convergence speed
Convergence speed refers to the rate at which an optimization algorithm approaches its solution. Faster convergence leads to reduced training times and can make iterative methods more efficient.
- Adaptive Riemannian ADMM for Nonsmooth Optimization: Optimal Complexity without Smoothing
- Asymptotics of SGD in Sequence-Single Index Models and Single-Layer Attention Networks
- Convergence of the Gradient Flow for Shallow ReLU Networks on Weakly Interacting Data
- Curriculum Abductive Learning
- FedQS: Optimizing Gradient and Model Aggregation for Semi-Asynchronous Federated Learning
- Harmony in Divergence: Towards Fast, Accurate, and Memory-efficient Zeroth-order LLM Fine-tuning
- Imitation Learning with Temporal Logic Constraints
- Improving Progressive Generation with Decomposable Flow Matching
- Learning in Compact Spaces with Approximately Normalized Transformer
- Learning in Stackelberg Mean Field Games: A Non-Asymptotic Analysis
- SageAttention3: Microscaling FP4 Attention for Inference and An Exploration of 8-Bit Training
- Scaling Diffusion Transformers Efficiently via $\mu$P
- SharpZO: Hybrid Sharpness-Aware Vision Language Model Prompt Tuning via Forward-Only Passes
- Sinusoidal Initialization, Time for a New Start
- Sparse MeZO: Less Parameters for Better Performance in Zeroth-Order LLM Fine-Tuning
- Structured Reinforcement Learning for Combinatorial Decision-Making
- Tight Lower Bounds and Improved Convergence in Performative Prediction
- Transfer Faster, Price Smarter: Minimax Dynamic Pricing under Cross-Market Preference Shift