Evolutionary Algo and ANN Implementation in AWS

Project entails the following: 1) Design and implement an evolutionary algorithm framework in Python that will deploy across AWS cloud to perform tuning optimization and feature selection for an ANN. This can be genetic algo, firefly, particle swarm, etc., with current preference for firefly. 2) Construct a customized ANN (or set of ANNs) as most of the pre-existing toolkits do not allow the full suite of parameter tuning (e.g., MATLAB GRNN does not allow for multiple, optimized kernel bandwidths). ANN architectures we have primarily used in the past are feedforward mlp's and grnn's. 3) Both the ANN and evolutionary algos must be parallelized and set up to run on AWS. This will include handling all the EC2 admin such as requesting spot instances, shutting down instances, etc. We are working on some very interesting problems across finance, media/entertainment, marketing, and other more proprietary areas. This is not a high-frequency trading project. Datasets are often high dimensional and sparse with a wide range of feature types. An understanding of time series modeling will be helpful as will experience with predictive analytics applied to social media data. More detail will be provided on project and datasets once an NDA has been executed. Alternative implementation/architecture suggestions are welcome but the use of a distributed computing implementation is required. We have considerable experience with the core datasets and simulation times in a single server environment is one of our key bottlenecks that we hope to overcome through this project.