convergence guarantees
Theoretical assurances that an algorithm will converge to a solution or optimal point within a defined number of iterations or under certain conditions. In AI, convergence guarantees provide confidence in model training and performance.
- 3BASiL: An Algorithmic Framework for Sparse plus Low-Rank Compression of LLMs
- A Difference-of-Convex Functions Approach to Energy-Based Iterative Reasoning
- Absorb and Converge: Provable Convergence Guarantee for Absorbing Discrete Diffusion Models
- Adaptive Algorithms with Sharp Convergence Rates for Stochastic Hierarchical Optimization
- AltLoRA: Towards Better Gradient Approximation in Low-Rank Adaptation with Alternating Projections
- Asymptotically exact variational flows via involutive MCMC kernels
- Breaking AR’s Sampling Bottleneck: Provable Acceleration via Diffusion Language Models
- ComPO: Preference Alignment via Comparison Oracles
- Discrete Diffusion Models: Novel Analysis and New Sampler Guarantees
- Efficient Quadratic Corrections for Frank-Wolfe Algorithms
- Efficiently Escaping Saddle Points under Generalized Smoothness via Self-Bounding Regularity
- Error Feedback under $(L_0,L_1)$-Smoothness: Normalization and Momentum
- Extragradient Method for $(L_0, L_1)$-Lipschitz Root-finding Problems
- FedFACT: A Provable Framework for Controllable Group-Fairness Calibration in Federated Learning
- Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning
- Functional data analysis for multivariate distributions through Wasserstein slicing
- Generating Informative Samples for Risk-Averse Fine-Tuning of Downstream Tasks
- GeoClip: Geometry-Aware Clipping for Differentially Private SGD
- Globally Optimal Policy Gradient Algorithms for Reinforcement Learning with PID Control Policies
- Iterative Self-Incentivization Empowers Large Language Models as Agentic Searchers
- Learning quadratic neural networks in high dimensions: SGD dynamics and scaling laws
- Leveraging semantic similarity for experimentation with AI-generated treatments
- MGUP: A Momentum-Gradient Alignment Update Policy for Stochastic Optimization
- NPN: Non-Linear Projections of the Null-Space for Imaging Inverse Problems
- Natural Gradient VI: Guarantees for Non-Conjugate Models
- Nonlinearly Preconditioned Gradient Methods: Momentum and Stochastic Analysis
- Personalized Bayesian Federated Learning with Wasserstein Barycenter Aggregation
- Progress Reward Model for Reinforcement Learning via Large Language Models
- Risk-Averse Total-Reward Reinforcement Learning
- Transformers Learn Faster with Semantic Focus
- Understanding outer learning rates in Local SGD