Evolution Strategies
PulseAugur coverage of Evolution Strategies — every cluster mentioning Evolution Strategies across labs, papers, and developer communities, ranked by signal.
1 day(s) with sentiment data
-
EGGROLL method enhances LLM training with low-rank evolution strategies
Researchers have developed EGGROLL, a method to make evolution strategies more practical for large language models by using low-rank Gaussian products instead of dense weight perturbations. This approach, while computat…
-
Evolution Strategies vs GRPO: LLM post-training yields distinct model updates
A new arXiv paper compares Evolution Strategies (ES) and Group Relative Policy Optimization (GRPO) for post-training large language models. While both methods achieve comparable accuracy on single-task and sequential le…
-
Evolution Strategies show promise for continual AI control tasks
Researchers have explored the use of Evolution Strategies (ES) for continual control tasks, where AI agents must adapt to new challenges without losing previously acquired knowledge. Experiments on sequential MuJoCo loc…
-
New framework unifies black-box optimization methods, introduces hybrid algorithms
Researchers have developed a unified theoretical framework for black-box optimization (BBO) methods, including Evolution Strategies (ES), Consensus-Based Optimization (CBO), and Optimization via Integration (OVI). This …
-
New Quantized Evolution Strategies Enable Fine-Tuning of LLMs
Researchers have introduced Quantized Evolution Strategies (QES), a novel optimization paradigm designed for fine-tuning quantized large language models (LLMs) directly within their discrete parameter space. This method…
-
New Operator Calculus Unifies Optimization Method Convergence Theory
Researchers have developed a new operator calculus framework to unify the convergence analysis of various population-based optimization methods. This approach describes algorithms like evolution strategies and stochasti…
-
New model analyzes Evolution Strategies for machine learning optimization
Researchers have developed a novel model to analyze the fitness progress of Evolution Strategies (ES) in generic problems. This model simplifies the analysis by focusing on the fitness relationship between parent and of…
-
New method trains energy-efficient spiking neural networks faster
Researchers have developed EGGROLL, a novel gradient-free method for training Spiking Neural Networks (SNNs) that significantly reduces computational cost. This approach uses low-rank factorization of Evolution Strategi…
-
AE Studio uses Modal to train AI for math theorem proving
AE Studio, a consulting partner for Modal, has developed a workflow for training AI models to prove mathematical theorems using reinforcement learning. They compared two methods: Group Relative Policy Optimization (GRPO…
-
New theory analyzes evolution strategies for mixed-integer optimization
Researchers have developed a theoretical framework to analyze the convergence of evolution strategies (ES) when applied to mixed-integer optimization problems. They introduced two variants, (1+1)-LB-ES and (1+1)-LUB-ES,…
-
OpenAI finds evolution strategies rival reinforcement learning for AI training
OpenAI researchers have found that evolution strategies (ES), a decades-old optimization technique, can rival the performance of modern reinforcement learning (RL) methods on benchmarks like Atari and MuJoCo. ES offers …