Day 10 kicked off with research project lectures! The computer vision group worked as a group through a Jupyter notebook tutorial on Python Numpy, while the robotics group learned about how control algorithms and feedback loops can be applied to autonomous vehicles. Campers in the robotics group also tried to program their robots to follow a line. The NLP group discussed conditional probability, Bayes’ rule, and marginalization, then did a short problem set to practice what they learned. In the computational biology group, students finished learning about random hyperplane projection classification and learned LaTeX.
A guest lecture by Assistant Prof. Stefano Ermon followed the research project lectures. Prof. Ermon discussed how machine learning and computer vision can be applied to satellite data to determine the poverty levels of a specific region. This can be done by examining an area’s night time light, roads, tall buildings, houses, traffic, farmland, swimming pools, and more. Ermon talked about the challenges, purpose, and solution to the project, and introduced the idea of transfer learning, which is when knowledge learned from one task is used to solve a different but related task. Ermon also discussed how this can be applied to evaluating farmland for food security and productivity purposes.
The campers had a lunch outdoors with their research instructors, and returned to free time with their research mentors. Many research projects used this time to prepare for their banquet presentations tomorrow.
After independent work time, PhD. student Ajay Mandlekar presented about cloud crowdsourcing for task automation. Ajay’s project aimed to create a system for collecting task demonstrations as scale, so that robots could use the data for machine learning. The current problem is that there lacks a large amount of training data for robotics. This project tries to resolve this. The campers had the opportunity to test the system out, and some were able to control the robots from a phone.
The day concluded with independent work time.
Blog post and photos by Lauren Yang