Neural Search is a new way to incorporate Monte Carlo Tree Search into neural networks. Unlike similar techniques, Neural Search builds the exploration and optimization of the search into the layers of the network. The search itself is learnable, and provides us with a means to develop guided search algorithms trained on data. We have demonstrated that the algorithm can learn robust solutions to search problems. We propose further development of the approach to large scale problems, and to demonstrate the effectiveness of the algorithm by training it to play Go, a standard achieved AlphaGo.
Total for the project is $150K over 11 months, and it will be resourced by 3 senior MLabs members.
