DeepSense
Deep Sense architecture partitions the input time series based upon their timestamps.
- processes each partition using a multi-layered convolutional network to generate a high level representation of each partition which contains all the intra partition dependencies
- uses a Recurrent Neural Network (Stacked GRU in this case) to combine the high level representations of each partition obtained from the convolutional network into a single vector (the final internal state of the Recurrent Network) The final internal state of the Recurrent Network is used as the representation of the current state of the agent.
