Abstract: In standard neuro-evolution, a population of networks is evolved in a task, and the network that best solves the task is found. This network is then fixed and used to solve future instances of the problem. Networks evolved in this way do not handle real-time interaction very well. It is hard to evolve a solution ahead of time that can cope effectively with all the possible environments that might arise in the future and with all the possible ways someone may interact with it. This paper proposes evolving feedforward neural networks online to create agents that improve their performance through real-time interaction. This approach is demonstrated in a game world where neural-networkcontrolled individuals play against humans. Through evolution, these individuals learn to react to varying opponents while appropriately taking into account conflicting goals. After initial evaluation offline, the population is allowed to evolve online, and its performance improves considerably. The population not only adapts to novel situations brought about by changing strategies in the opponent and the game layout, but it also improves its performance in situations that it has already seen in offline training. This paper will describe an implementation of online evolution and shows that it is a practical method that exceeds the performance of offline evolution alone.

Abstract: CC4 neural network is a new type of corner classification training algorithm for three-layered feedforward neural networks. On the condition that documents are almost of the same size, CC4 neural network is an efficient document classification algorithm. It would be impractical however to assume that all documents on WWW is of the same size in reality. To solve the problem incurred by the great difference in document sizes, we propose an MDS-NN based data indexing method thus making all documents be mapped to k-dimensional points while their distance information is kept well. We also extend CC4 neural network so that it can accept k-dimensional indexes of documents as its input, then transforms these indexes to binary sequences required by CC4 neural network. Our experimental results show that the performance of our method is much better than that of original CC4.

Abstract: This paper argues that multiagent learning is a potential \killer application" for generative and developmental sys- tems (GDS) because key challenges in learning to coordinate a team of agents are naturally addressed through indirect encodings and information reuse. For example, a signicant problem for multiagent learning is that policies learned sepa- rately for dierent agent roles may nevertheless need to share a basic skill set, forcing the learning algorithm to reinvent the wheel for each agent. GDS is a good match for this kind of problem because it specializes in ways to encode patterns of related yet varying motifs. In this paper, to establish the promise of this capability, the Hypercube-based NeuroEvo- lution of Augmenting Topologies (HyperNEAT) generative approach to evolving neurocontrollers learns a set of coor- dinated policies encoded by a single genome representing a team of predator agents that work together to capture prey. Experimental results show that it is not only possible, but benecial to encode a heterogeneous team of agents with an indirect encoding. The main contribution is thus to open up a signicant new application domain for GDS.

Abstract: Time Series Forecasting is an extremely important problem that has received a lot of attention from many different quarters of science and engineering because of its intrinsic difficulties and practical applications. We present a Neuro-Evolutionary approach to time series forecasting, wherein we evolve neural network models to do time series prediction using an advanced evolutionary computation algorithm, called Neuro-Evolution using Augmenting Topologies. Changes in the original algorithm are proposed and to reach the high level of precision desired by time series forecasting, a novel ensembling method is also proposed and implemented. The algorithms ability to evolve networks for a chaotic time series prediction task is tested by applying it to a benchmark series, the Mackey-Glass Time Series, and finally to test the proposed ensembling technique, the system is applied to predicting a complex currency exchange rate time series, that of the Yen-$ daily exchange rates. The results suggest that the system is robust, and when combined with the proposed ensembling technique, it is able to beat the predictions of the Random Walk model in the currency exchange rate prediction task. In the Mackey Glass test, the system successfully demonstrates its ability to evolve minimal structures to solve complex problems and multiple runs on the task confirm its ability to create diversity in terms of the architectures of the neural networks it produces.