Session: Advanced Computational Solvers 1
Chair: Nai-Yuan Chiang
Cluster: Computational Software
Talk 1: Simultaneous approaches for estimation and optimization of hybrid dynamic systems
Speaker: Carl Laird
Abstract: Hybrid modeling is becoming increasingly popular in science and engineering applications where it benefits from first-principles domain knowledge and increasing availability of data to train and validate machine learning components. Notable recent methodologies include the use of neural networks within DAE (differential algebraic equation) systems, namely Neural ODEs and Universal Differential Equations (UDE), which are powerful tools that enable effective hybrid modeling of systems with complex dynamics. In this presentation, we demonstrate the integration of machine learning models with simultaneous approaches for estimation and optimization of hybrid dynamic systems. We investigate both algorithmic aspects for improving computational performance and integration with modeling languages.
Talk 2: Recent Progress in the Cardinal Optimizer
Speaker: Nils-Christian Kempke
Abstract: The Cardinal Optimizer (COPT) is a mathematical optimization solver for large-scale optimization problems. It includes high-performance solvers for LP, MIP and conic programming etc. In this talk, we present the recent advances in COPT, including our work on GPU-accelerated optimization. We highlight some key computational and theoretical improvements we have recently integrated and evaluate their impact on the performance of our solver. We provide an overview of the performance numbers of the latest COPT release for all supported problem classes.
Talk 3: Latest Developments and Improvements in the MindOpt Interior Point Optimizer
Speaker: Kuoling Huang
Abstract: We focus on practical improvements in the MindOpt interior point optimizer, with a particular emphasis on enhancements in the sparse solver. We discuss various algorithmic techniques, such as advanced Cholesky factorization and reordering strategies, that enhance numerical stability and efficiency. Key numerical results on large-scale optimization problems will be presented.
