parameterization
Parameterization in AI models refers to the process of defining a model using parameters that can be adjusted during training. This includes selecting the appropriate structure and the number of parameters that can be learned to optimize performance.
- Discovering Important Experts for Mixture-of-Experts Models Pruning Through a Theoretical Perspective
- Higher-Order Learning with Graph Neural Networks via Hypergraph Encodings
- Improved Representation Steering for Language Models
- Mamba Modulation: On the Length Generalization of Mamba Models
- Optimization Inspired Few-Shot Adaptation for Large Language Models
- PINNs with Learnable Quadrature
- PoLAR: Polar-Decomposed Low-Rank Adapter Representation
- Self-Supervised Learning of Motion Concepts by Optimizing Counterfactuals
- Thumb on the Scale: Optimal Loss Weighting in Last Layer Retraining
- Uncovering the Spectral Bias in Diagonal State Space Models