Rabbi Dr. Ari Berman, President and Rosh Yeshiva | Yeshiva University
Rabbi Dr. Ari Berman, President and Rosh Yeshiva | Yeshiva University
David Sweet, an industry professor at the Katz School's Graduate Department of Computer Science and Engineering, has introduced a new method to enhance computer optimization of complex tasks. His paper, "Fast, Precise Thompson Sampling for Bayesian Optimization," was presented at NeurIPS, a prominent AI conference in December.
Sweet's research is linked to technologies that are part of daily life. For instance, streaming services use similar systems to recommend shows by testing various approaches and selecting the most effective one. Bayesian optimization aids computers in efficiently reducing the number of tests needed to identify optimal solutions.
"The fewer experiments required, the faster and more cost-effective the process becomes," Sweet stated. He emphasized its importance in fields like healthcare and quantitative trading where quick optimization can save lives or prevent financial losses.
Thompson Sampling (TS) is a technique used to choose the best options by predicting their success likelihood. While TS works well with simpler systems, it faces challenges with more complex ones. Sweet's team developed an advanced version called the Stagger Thompson Sampler (STS), which is both faster and more precise for handling larger challenges.
The method has practical applications across various sectors:
- Improved recommendations: Shopping apps and streaming services can quickly learn user preferences for better suggestions.
- Accelerated drug development: Scientists can test fewer configurations, expediting medication creation.
- Enhanced green technology: Electric car charging schedules can be optimized to reduce battery wear.
- Advanced AI: Models can learn faster, perform better, and consume less power.
Sweet's approach not only increases speed but also handles high-dimensional problems effectively. STS surpasses older methods in dealing with complex issues involving numerous variables.
"Even though you may never directly interact with Bayesian optimization or Thompson Sampling, these advancements are shaping the tools, apps and technologies we use every day," said Sweet.