论文信息 - Evolving Neural Turing Machines

Evolving Neural Turing Machines

Instead of training a Neural Turing Machine (NTM) with gradient descent [1], in this work NTMs are trained through an evolutionary algorithm. Preliminary results suggest that this setup can greatly simplify the neural model, generalizes better, and does not require accessing the entire memory content at each time-step. We show preliminary results on a simple copy and T-Maze learning task.

Emil Juul Jacobsen | Rasmus Boll Greve | R. Greve | E. Jacobsen | Sebastian Risi

[1] Dario Floreano,et al. Evolutionary Advantages of Neuromodulated Plasticity in Dynamic, Reward-based Scenarios , 2008, ALIFE.

[2] Risto Miikkulainen,et al. Evolving Neural Networks through Augmenting Topologies , 2002, Evolutionary Computation.

[3] Alex Graves,et al. Neural Turing Machines , 2014, ArXiv.

[4] Jürgen Schmidhuber,et al. Deep learning in neural networks: An overview , 2014, Neural Networks.

[5] Wojciech Zaremba,et al. Reinforcement Learning Neural Turing Machines , 2015, ArXiv.

[6] Sergey Levine,et al. End-to-End Training of Deep Visuomotor Policies , 2015, J. Mach. Learn. Res..