Creation of Autonomous Artificial Intelligent Agents using Novelty Search method of fitness function optimization

For billions of years of evolution, biological intelligent agents have mastered the power to find optimal solutions in deceptive environments that we encounter in our daily interactions with the real world. We can easy navigate themselves through the maze of a big city subways and roadways. But for artificial intelligent systems, this is too difficult to be easily solved using the optimal computing resources. This is especially true for offline autonomous agents that are not backed by super power of cloud servers.

Self Replication to Preserve Innate Learned Structures in Artificial Neural Networks

It’s interesting to investigate combination of deep neuro-evolution and self-replication1 to evolve Artificial Neural Networks (ANNs) able to keep and complexify innate learned structures aimed to fulfill auxiliary tasks (orthogonal to the self-replication). In such a way, it may became possible to build evolutionary lineage tree of ANNs specialized to complete specific tasks under different environmental settings through subsequent training sessions. And the knowledge acquired during all these training sessions will accumulate not only in the form of learned connections’ weights, but in a neural network’s topology as well.

Neuroevolution - Evolving Artificial Neural Networks Topology From the Scratch

The most popular method of Artificial Neural Networks (ANN) training - at the time of this essay writing - is to use some form of Gradient Descent (GD) combined with error back propagation w.r.t. objective function defining our learning goal. This methodology was invented about 30 years ago by Geoffrey Hinton and become a foundation of all modern research activities in the Deep Machine Learning (ML) and the Artificial Intelligence (AI).