Wednesday, June 19, 2019 - 4:15pm - 5:05pm
Moritz Hardt (University of California, Berkeley)
We clarify what fairness guarantees we can and cannot expect to follow from unconstrained machine learning. Specifically, we characterize when unconstrained learning on its own implies group calibration, that is, the outcome variable is conditionally independent of group membership given the score. We show that under reasonable conditions, the deviation from satisfying group calibration is upper bounded by the excess risk of the learned score relative to the Bayes optimal score function. A lower bound confirms the optimality of our upper bound.
Monday, December 3, 2018 - 9:00am - 10:00am
Kaveh Khodjasteh (Target Corporation)
Target is a unique retailer with a large and complex supply chain network supporting a diverse set of SKUs that it offers in store across the US and online. Making this network efficient entails solving multiple interconnected optimization problems using techniques in stochastic modeling and optimization, algorithm development and scaling, and artificial intelligence. This talk will give an overview of our approach as well as some examples in inventory management optimization, space planning and transportation, and software design using functional programming.
Thursday, October 4, 2018 - 11:00am - 11:45am
Adam Elmachtoub (Columbia University)
Many real-world analytics problems involve two significant challenges: prediction and optimization. Due to the typically complex nature of each challenge, the standard paradigm is to predict, then optimize. By and large, machine learning tools are intended to minimize prediction error and do not account for how the predictions will be used in a downstream optimization problem.
Tuesday, February 14, 2017 - 9:00am - 9:50am
Joyce Mclaughlin (Rensselaer Polytechnic Institute)
In biomechanical Imaging of tissue and in geophysics, viscoelastic models are used in order for the mathematical models: (1) to accurately predict the data; and (2) given the data, to enable the imaging functional to accurately compute biomechanical properties of tissue or physical properties of the earth. The mathematical structure of these integro-differential operators, in the time/space domain, have new properties.
Thursday, June 9, 2016 - 3:15pm - 4:15pm
Andrew Barker (Lawrence Livermore National Laboratory)
Optimization of controls and parameters coming from realistic full-scale simulation requires enormous computational effort. To make such optimization practical requires optimal multilevel solvers and scalable parallel algorithms. Even in the case where such solvers and algorithms are well understood for the forward problem, adapting them to the optimization context can be interesting and complicated.
Thursday, June 9, 2016 - 2:00pm - 3:00pm
Thomas Grandine (The Boeing Company)
PDE-constrained optimization is an essential technology for product development at The Boeing Company. This talk will survey three separate applications of increasing mathematical and computational complexity. The first application is a relatively straightforward parametric surface lofting application. The second application is structural analysis. In practice, there are at least three ways in which PDE-constrained optimization can be carried out, and these three approaches will all be reviewed.
Wednesday, June 8, 2016 - 11:15am - 12:15pm
Sven Leyffer (Argonne National Laboratory)
Many complex applications can be formulated as optimization problems constrained by partial differential equations (PDEs) with integer decision variables. Examples include the remediation of contaminated sites and the maximization of oil recovery; the design of next generation solar cells; the layout design of wind-farms; the design and control of gas networks; disaster recovery; and topology optimization.
Thursday, March 17, 2016 - 10:30am - 11:00am
Carlos Rautenberg (Humboldt-Universität)
We address the problem of optimally placing sensor networks for convection-diffusion processes where the convective part is perturbed. The problem is formulated as an optimal control problem where the integral Riccati equation is a constraint and the design variables are sensor locations. The objective functional involves a term associated to the trace of the solution to the Riccati equation and a term given by an additional constrained optimization problem subject to a set of admissible perturbations.
Tuesday, March 15, 2016 - 10:30am - 11:00am
Chunming Wang (University of Southern California)
Assimilation of observation data in meteorology and space weather consists of using these data to estimate the current state and the spatially and temporally distributed parameters of Numerical Weather Prediction (NWP) models, which are often fluid dynamical equations. The aim of the data assimilation is to provide wider monitoring of the weather condition beyond the locations where data are collected, also referred to as now-casting and to provide forecasting of weather conditions using NWP.
Thursday, November 5, 2015 - 11:00am - 12:00pm
J. William Helton (University of California, San Diego)
One of the main developments in optimization over the last 20 years is
Semi-Definite Programming. It treats problems which can be expressed as a
Linear Matrix Inequality (LMI). Any such problem is necessarily convex,
so the determining the scope and range of applicability comes down to the

How much more restricted are LMIs than Convex Matrix Inequalities?

The talk gives a survey of what is known on this issue and will be
accessible to about anybody.

There are several main branches of this pursuit.


Subscribe to RSS - Optimization