Neuroevolution
Neuroevolution enable the unavailable capabilities through traditional gradient-based approaches, such as learning neural network building blocks (e.g., activation functions), hyperparameters, architectures, and even the algorithms for learning themselves. Over the years, the methods for evolving neural networks have greatly advanced. A combination of neuroevolution and another neural network learning could be used for model biological adaptation. Neuroevolution is a powerful approach to solve reinforcement learning problems and is successfully used for tasks as diverse as modeling biological phenomena and others.
- Artificial Neural Networks (ANNs)
ANNs are the computing systems inspired by the biological neural networks that constitute animal brains. The use of ANNs is a widely-covered research topic. The complexity of control tasks often makes it difficult to design ANNs manually. Therefore, it is a common approach to use evolutionary algorithms (EAs) for this kind of problem. Not only the synaptic weights but also the structure of the neural network can be subject to neuroevolution.
A system that automated configuration and training of DNNs using EAs inspired by the fact that natural brains are the products of the evolutionary process. EAs are gaining momentum as a computationally feasible method for the automated optimization of DNNs. Analog genetic encoding (AGE) is an implicit method, which-sofar-has only been applied to some simple problems of neuroevolution.
Fig.1 Compositional pattern-producing networks and HyperNEAT. (Stanley, 2019)
Applications of Neuroevolution
Neuroevolution experiments are possibly designed on how behaviors respond to environmental pressure, including but not limited to foraging, pursuit, and evasion, hunting and herding, collaboration, and even communication. In addition, abstract evolutionary tendencies can also be investigated in neuroevolution. Finally, neuroevolution, especially for analyzing evolved neural systems and their function, can lead to insights into biological networks.
From a scientific point of view, artificial evolution can test hypotheses of brain development from many scales, such as genetic, developmental, learning, and behavioral phenomena. Neuroevolution provides efficient solutions to many hard problems. Within this context, the evolution of neural systems will generate the understanding of reinforcement-learning-like structures important for various biological research.
Fig.2 ANNs represented with analog genetic encoding. (Floreano, 2008)
With long-term expertise in neuroscience, Creative Biolabs is the ideal partner to provide a wide range of leading technologies and products for our customers worldwide. Our dedicated team of specialists has an in-depth understanding of neuroscience and will help you choose the right solution for your research requirements. Please do not hesitate to contact us for more detailed information.
References
- Stanley, K. O.; et al. Designing neural networks through neuroevolution. Nature Machine Intelligence. 2019, 1(1), 24-35.
- Floreano, D.; et al. Neuroevolution: from architectures to learning. Evolutionary intelligence. 2008, 1(1), 47-62.
