Machine learning models are widely used in scientific research, enabling estimates and analyses that were once unattainable. However, these models often struggle with slow convergence and unreliable results when dealing with complex interactions. To address this, Dr. Xiaowu Dai, an assistant professor of Statistics and Data Science and of Biostatistics at UCLA, has developed a new statistical technique to improve the accuracy and reliability of machine learning estimates.

Dai’s method, called nonparametric estimation via partial derivatives, uses gradient information—either observed or estimated—to accelerate the convergence of machine learning models in complex, high-dimensional studies. This approach demonstrates that nonparametric estimation can achieve near-parametric rates when gradient data is incorporated, allowing the central limit theorem to be applied for the inference of nonparametric functions. Dai’s work builds on a long-standing conjecture by Karlin (1969) and Wahba (1971), who showed that incorporating gradient information adds no value for exact, one-dimensional functions. However, this result does not hold for noisy data. Dai’s method addresses this gap, improving accuracy in noisy, complex datasets.

This new technique has been applied to various questions in econometrics, queuing systems, and biological modeling, consistently outperforming traditional methods. “By incorporating gradient data, we can accelerate convergence and improve estimate accuracy, addressing a major limitation of conventional models that often require large datasets for reliable results,” says Professor Dai. “We believe this method will be a useful tool for practitioners in data-driven fields, ranging from science and social science to engineering.”

This work has been published in the prestigious Journal of the Royal Statistical Society Series B and can be accessed here. Here is another link to the UCLA Division of Physical Sciences article that discusses this work.

The Conference on Information Processing and Management of Uncertainty (IPMU 2024) has awarded Professor Judea Pearl, a joint faculty member in our department, the Kampe de Feriet Award for “his seminal work on probabilistic reasoning in artificial intelligence, including Bayesian networks and causality”. This prestigious award recognizes significant contributions to the field of information processing and the management of uncertainty. For more details, please visit this link.

Please join us in congratulating Professor Pearl!

The management of uncertainty was a hot AI issue in the 1970’s. Please see the following article by Professor Pearl on this topic: “A Personal Journey Into Bayesian Networks“.

Karen McKinnon is a faculty member in our department and is also an affiliated faculty member in the UCLA Department of Atmospheric and Oceanic Sciences. Her work focuses on climate variation and climate change. Please congratulate Karen for her promotion!