3 UC Berkeley affiliates listed among MIT’s ‘35 Innovators Under 35′

Related Posts

Three UC Berkeley affiliates — electrical engineering and computer science, or EECS, assistant professor Alessandro Chiesa, EECS graduate student Chelsea Finn and research scientist John Schulman, who earned a doctorate from the UC Berkeley EECS department — were recognized in the Massachusett Institute of Technology, or MIT, Technology Review’s 2018 list of “35 Innovators Under 35.”

They will be recognized at the Emerging Technologies Conference, or EmTech, taking place from September 11 to 14 at MIT.

Chiesa is the co-founder of Zcash, a cryptocurrency launched four years ago that ensures absolute privacy in online transactions, unlike public blockchain systems.

Zcash, with a market cap of over a billion dollars, employs a zero-knowledge proof construction that protects users’ sensitive data and protects against identity theft.

“These constructions allow the network to maintain a secure ledger of balances without disclosing the parties or amounts involved,” Zcash’s website reads.

As a research scientist at the nonprofit AI research company OpenAI, Schulman has been developing reinforcement learning algorithms, a branch of machine learning that provides rewards when the robot does a certain action. Some of these algorithms include TRPO, or trust region policy optimization, and PPO, or proximal policy optimization, which are two of the most widely used algorithms used on simulated robots.

In order to test his machine-learning algorithms, Schulman used Sonic the Hedgehog, a video game that incorporates varying speeds and physics.

“I believe that the biggest missing piece in the reinforcement learning subfield of AI is the ability to learn quickly (and) I think Sonic is a good testbed,” Schulman said in an email.

Finn, another recipient, has developed meta-learning algorithms and applied them to robots in recent years. Currently, robots are being programmed with reactions to certain stimuli and are only able to complete tasks that are written in their code, but Finn’s algorithms enable robots to self-learn with minimal code.

Her research developed algorithms for self-supervised learning, which allows robots to learn new tasks on their own by implementing previous knowledge and observation.

“In general, machine learning, learning from scratch, is seen as a good thing from other people, but ultimately we don’t want the systems resetting their knowledge,” Finn said.

At the lab where Finn works, the Berkeley Artificial Intelligence Research Lab, the algorithms are tested based on the robots’ ability to sort wooden blocks, and later to set a table and arrange other objects.

Finn’s research is ultimately aimed at developing machines that can acquire skills just by observing, not from programming.

“I remember in high school reading about Sebastian Thrun on a similar list, and being inspired by his endeavors in self-driving cars,” Finn said. “In the same way that I was inspired by his pioneering work, I hope to inspire others.”

Contact Aarya Gupta and Nathan Chin at [email protected].