weak-to-strong generalization
The transition in machine learning performance from being able to generalize well on easy or similar tasks (weak) to performing effectively on a wider range of more complex or differing tasks (strong).
- Disentangling Latent Shifts of In-Context Learning with Weak Supervision
- From Linear to Nonlinear: Provable Weak-to-Strong Generalization through Feature Learning
- On the Mechanisms of Weak-to-Strong Generalization: A Theoretical Perspective
- Robust SuperAlignment: Weak-to-Strong Robustness Generalization for Vision-Language Models
- Weak-to-Strong Generalization under Distribution Shifts