
Pubblicazione: (2025)

Controlling Multimodal LLMs via Reward-guided Decoding
di: Mañas, Oscar, et al.
Pubblicazione: (2025)

MorphNAS: Differentiable Architecture Search for Morphologically-Aware Multilingual NER
di: Devadiga, Prathamesh, et al.
Pubblicazione: (2025)

HiSpec: Hierarchical Speculative Decoding for LLMs
di: Kumar, Avinash, et al.
Pubblicazione: (2025)

On Speculative Decoding for Multimodal Large Language Models
di: Gagrani, Mukul, et al.
Pubblicazione: (2024)

Draft-Conditioned Constrained Decoding for Structured Generation in LLMs
di: Reddy, Avinash, et al.
Pubblicazione: (2026)

Evolutionary Guided Decoding: Iterative Value Refinement for LLMs
di: Liu, Zhenhua, et al.
Pubblicazione: (2025)

When Attention Collapses: How Degenerate Layers in LLMs Enable Smaller, Stronger Models
di: Sanyal, Sunny, et al.
Pubblicazione: (2024)

Quantifying Modality Contributions via Disentangling Multimodal Representations
di: Amit, Padegal, et al.
Pubblicazione: (2025)

RL-PLUS: Countering Capability Boundary Collapse of LLMs in Reinforcement Learning with Hybrid-policy Optimization
di: Dong, Yihong, et al.
Pubblicazione: (2025)

Nudging: Inference-time Alignment of LLMs via Guided Decoding
di: Fei, Yu, et al.
Pubblicazione: (2024)

Weak-to-Strong Elicitation via Mismatched Wrong Drafts
di: Deng, Wei
Pubblicazione: (2026)

Diagnosing Training Inference Mismatch in LLM Reinforcement Learning
di: Zhong, Tianle, et al.
Pubblicazione: (2026)

Defeating the Training-Inference Mismatch via FP16
di: Qi, Penghui, et al.
Pubblicazione: (2025)

Connect, Collapse, Corrupt: Learning Cross-Modal Tasks with Uni-Modal Data
di: Zhang, Yuhui, et al.
Pubblicazione: (2024)

Calibration Collapse Under Sycophancy Fine-Tuning: How Reward Hacking Breaks Uncertainty Quantification in LLMs
di: Sahoo, Subramanyam
Pubblicazione: (2026)

Multimodal Prompt Optimization: Why Not Leverage Multiple Modalities for MLLMs
di: Choi, Yumin, et al.
Pubblicazione: (2025)

ZebraLogic: On the Scaling Limits of LLMs for Logical Reasoning
di: Lin, Bill Yuchen, et al.
Pubblicazione: (2025)

When Correct Beliefs Collapse: Epistemic Resilience of LLMs under Clinical Pressure
di: Xiao, Boyu, et al.
Pubblicazione: (2026)

Dual-Modality Multi-Stage Adversarial Safety Training: Robustifying Multimodal Web Agents Against Cross-Modal Attacks
di: Liu, Haoyu, et al.
Pubblicazione: (2026)

Cross-Modal Augmentation for Few-Shot Multimodal Fake News Detection
di: Jiang, Ye, et al.
Pubblicazione: (2024)

Direct Alignment of Draft Model for Speculative Decoding with Chat-Fine-Tuned LLMs
di: Goel, Raghavv, et al.
Pubblicazione: (2024)

Information-Theoretic Reward Decomposition for Generalizable RLHF
di: Mao, Liyuan, et al.
Pubblicazione: (2025)

BiLLM: Pushing the Limit of Post-Training Quantization for LLMs
di: Huang, Wei, et al.
Pubblicazione: (2024)

Accelerating Diffusion LLMs via Adaptive Parallel Decoding
di: Israel, Daniel, et al.
Pubblicazione: (2025)

A Surprising Failure? Multimodal LLMs and the NLVR Challenge
di: Wu, Anne, et al.
Pubblicazione: (2024)

Calibrated Surprise: An Information-Theoretic Account of Creative Quality
di: Zou, Bo, et al.
Pubblicazione: (2026)

KITE: Kernelized and Information Theoretic Exemplars for In-Context Learning
di: Singh, Vaibhav, et al.
Pubblicazione: (2025)

Decomposing the Depth Profile of Fine-Tuning
di: Billa, Jayadev
Pubblicazione: (2026)

Assessing Modality Bias in Video Question Answering Benchmarks with Multimodal Large Language Models
di: Park, Jean, et al.
Pubblicazione: (2024)

A Measure-Theoretic Analysis of Reasoning: Structural Generalization and Approximation Limits
di: Zhang, Yuyang, et al.
Pubblicazione: (2026)