The machine exposes quantum atomic vibration in learning material

Calcutic scientists have developed an artificial intelligence (AI) method that dramatically accelerates the quantum interactions that occur in content. In the new work, the group focuses on the interaction between nuclear vibration, or the phones – which rule the negotiating material features, including heat transport, thermal expansion, and phase transfer. The new approach to machine learning can be enhanced by counting all quantum interaction, which potentially enables encyclopedic knowledge to deal with particles and enthusiasm.

Scientists like Marco Bernardi, Applied Physics, Physics, and Materials Science in Caltek, and its graduate student Yao Lou (MS), are trying to find ways to accelerate the need to accelerate the large calculations needed to understand such particle interaction in real substances.

Last year, Bernardi and Lou developed a technique -based data -based data -based data -based data -based data -based data -based data -based data -based data -based data -based data -based data -based data -based data -based data -based data -based data -based data -based data -driven data -based data -driven data -based data -driven data -based data.

The issue of Phonen talks is even more complicated. These interactions are encoded in multi -dimensional items that are commonplace in the general status of tensors, vectors and high dimensions. The complexity of these tensors increases rapidly with the number of particles involved, which limit the understanding of scientists’ interaction with three or more phones.

Now, inspired by the recent developments in the machine learning, Bernardi and Lou have developed an AI -based technique that encodes the phone’s conversation in a substance through high -order tensors and removes the only important bits necessary to complete the thermal transport explaining the calculation. They describe the task in an article that appears in the journal Physical review posts.

Using existing sophisticated techniques, a supercomputer takes hours or days to calculate an interaction between three or four phones in a content. The new procedure enables computers to complete the same thermal transport and phone dynamics calculations at 1,000 to 10,000 times faster, while maintaining accuracy.

Bernardi says, “Counting four Four Four talks is a nightmare.” This work, this work, will include weeks of calculations. Now we can do them in 10 seconds. “

Bernardi explained more about this procedure:

“We use a machine learning technique named Candcump/Parafac tanker rotten, but we had to adopt it to meet the balance of this particular physical problem. We first set up a nervous network and then run it on the GPU and ask: ‘Who are the best things to do with the original tension?

“Once we decide the number of products that we want to keep, the machine learning process returns the best tasks to estimate the full tensor. We usually need some of these products, which saves orders to expand in computational complexity with full tensor.” This procedure allows the phonon to use the same thing of bilateral contacts.

Bernardi added that the new method is suitable for the thermal physics and the high -thropped screening of thermal physics and heat transport in the large material database, which is a great effort in the content community. As is a matter of future work, he says, “Right now, my vision is to compress all the different types of quantum interaction and high order process in similar techniques. The key to the formation of large tensors must be completely ignored and direct interaction in the compressed form.”

The title of this article is “Compression of Tancer Learning and Ann Following Interaction”. Additional authors are Dhu Mangatani, who worked on the project as a surf student in Bernardi’s lab. Shiv Peng, a Post Documentary Scholar Research Associate. And Calcutic graduating students Jia Yao (MS) and Sergey Kalyanic.