In the Informatics for Design, Engineering And Learning (IDEAL) Lab, we study how to make machines that learn how to design and build other machines. To do this, we use Machine Learning, Artificial Intelligence, and Crowdsourcing to understand how large groups of people design things and how complex engineered systems work, so that we can use the data they produce to make them better.

Some of the fundamental scientific questions we study include: What are efficient and useful ways to computationally and mathematically represent designs? How do we combine physics-driven and data-driven models to design better products? What makes design collaboration between large groups of people work well or poorly? How can we use tools from applied mathematics (such as graph theory, category theory, and statistics) and computer science (such as complexity theory, submodular optimization, and artificial intelligence) to better understand how humans design?

Some past practical applications of our research include: a fully automated system for inferring what makes designs creative given human feedback; the world’s first polynomial time algorithm for diverse bi-partite b-matching; algorithms for exploring and optimizing high-dimensional design spaces (e.g., aircraft) that accelerate optimization by an order of magnitude or more; software for helping novices 3D print working mechanical devices; and network analyses of online collaborative design networks such as OpenIDEO.

Graduate and Post-Doctoral alumni from our lab have gone on to pursue productive research careers in Academia (MIT, American University), National Labs (NIH, NIST), and Industry (Siemens, Capital One, Amazon).

You can read more about our work by looking at our papers or by reproducing any of our open-source code.

Mark recently had to give a pre-recorded invited talk as part of a conference. We uploaded the video to YouTube in case others wanted to watch as well. Video is 18 minutes long and covers much of the work covered in Wei Chen’s dissertation, including the following papers:

  • http://ideal.umd.edu/papers/paper/aiaaj-beziergan
  • http://ideal.umd.edu/papers/paper/jmd-hgan
  • http://ideal.umd.edu/papers/paper/scitech-airfoil-opt
  • http://ideal.umd.edu/papers/paper/arxiv-bezier-gan
  • http://ideal.umd.edu/papers/paper/samo-aes
  • http://ideal.umd.edu/papers/paper/jmd-feasible-designs
  • http://ideal.umd.edu/papers/paper/jmd-design-manifolds
Posted 22 Jul 2020 by Mark

Our article entitled “Airfoil Design Parameterization and Optimization using Bézier Generative Adversarial Networks” was accepted for publication in the AIAA Journal. Check out the publications page for more information.

Posted 22 Jun 2020 by

Dr. Mark Fuge has received an NSF CAREER award to study new Machine Learning algorithms with the goal of enabling computers that learn how to design complex systems like new aircraft or vehicles. See the College of Engineering News Story for more details: https://enme.umd.edu/news/story/fuge-receives-nsf-career-award

Posted 29 Jan 2020 by Mark

Our article entitled “Synthesizing Designs with Inter-Part Dependencies using Hierarchical Generative Adversarial Networks” was accepted for publication in the ASME Journal of Mechanical Design. Check out the publications page for more information.

Posted 12 Jun 2019 by
Published 22 Jun 2020
Published 25 Feb 2020
Machine Learning for Engineering Design
Panchal, Fuge, Liu, Missoum, and Tucker JMD 2019
Published 09 Sep 2019
Algorithms for Optimal Diverse Matching
Ahmadi, Ahmed, Dickerson, Fuge, Khuller arXiv 2019
Published 07 Sep 2019
Published 22 Aug 2019
Published 12 Jul 2019
Published 12 Jun 2019
Measuring and Optimizing Design Variety using Herfindahl Index
Faez Ahmed, Sharath Kumar Ramachandran, Mark Fuge, Sam Hunter and Scarlett Miller ASME IDETC 2019
Published 26 May 2019