Learning recurrent representations for hierarchical behavior modeling
This work addresses behavior modeling for domains like animal studies and handwriting, but it is incremental as it builds on existing neural network methods.
The paper tackled the problem of detecting action patterns from motion sequences and modeling sensory-motor relationships in animals, using a generative recurrent neural network with lateral connections; results showed improved action detection with unlabeled data, unsupervised learning of high-level phenomena like writer identity and fly gender, and realistic simulated motion trajectories.
We propose a framework for detecting action patterns from motion sequences and modeling the sensory-motor relationship of animals, using a generative recurrent neural network. The network has a discriminative part (classifying actions) and a generative part (predicting motion), whose recurrent cells are laterally connected, allowing higher levels of the network to represent high level phenomena. We test our framework on two types of data, fruit fly behavior and online handwriting. Our results show that 1) taking advantage of unlabeled sequences, by predicting future motion, significantly improves action detection performance when training labels are scarce, 2) the network learns to represent high level phenomena such as writer identity and fly gender, without supervision, and 3) simulated motion trajectories, generated by treating motion prediction as input to the network, look realistic and may be used to qualitatively evaluate whether the model has learnt generative control rules.