EVOLUTIONARY ARTIFICIAL INTELLIGENCE DEVELOPMENT
Our Evolutionary AI Solutions: Evolutionary Computation, Deep Learning, Neuroevolution, Surrogate Optimization, Meta-learning, Trustworthy AI
Our Evolutionary AI Solutions: Evolutionary Computation, Deep Learning, Neuroevolution, Surrogate Optimization, Meta-learning, Trustworthy AI
In contrast to deep learning, which focuses on modeling known behaviors, Evolutionary Computation creates new solutions, by reassembling, transforming, and transforming the population over and over again. Our latest research includes finding design strategies in architecture, business, and sports, encouraging AI to find creative and novel solutions, using multiple purposes, and creating AI that can explain its decisions in terms of rules.
Deep learning is another foundation for AI research with the Evolutionary AI team. The main focus of our recent research is to improve deep learning of operational properties and cost calculations, using multiple data sets by learning multiple functions and finding useful neural network building blocks.
Neuroevolution is a powerful way of combining evolution with deep learning: evolution is used to automatically develop deeper learning structures, namely topology, components, parameters, and weight limits of sensory networks. To put it another way, AI designs AI. Our latest work focuses on neural architecture research, improving the state of the art in several machine learning benchmarks.
The idea is to start building a domain model by e.g. In-depth Learning, then use the model as a substitute to improve interaction using Evolutionary Computation. We have used this development method e.g. agricultural growth recipes, animal behavior agents, and non-pharmacological interventions in the COVID-19 epidemic. In this way, it is possible to get smart and effective decision-making strategies safely and effectively.
Modern models of deep learning have many features that need to be adjusted by hand, a tedious process that requires specialized technology. Metalearning is a family of strategies that allow for structures, loss functions, hyperparameters, opening functions, and other automated functions, leading to more efficient models.
To trust the predictions and instructions of the AI system, it needs to show how certain it is, it needs to allow testing of other solutions, and in some cases, it needs to explain its behavior with clear rules. The LEAF platform incorporates technologies developed specifically for these three principles.