A naukri.com initiative
Mit
1d
157
Image Credit: Mit
3 Questions: Inverting the problem of design
- The DeCoDE Lab at MIT is combining machine learning and generative AI techniques, physical modeling, and engineering principles to tackle design challenges and enhance the creation of mechanical systems.
- Linkages, one of the group's recent projects, explores ways planar bars and joints can be connected to trace curved paths.
- The project uses the idea of self-supervised contrastive learning approaches to learn the representation of the design and how it works.
- This technique helps to incorporate precision into generative AI models and contributes to automated discovery.
- With the Linkages project, the team showed that the approach enabled solving problems more precisely and significantly faster, at 28 times less error and 20 times faster than prior state-of-the-art approaches.
- The Linkage method uses contrastive learning between the represented mechanism joints in graphs and another model that creates an embedding for the curves, and then connects these two modalities using contrastive learning.
- The method can find new mechanisms with precision through additional optimization steps beyond identifying candidate mechanisms.
- This proof of concept shows that the method is effective on discrete and continuous systems and can potentially transfer to many engineering applications.
- Future work for the project includes exploring more complex mechanical systems and more physics, as well as thinking about how precision in langue models can be incorporated.
- The project was supported by the MIT-IBM Watson AI Lab.
Read Full Article
9 Likes
For uninterrupted reading, download the app