We’ve discovered that evolution strategies (ES), an optimization technique that’s been known for decades, rivals the performance of standard reinforcement learning (RL) techniques on modern RL benchmarks (e.g. Atari/MuJoCo), while overcoming many of RL’s inconveniences.
Survival of the fittest.
Evolutionary Complexity (EPlex) Research Group at UCF. Our research focuses on abstracting the essential properties of natural evolution that made it possible to discover astronomically complex structures such as the human brain. Our work is in part an approach to artificial intelligence.
Neuroevolution: an alternative approach to neural networks. Works this way: generate random neural networks and test them to perform a specific task. Choose the best one and let it breed.. with some statistic variation and then test the children for the task... and soo on. Besides changing the weight of the connections... as neural networks do... you also can change the topology of the network.