gradient-based methods
Gradient-based methods are optimization algorithms that utilize gradient information to iteratively adjust model parameters, with techniques such as gradient descent being commonly used in training deep learning models.
- A Single-Loop Gradient Algorithm for Pessimistic Bilevel Optimization via Smooth Approximation
- Better NTK Conditioning: A Free Lunch from (ReLU) Nonlinear Activation in Wide Neural Networks
- Better Training Data Attribution via Better Inverse Hessian-Vector Products
- Final-Model-Only Data Attribution with a Unifying View of Gradient-Based Methods
- Generalization Bounds for Rank-sparse Neural Networks
- Identifying multi-compartment Hodgkin-Huxley models with high-density extracellular voltage recordings
- LARGO: Latent Adversarial Reflection through Gradient Optimization for Jailbreaking LLMs
- Learning Individual Behavior in Agent-Based Models with Graph Diffusion Networks
- Learning Theory for Kernel Bilevel Optimization
- Minimizing False-Positive Attributions in Explanations of Non-Linear Models
- Sharper Convergence Rates for Nonconvex Optimisation via Reduction Mappings
- Stable Coresets via Posterior Sampling: Aligning Induced and Full Loss Landscapes
- What is Your Data Worth to GPT? LLM-Scale Data Valuation with Influence Functions