Project Details
Grant Program
Faculty-development competitive research grants program for 2023-2025
Project Description
The proposed research lies at the boundary between optimization and machine learning. Both fields attract much attention nowadays due to ubiquitous contributions to practical applications. However, despite their widespread use, many problems still remain unsolved. One of those challenges is tackling the ever-increasing amount of data and the associated high-dimensional mathematical models which require large computational costs to process and analyze. Machine learning algorithms which heavily rely on the data often end up with a task of optimizing high-dimensional functions. Optimizing these functions locally can be performed with a relatively high degree of success. However, in many practical situations, one requires to know the global optimum of the objective function. Current global optimization solvers are only successful in tackling problems in the order of tens of variables while size requirements in many real-life problems especially the ones that arise in machine learning are at least three orders of magnitude larger. Therefore, there is a need to improve the scalability of the current global optimization algorithms.
Project Impact
The outcome of this project may lead to a better understanding of a particular strand in the field of optimization, result in a more scalable global optimization algorithms for a particular class of functions and, thus, may bring about an incremental change in the advancement of technology globally. However, we do not see any specific and direct socio-economic impact on Kazakhstan.
Status | Active |
---|---|
Effective start/end date | 1/1/23 → 12/31/25 |
Keywords
- global optimization
- dimensionality reduction
- random embeddings
- low effective dimensionality
- active subspaces
- random matrices
- linear programming
Fingerprint
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.