|
Abstract:
|
Typical problems in engineering navigate a complex space of trade-offs between different optimization objective functions. A manufactured object, for example, must be lightweight, strong, aesthetic, and inexpensive---but it is impossible to engineer an object that extremizes any one of these criteria without forsaking another. Moreover, many practical engineering objective functions exhibit wide regions of near-optimal designs, especially after accounting for tolerances and uncertainty in measuring the optimization objective. In this talk, I will summarize efforts to navigate and parameterize the wide space of near-optimal points in multi-objective engineering problems, including links to probabilistic sampling. After explaining the many open challenges in this space, I will describe preliminary efforts to design Markov chain Monte Carlo-inspired algorithms for sampling from the space of near-Pareto optimal points in multi-objective optimization, with application to computational fabrication. [Joint work with Adriana Schulz, Wojciech Matusik, Liane Makatura, and Haisen Zhao.]
|
