Evolutionary Algorithms
We explore the use of evolutionary methods, such as evolution strategies and genetic algorithms, for training AI models.
The goal is to overcome some of the limitations of backpropagation and gradient-based learning, and to study whether evolutionary optimization can support stronger adaptation, reasoning, and generalization.
Recent advances in the literature suggest that these methods can be applied effectively even to large language models.
Evolution Strategies
Black-box optimization methods for improving model behavior without relying on gradients.
Genetic Algorithms
Population-based search methods for exploring alternative solutions and training dynamics.
Beyond Backpropagation
Studying training methods that do not depend entirely on standard gradient-based pipelines.
LLM Training
Investigating whether evolutionary methods can contribute to training or adapting large language models.