Somayeh Sojoudi
University of California-Berkeley, CA·Engineering·Early-career
Many real-world problems come down to finding the best choice among a huge number of options. Examples include running an electric power grid at the lowest cost, or training a machine-learning model.
- Fields
Engineering
Computer Science
Mathematics & Statistics
- Topics
non-convex optimization
optimization theory
machine learning
- Grants
- 1 known grant, through Dec 2026
What they work on
- Analyze when key non-convex machine-learning problems have spurious solutions (points that look locally best but are not globally best), and how the amount of data and problem structure remove them.
- Convert arbitrary polynomial optimization problems into machine-learning problems to discover which structural properties guarantee that no spurious solutions exist, using the theory of global functions and kernel structure.
- Approximate polynomial optimization problems that do have spurious solutions with a sequence of spurious-minima-free non-convex problems in a higher-dimensional space.
- Develop software and run case studies applying these low-complexity local search methods to key optimization problems in power systems and machine learning.
More about the research
When the problem is "convex", meaning it has a single bowl-shaped valley, fast algorithms can reliably find the best answer. Most practical problems are non-convex. They have many valleys, and a search can get stuck in a poor one. According to the project, this difficulty is a major bottleneck in upgrading aging power grids and costs billions of dollars a year in the United States.
Why it matters
Faster, more reliable ways to solve hard optimization problems could help power grid operators run electricity networks more efficiently, reliably and sustainably. The project describes current limitations as costing billions of dollars a year in the United States. The same tools could also make it safer to use artificial intelligence in safety-critical systems, where getting stuck on a poor solution is unacceptable.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Weilin Li
CUNY City College, NY·Mathematics & Statistics
Construct objective functions for continuous parameter estimation whose informative minima lie close to the true parameters and have large basins of attraction, so standard local optimization methods can locate them efficiently.
Yuxin Chen
University of Pennsylvania, PA·Mathematics & Statistics
Characterize the fundamental limits on sample and computational complexity of reinforcement learning, meaning how much data and computing an agent needs to learn good decisions in an unknown environment.
Simon Foucart
Texas A&M University-College Station, TX·Mathematics & Statistics
Extend Optimal Recovery theory, which guarantees worst-case accuracy for learning functions from data, from functions with single real-valued outputs to functions with more complex outputs, using optimization theory.
The school
University of California-Berkeley: 46 labs on PI Pages, $1.1B a year on research, 825 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.