Ross G
- Research Program Mentor
PhD candidate at University of California San Diego (UCSD)
Expertise
Artificial Intelligence, Computer Vision, Machine Learning, Autonomous Vehicles, Human-Robot (and Human-Computer) Interaction, Music, Intelligent Systems, Image Processing
Bio
I use computer vision & artificial intelligence to create human-aware systems for applications in intelligent vehicles, virtual classrooms, and even concert halls! At school, you'll find me spending my time in two labs: the Laboratory for Safe & Intelligent Vehicles (PI Mohan Trivedi) & the Center for Research in Entertainment & Learning (PI Shlomo Dubnov). Some problems I solve with my research are prediction of vehicle trajectories in real-world driving scenarios, estimation of driver state & readiness behind the wheel, and understanding & communicating human gaze & gesture during live musical performances. My work even teaches computers how to compose music using artificial intelligence, working towards human-AI "co-creativity"! I love to create music, and I feel lucky to combine two of my passions in part of my research & teaching. Outside of my academic work, I advise the Symphonic Student Association at UCSD, and field direct & write music for the Marching Band at UC Berkeley. One of my favorite hobbies is traveling to conducting workshops around the world! I'm really excited about projects that keep people safe & happy. If your creative ideas fall into this category, please reach out - I'd love to work with you!Project ideas
Computer Recognition of Directed Gestures
After spending a year in virtual classrooms, what are some of the small, nonverbal interactions you miss from in-person conversations? Eye contact & directed gestures are two modalities of communication we usually take for granted, but over Zoom, it can be very hard to tell to whom someone is waving or pointing. What if a computer vision system could help in determining & communicating this information? In this project, you can research methods in computer vision and machine learning that will allow for gesture recognition. Beginning from a standard desktop workstation, you can further experiment with your methods, extending to other poses and contexts.
Machine Learning for Autonomous Vehicles
There are many, many tasks involved in the realm of self-driving vehicles, which involve analysis of activities happening inside the vehicle, outside the vehicle, and even from a bird's-eye-view. Some projects you may want to explore: Driver activity analysis using inside-facing cameras, exploring a driver's eyes, attention, posture, hand activity, etc. Detecting pedestrians and other vehicles in the outside scene using machine learning. Path planning and trajectory prediction for intelligent agents. If your idea involves an intelligent vehicle, I'd love to connect and help guide your research!
Enhanced Transfer Learning for Convolutional Neural Networks
Transfer learning is a popular mechanism to improve performance and reduce training time of Convolutional Neural Network (CNN) models. However, when a pretrained model well-suited for a task operates on a different scale than the original training images, it may be beneficial to change the size of the convolutional kernel. But, this requires retraining the network, making transfer learning impossible. In this project, you will develop a method for initializing kernels that maintain the effects of learned weights while having flexibility to accommodate new kernel sizes in convolutional layers.
Guiding Young Musicians with Machine Learning
Many young musicians need a little help from their teacher in managing hand placement, tone quality, rhythmic or pitch correction, or other actions on their instrument. Using your camera and/or microphone, we can explore ways to use machine learning and artificial intelligence to identify mistakes and help suggest corrections. This would be a great area of exploration for students who have experience playing a musical instrument!
Win a Computer Vision or Machine Learning Challenge!
The premier conferences in the computer vision and machine learning fields often put out seasonal challenges, allowing researchers to develop a rapid solution (usually in just a few months) to an exciting problem. For advanced students who have time to devote towards one of these competitions, I would be excited to serve as your coach and mentor towards these projects. Please visit the CVPR website for an idea of past competitions!
Facial Emotions to Affect Vehicle Control
Some riders may be nervous to be in an autonomous vehicle, and it's important to make smooth, safe rides for these passengers. What if their car could understand how they might be feeling, and drive slower to accommodate their cautionary feelings?
Can computer vision beat a person at a card game?
Have you ever played a pattern-matching game like Spot-It? What if your opponent had really fast eyes & a brain (e.g. a computer), but a bit of uncertainty in its choices? Who would win? Can you engineer a system such that the computer consistently beats the human?
How can machine learning leverage multiple views of the same scene, with noise and sensor failures?
Sometimes, we add additional sensors to systems to make our predictions more accurate (for example, multiple views of a driver, to estimate whether they are alert). But, sometimes these views become blocked, or sensors temporarily fail. How can we use machine learning to work around these issues in a way that allows for consistently-available predictions?
Is AI-generated music created by an "intelligent" system? How can we evaluate our models?
In this project, we will develop a musical "Turing Test" to evaluate the brains of our musical machine learning systems.