model collapse
A scenario where a model, especially in generative contexts, begins to produce homogeneous and low-diversity outputs, losing its ability to capture the variability in the training data.
- A Closer Look at Model Collapse: From a Generalization-to-Memorization Perspective
- A solvable model of learning generative diffusion: theory and insights
- Balanced Conic Rectified Flow
- BitMark: Watermarking Bitwise Autoregressive Image Generative Models
- Hippocampal-like Sequential Editing for Continual Knowledge Updates in Large Language Models
- ReservoirTTA: Prolonged Test-time Adaptation for Evolving and Recurring Domains
- Self-Verification Provably Prevents Model Collapse in Recursive Synthetic Training
- Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning
- When Models Don’t Collapse: On the Consistency of Iterative MLE