approximation error
Approximation error refers to the discrepancy between the predicted output of a model and the actual output it seeks to predict. It reflects the limitations in the model's capacity to accurately represent or learn the underlying data distribution.
- Capturing Individual Human Preferences with Reward Features
- Degrees of Freedom for Linear Attention: Distilling Softmax Attention with Optimal Feature Efficiency
- Effective Neural Approximations for Geometric Optimization Problems
- Efficient Large Language Model Inference with Neural Block Linearization
- F-Adapter: Frequency-Adaptive Parameter-Efficient Fine-Tuning in Scientific Machine Learning
- In-context Learning of Linear Dynamical Systems with Transformers: Approximation Bounds and Depth-separation
- On the Hardness of Approximating Distributions with Tractable Probabilistic Models
- Perturbation Bounds for Low-Rank Inverse Approximations under Noise
- Purifying Shampoo: Investigating Shampoo's Heuristics by Decomposing its Preconditioner
- Quantitative convergence of trained neural networks to Gaussian processes
- Scalable Policy-Based RL Algorithms for POMDPs
- Stability and Oracle Inequalities for Optimal Transport Maps between General Distributions
- What Does It Take to Build a Performant Selective Classifier?