A series of miracles such as autonomous driving, speech recognition, and music creation are being created by deep learning. Today, scientists at the US Department of Energy's Oak Ridge National Laboratory (ORNL) have developed a new software that enables deep learning to further aid scientific discovery.
Interior of the Fermi National Accelerator Laboratory MiniBooNE Neutrino Detector
ORNL's team is working on GPU-accelerated Titan supercomputers and has developed an algorithm that automatically generates neural networks. These neural networks are freely constructed with connections in the human brain and are "learned" in a deep learning manner.
MENNDL (full name Multi-node Evolutionary Neural Networks for Deep Learning) is used to evaluate and test neural networks such as unique data sets collected by scientists and provide advice. With the acceleration of the GPU, the speed is very fast, and the work that took a few months to complete can be shortened to a few weeks.
“MENNDL saves people time and speeds up scientific discovery,†said Steven Young, research scientist at ORNL's Nature Inspired Machine Learning team.
AI for scientists
Although the ORNL team is a MENNDL created for scientists, the scope of AI may be expanded to a wider extent. By training neural networks, researchers created a software that performs specific tasks. ORNL's software itself can create networks, thus eliminating the trial and error correction processes that are typically required to configure the network.
The algorithm scales up on Titan's 18,688 NVIDIA Tesla GPUs while testing and training thousands of potential networks to predict a network that is more suitable for the job.
In many areas, researchers use existing neural networks or data sets as a platform for deep learning research. However, this approach does not apply to scientists because the data comes from scientific instruments and is very different from the data used to teach computers for facial recognition or speech recognition.
“The data we use in the lab is extracted from neutrino detectors, electron microscopes or some other scientific instrument,†Young said. “This is very different from the pictures of cats and dogs.â€
MINERvA Neutrino Detector at Fermi National Accelerator Laboratory
Reduce work cycle to 24 hours
MENNDL has accelerated the research speed of neutrino physics. Neutrinos are subatomic particles that scientists believe can uncover the mysteries of the origin and material nature of the universe.
Because neutrinos are extremely difficult to detect, the US Department of Energy's Fermi National Accelerator Laboratory (Fermilab) uses high-intensity beams to study how they interact with common materials. This will generate massive amounts of data that researchers must analyze to accurately identify where interactions occur.
Young pointed out that in the past, the Fermilab team had to spend months testing the neural network to find a network that would help solve the problem, and MENNDL only took 24 hours to complete.
“With MENNDL, scientists no longer have to spend months looking for a suitable deep learning framework to find a network that can help process data in just one day,†Young said.
As a result, researchers will be able to conduct more trials in less time and drive scientific advances faster.
Maskking is an electronic cigarette brand. Maskking vape is a popular brand of Disposable Pod systems and e-cigarettes that are designed to provide a convenient and easy-to-use alternative to traditional smoking. The company has a range of products that are known for their compact and sleek design, with a variety of flavors and nicotin strengths to choose from.
Maskking Vape,Maskking Vanilla Disposable Vape,Nicotine Free Disposable Vape,Maskking Vape Kit,Maskking Vape Pod
TSVAPE Wholesale/OEM/ODM , https://www.tsecigarette.com