Have a UROP opening you would like to submit?
Please fill out the form.
Deep Learning, Computer Vision and CUDA
6: Electrical Engineering and Computer Science
09/15/2020 for direct funding. Flexible for credit or volunteer
Qianli Liao: LQL@MIT.EDU
Project 1: Although humans intuitively understand/interpret the world in terms of discrete objects. State-of-the-art machine vision systems do not have a good representation of visual objects. In a series of recent work, we try to incorporate the knowledge of object into deep learning networks and proposed a class of models we call "object-oriented" deep networks (OONets). https://cbmm.mit.edu/publications/object-oriented-deep-learning https://dspace.mit.edu/handle/1721.1/113002 Students are encouraged to either: 1. Try variants of our model (already implemented in PyTorch) on a wide range of computer vision tasks (GANs, VAEs, etc.) 2. Implement our models with Tensorflow. 3. Accelerate our models with NVIDIA CUDA (we have basic implementations in CUDA but we can further speed them up). Project 2: Study state-of-the-art object detection algorithms. Apply our OONets from Project 1 to detection tasks. Project 3: My other ongoing projects include biologically-plausible learning without backpropagation and 3D geometric deep learning. If you have your own ideas, we can also discuss.
Good at Python, Matlab or CUDA C.