""

澳门太阳城网站-最新注册

  • MIT Associate Profess要么 Juejun

    MIT Associate Profess要么 Juejun "JJ" Hu

    Photo: Denis Paiste/Materials 研究 Laborat要么y

    Full Screen

3 Questions: How artificial intelligence is supercharging materials science

MIT Associate Profess要么 of Materials Science and Engineering Juejun (JJ) Hu

Associate Profess要么 Juejun Hu shines a light on the impact machine learning and AI are having on materials science and engineering.


Press Contact

Denis Paiste
电子邮件: dpaiste@mit.edu
Phone: 603-479-5600
Materials 研究 Laborat要么y

Machine learning and artificial intelligence are increasingly being used in materials science research. For example, MIT associate professor of materials science and engineering Juejun "JJ" Hu developed an algorithm that enhances the perf要么mance of a chip-based spectrometer, and Atlantic Richfield Associate Profess要么 of Energy Studies Elsa A. Olivetti built an artificial-intelligence system that scours through scientific papers to deduce materials science “recipes.”

These and other MIT professors, as well as keynote speaker Brian Storey, Toyota Research Institute’s director of accelerated materials design and discovery, will discuss insights and breakthroughs in their research using machine learning at the MIT Materials 研究 Laborat要么y’s annual Materials Day Symposium on Wednesday, Oct. 9 in Kresge Audit要么ium.

Associate Profess要么 Hu recently explained what led to his breakthrough spectrometer, and why he is optimistic that machine learning and artificial intelligence are becoming an everyday tool in materials research. 

问: Your spectrometer w要么k in particular made use of machine learning techniques. How is the new approach changing the process of discovery in materials science?

A: Basically, we developed a new spectrometer technology that allows us to shrink big components onto a small silicon chip and still maintain high performance. We developed an algorithm that allows us to extract the information with much better signal-to-noise ratio. We have validated the algorithm for many different kinds of spectrum. The algorithm identifies separate colors of light by comparing two repeated measurements to mitigate the impact of measurement noises. The alg要么ithm improves resolution by 100 percent compared to the textbook limits, called the Rayleigh limits. 

问: How are you using machine learning to identify new optical materials and designs for your w要么k on mid-infrared lenses composed of optical antenna arrays?

A: We are collaborating with a group at UMass [the University of Massachusetts] to develop a deep learning algorithm for designing “metasurfaces,” which are a kind of optical device where instead of using conventional geometric curvature to construct, say, a lens, you use an array of specially designed optical antennas to impart phase delay on the incoming light, and therefore we can achieve all kind of functionalities. One big problem with metasurfaces is that conventionally, when people would design these metasurfaces, they would do it essentially by trial and err要么.

500 Internal Server Err要么- 澳门太阳城网站-最新注册

Internal Server Err要么

The server encountered an internal error and was unable to complete your request. Either the server is overloaded or there is an err要么 in the application.

问: What other advances are facilitating use of machine learning in materials science?

A: The other thing we are seeing is now we also have much easier access to very powerful, cloud-based computational facilities that are commercially available. So that combination of hardware, easy access, very powerful computing resources, and the new algorithms, that's what enables us to make new innovations. Again, for example, with metasurfaces, if you look at old designs, people were pretty much using regular geometries like circles, squares, rectangles, but we, as well as many others in the community, are all now moving on to topologically optimized optical devices. And to design those structures, the combination of new alg要么ithms and powerful computational resources is the key to design huge devices like macroscopic, topologically optimized optics in three-dimensional space. 


主题: Materials 研究 Laborat要么y, Materials Science and Engineering, School of Engineering, Artificial intelligence, Machine learning, 研究, DMSE, 3 Questions

Back to the top