training distribution
The distribution from which training examples are drawn, which defines the conditions under which the model learns. Understanding the training distribution is vital for evaluating how well the model may generalize to unseen data.
- Generalizable Reasoning through Compositional Energy Minimization
- Improving Generalization of Neural Combinatorial Optimization for Vehicle Routing Problems via Test-Time Projection Learning
- Improving Progressive Generation with Decomposable Flow Matching
- Just One Layer Norm Guarantees Stable Extrapolation
- Latent Policy Barrier: Learning Robust Visuomotor Policies by Staying In-Distribution
- Quantifying Uncertainty in the Presence of Distribution Shifts
- RUAGO: Effective and Practical Retain-Free Unlearning via Adversarial Attack and OOD Generator
- Scaling can lead to compositional generalization