22-23 AI @ CSE – CSE Annual Report

AI @ CSE

CSE’s AI-Powered Technologies You Should Know About

A social robot to help people with cognitive impairments. Self-driving vehicles for delivery and micro-transit. A chatbot that gives movie recommendations

All are the result of AI research from CSE professors. Laurel Riek, director of the Healthcare Robotics Lab, Henrik Christensen, director of the Contextual Robotics Institute, and Julian McAuley, who has funding from Netflix, are among the professors featured across UC San Diego for work that could lead to the next developments in the “AI revolution.”

Accelerating Decision-Making for Climate Scientists

Why is Texas experiencing extreme heat waves? And why does El Niño cause temperatures to rise across North America? CSE’s Rose Yu is seeking to use AI to help address such broad, weighty questions in climate science.

Thanks to a grant from the Department of Energy, Yu and colleagues have been working to develop new machine learning methods that can speed up these climate models, better predict the future, and improve understanding of climate extremes.

In This Era of AI, Will Everyone Be a Programmer?

ChatGPT and Large Language Models (LLMs) are viewed by many as a threat to the field of computer science education because they’re able to produce code on command. But for Leo Porter, a teaching professor with CSE, these are tools that might actually help bring programming to a wider, more inclusive and diverse audience.

In the retooled, foundational course, “Introduction to Programming and Computational Problem Solving 1” launched this fall, Porter is teaching a class of mostly first-year UC San Diego students with a curriculum that incorporates LLMs. He is also the coauthor of the new book “Learn AI-Assisted Python Programming with GitHub Copilot and ChatGPT.”

AI Methods Fast-Track 3D Object Generation and Robotics

It takes roughly one minute. That’s how fast One-2-3-45++, an innovative AI method for 3D object generation, can create a high-fidelity 3D object from a single RGB image.

CSE Associate Professor Hao Su and collaborators from other universities proposed One-2-3-45++ in new research, hoping to overcome the shortcomings of its predecessor, One-2-3-45.

Breakthrough in 3D Representation with NeRFs

CSE’s Ravi Ramamoorthi, director of the Center for Visual Computing, introduced Neural Radiance Fields (NeRF) at the European Conference on Computer Vision. This breakthrough allows users to create a 3D representation of the scene of interest from a few views, allowing for immersive experiences of viewing it from different directions and creating compelling fly-throughs. NeRFs have been used by Google Maps, Google Street View, Luma AI and other technologies.

Recently, Ramamoorthi and researchers presented an extension to real-time radiance fields for portraits at SIGGRAPH 23 that can enable view synthesis in real-time from a single portrait image captured with a conventional cell phone camera (Image credit: Trevithick et al. SIGGRAPH 23).

Blocking Toxic Chat

Despite remarkable advances that large language models (LLMs) have achieved in chatbots, maintaining a non-toxic user-AI interactive environment has become increasingly critical. However, previous efforts in toxicity detection have been mostly based on benchmarks derived from social media content, leaving the unique challenges inherent to real-world user-AI interactions insufficiently explored. But new work from CSE professor Jingbo Shang and his research team introduces ToxicChat, a novel benchmark based on real user queries from an open-source chatbot.

New, Generative AI Transforms Poetry into Music

AI shows its artistic side in a new algorithm created thanks in part to research from UC San Diego computer scientists. The new algorithm helps create music out of “sound poetry,” a practice that uses nonverbal sounds created by the human voice to inspire feeling.

CSE-affiliate professor Shlomo Dubnov collaborated with CSE Associate Professor Taylor Berg-Kirkpatrick and Ph.D. candidate Ke Chen to create a recent performance based on principles also used in popular text-to-image AI generators.