A Theoretical Perspective: How to Prevent Model Collapse in Self-consuming Training Loops

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Fu, Shi, Wang, Yingjie, Chen, Yuzhu, Tian, Xinmei, Tao, Dacheng
Natura:	Preprint
Pubblicazione:	2025
Soggetti:	Machine Learning Artificial Intelligence
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

Towards Theoretical Understandings of Self-Consuming Generative Models
di: Fu, Shi, et al.
Pubblicazione: (2024)

A Theoretical Survey on Foundation Models
di: Fu, Shi, et al.
Pubblicazione: (2024)

Drawback of Enforcing Equivariance and its Compensation via the Lens of Expressive Power
di: Chen, Yuzhu, et al.
Pubblicazione: (2025)

CoVeR: Conformal Calibration for Versatile and Reliable Autoregressive Next-Token Prediction
di: Chen, Yuzhu, et al.
Pubblicazione: (2025)

HRP: High-Rank Preheating for Superior LoRA Initialization
di: Chen, Yuzhu, et al.
Pubblicazione: (2025)

Why Self-Rewarding Works: Theoretical Guarantees for Iterative Alignment of Language Models
di: Fu, Shi, et al.
Pubblicazione: (2026)

Preventing Curriculum Collapse in Self-Evolving Reasoning Systems
di: Mishra, Vaibhav
Pubblicazione: (2026)

Preventing Dimensional Collapse in Self-Supervised Learning via Orthogonality Regularization
di: He, Junlin, et al.
Pubblicazione: (2024)

Understanding Representation of Deep Equilibrium Models from Neural Collapse Perspective
di: Sun, Haixiang, et al.
Pubblicazione: (2024)

Preventing Model Collapse via Contraction-Conditioned Neural Filters
di: Han, Zongjian, et al.
Pubblicazione: (2025)

Self-Correcting Self-Consuming Loops for Generative Model Training
di: Gillman, Nate, et al.
Pubblicazione: (2024)

InfoRM: Mitigating Reward Hacking in RLHF via Information-Theoretic Reward Modeling
di: Miao, Yuchun, et al.
Pubblicazione: (2024)

How Bad is Training on Synthetic Data? A Statistical Analysis of Language Model Collapse
di: Seddik, Mohamed El Amine, et al.
Pubblicazione: (2024)

Preventing Collapse in Contrastive Learning with Orthonormal Prototypes (CLOP)
di: Li, Huanran, et al.
Pubblicazione: (2024)

CURE: Critical-Token-Guided Re-Concatenation for Entropy-Collapse Prevention
di: Li, Qingbin, et al.
Pubblicazione: (2025)

Theoretical Perspectives on Data Quality and Synergistic Effects in Pre- and Post-Training Reasoning Models
di: Javanmard, Adel, et al.
Pubblicazione: (2026)

TQL: Scaling Q-Functions with Transformers by Preventing Attention Collapse
di: Dong, Perry, et al.
Pubblicazione: (2026)

Low-Precision Training of Large Language Models: Methods, Challenges, and Opportunities
di: Hao, Zhiwei, et al.
Pubblicazione: (2025)

Understanding and Preventing Entropy Collapse in RLVR with On-Policy Entropy Flow Optimization
di: Xu, Huimin, et al.
Pubblicazione: (2026)

Theoretical Modeling of Large Language Model Self-Improvement Training Dynamics Through Solver-Verifier Gap
di: Sun, Yifan, et al.
Pubblicazione: (2025)

Benefits and Pitfalls of Reinforcement Learning for Language Model Planning: A Theoretical Perspective
di: Wang, Siwei, et al.
Pubblicazione: (2025)

When and How Human Curation Backfires: Preference Alignment under Multi-Model Self-Consuming Loop
di: Zhang, Yang, et al.
Pubblicazione: (2026)

Revisiting Plasticity in Visual Reinforcement Learning: Data, Modules and Training Stages
di: Ma, Guozheng, et al.
Pubblicazione: (2023)

A Step Back: Prefix Importance Ratio Stabilizes Policy Optimization
di: Lei, Shiye, et al.
Pubblicazione: (2026)

Adaptive Defense against Harmful Fine-Tuning for Large Language Models via Bayesian Data Scheduler
di: Hu, Zixuan, et al.
Pubblicazione: (2025)

SMILE: Zero-Shot Sparse Mixture of Low-Rank Experts Construction From Pre-Trained Foundation Models
di: Tang, Anke, et al.
Pubblicazione: (2024)

LaDi-RL: Latent Diffusion Reasoning Prevents Entropy Collapse in Reinforcement Learning
di: Kang, Haoqiang, et al.
Pubblicazione: (2026)

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

Offline Behavior Distillation
di: Lei, Shiye, et al.
Pubblicazione: (2024)

Continual Learning on Graphs: Challenges, Solutions, and Opportunities
di: Zhang, Xikun, et al.
Pubblicazione: (2024)

Offline Behavioral Data Selection
di: Lei, Shiye, et al.
Pubblicazione: (2025)

Efficient Differentiable Causal Discovery via Reliable Super-Structure Learning
di: Ma, Pingchuan, et al.
Pubblicazione: (2026)

Generalized Category Discovery with Large Language Models in the Loop
di: An, Wenbin, et al.
Pubblicazione: (2023)

Multi-modal Synthetic Data Training and Model Collapse: Insights from VLMs and Diffusion Models
di: Hu, Zizhao, et al.
Pubblicazione: (2025)

Continual Task Learning through Adaptive Policy Self-Composition
di: Hu, Shengchao, et al.
Pubblicazione: (2024)

Enhancing Pre-Trained Model-Based Class-Incremental Learning through Neural Collapse
di: He, Kun, et al.
Pubblicazione: (2025)

ForTIFAI: Fending Off Recursive Training Induced Failure for AI Model Collapse
di: Shabgahi, Soheil Zibakhsh, et al.
Pubblicazione: (2025)

How to Synthesize Text Data without Model Collapse?
di: Zhu, Xuekai, et al.
Pubblicazione: (2024)

Distillation Traps and Guards: A Calibration Knob for LLM Distillability
di: Zhan, Weixiao, et al.
Pubblicazione: (2026)

Loop Corrections to the Training Error and Generalization Gap of Random Feature Models
di: Kim, Taeyoung
Pubblicazione: (2026)