1 — 16:20 — From Algorithms to Applications: developing and deploying analytical and optimization models with FICO® Xpress Insight
In this talk we will start by presenting the latest developments in the Python interface for the Xpress Solver, outlining recent updates to modelling capabilities and discussing some best practice guidance for developing readable and yet efficient implementations of large-scale optimization models.
To facilitate the deployment and utilization of optimization or analytical solutions implemented in Python or Xpress Mosel, FICO® Xpress Insight provides a flexible environment for creating user interfaces with little to no programming effort. Its interface is designed for both technical and non-technical users interested in running several scenarios and comparing outcomes to understand trade-offs and sensitivities implicit to the optimization problem.
We will demonstrate how an Xpress Insight application can seamlessly be built from a standard Python or Mosel optimization model with intuitive drag and drop of front-end visualization components, including multi-scenario charts and tables. Furthermore, we will demo the latest advancements for interacting with the solver during an optimization run via the new Custom Progress Reporting functionalities.
2 — 16:50 — OMLT: Solving inverse problems over trained graph neural networks
OMLT, the Optimization and Machine Learning Toolkit, is a Python package that represents machine learning models (neural networks, gradient-boosted trees, and linear decision trees) within the Pyomo optimization environment. This presentation discusses our recent extension of OMLT to incorporate graph neural networks and its possible application to computer-aided molecular design.
3 — 17:20 — Democratization of High-performance Computing Approaches for Optimization
Large-scale optimization approaches often require tailored solutions and advanced computational implementations that are not possible with off-the-shelf computational tools. The implementation of high-performance computing (HPC) approaches requires significant expertise in low-level compiled programming languages and knowledge of complex, architecture specific parallelization frameworks. The success of high-level languages like Python and Julia, have helped "democratize" advanced HPC approaches, providing students and researchers with access to advanced, performant capabilities in high-level languages. This allows for much more rapid innovation, refinement, and adoption of HPC approaches. I will discuss recent software advances that support democratization of HPC and scientific computing solutions for challenging large-scale optimization problems.