Deep Action Sequence Learning for Causal Shape Transformation

Deep learning (DL) became the method of choice in recent years for solving problems ranging from object recognition and speech recognition to robotic perception and human disease prediction. In this paper, we present a hybrid architecture of convolutional neural networks (CNN) and stacked autoencoders (SAE) to learn a sequence of actions that nonlinearly transforms an input shape or distribution into a target shape or distribution with the same support. While such a framework can be useful in a variety of problems such as robotic path planning, sequential decision-making in games and identifying material processing pathways to achieve desired microstructures, this paper focuses on controlling fluid deformations in a microfluidic channel by deliberately placing a sequence of pillars, which has a significant impact on manufacturing for biomedical and textile applications where highly targeted shapes are desired. We propose an architecture which simultaneously predicts the intermediate shape lying in the nonlinear transformation pathway between the undeformed and desired flow shape, then learns the causal action--the single pillar which results in the deformation of the flow--one at a time. The learning of stage-wise transformations provides deep insights into the physical flow deformation. Results show that under the current framework, our model is able to predict a sequence of pillars that reconstructs the flow shape which highly resembles the desired shape.