On Thursday, Oct. 15th,  Alaleh Maskooki will give a talk starting at 10:15 at lecture hall Arcanum 2 as part of the Joint Applied Mathematics and Statistics Seminars.

Title:

Application of Mathematical Programming in Pattern Recognition

Abstract:

Recent advances in digital data-storage technologies brought about massive data sets. The principal task in pattern recognition is to extract hidden relationships between data, and summarize them innovatively, to be understandable and thus practical for data owners. Recent literatures indicate lower error rate of mathematical programming approaches for multiclass classification and clustering on pattern with various characteristics, comparing to other common methods. This talk introduces two learning approaches based on integer linear programming (ILP) formulation. However, ILP problems are generally NP-complete. On the other hand, real-world patterns are often large-size datasets leading to large-scale ILPs that can be extremely time-consuming. Thus, efficient implementations are essential, along with keeping the accuracy. Based on this, the discussion will end with some suggestions as directions for improvement of the original methods.

All interested are warmly welcome! The talk is primary of optimization and ILP areas but it also contains some data classification and clustering tehnique ideas. So, it could also be interesting for statistians.

On Thursday, Oct. 15th,  Alaleh Maskooki will give a talk starting at 10:15 at lecture hall Arcanum 2 as part of the Joint Applied Mathematics and Statistics Seminars.

Title:

Application of Mathematical Programming in Pattern Recognition

Abstract:

Recent advances in digital data-storage technologies brought about massive data sets. The principal task in pattern recognition is to extract hidden relationships between data, and summarize them innovatively, to be understandable and thus practical for data owners. Recent literatures indicate lower error rate of mathematical programming approaches for multiclass classification and clustering on pattern with various characteristics, comparing to other common methods. This talk introduces two learning approaches based on integer linear programming (ILP) formulation. However, ILP problems are generally NP-complete. On the other hand, real-world patterns are often large-size datasets leading to large-scale ILPs that can be extremely time-consuming. Thus, efficient implementations are essential, along with keeping the accuracy. Based on this, the discussion will end with some suggestions as directions for improvement of the original methods.

All interested are warmly welcome! The talk is primary of optimization and ILP areas but it also contains some data classification and clustering tehnique ideas. So, it could also be interesting for statistians.