extrapolation
Extrapolation in AI refers to the method of predicting outcomes based on an extended understanding of a model beyond the range of training data. It is crucial for generalizing a model’s capability to unseen data or scenarios.
- Estimation of Treatment Effects in Extreme and Unobserved Data
- Extrapolation by Association: Length Generalization Transfer In Transformers
- FlexWorld: Progressively Expanding 3D Scenes for Flexible-View Exploration
- How Benchmark Prediction from Fewer Data Misses the Mark
- Implicit Generative Property Enhancer
- Just One Layer Norm Guarantees Stable Extrapolation
- LEDiT: Your Length-Extrapolatable Diffusion Transformer without Positional Encoding
- Normalized Attention Guidance: Universal Negative Guidance for Diffusion Models
- Scaling Laws for Optimal Data Mixtures
- Stable Port-Hamiltonian Neural Networks
- TimePerceiver: An Encoder-Decoder Framework for Generalized Time-Series Forecasting
- Transformers Provably Learn Chain-of-Thought Reasoning with Length Generalization