I’m working on deep reinforcement learning for training machine agents. We are attempting to borrow elements of human decision making and investigating its advantages when applied to reinforcement learning algorithms.
Ishu is researching on estimating the structure and motion from Videos. He will be using Deep Neural Networks to train a model that decomposes frame-to-frame pixel motion in terms of scene and object depth, camera motion and 3D object rotations and translations. The model when trained can be used to predict depth maps from the single monocular 2-d image and predict the pose and motion from two consecutive images of the video.
Santhoshini is studying approximating MAXDICUT in a streaming setting – They are working on finding a good approximation algorithm for finding Max DICUT of a directed graph with its edges coming in a streaming setting and they are working on lower bounds for the same