Dr. Shaokai Yang
High Energy Physics (HEP) offers the most exciting experimental possibilities to connect theoretical insights to long-lasting puzzles in our current understanding of the properties and interactions of fundamental particles. Deep learning (DL) is principally concerned with building neural networks that can understand the physics world successfully and solve assignments requiring intelligence. My primary academic interests are developing DL technologies to design physics-informed neural network architectures and, reciprocally, to use these algorithms to drive improvements in the physics reach of HEP projects, e.g.,the HEP experiments, perturbation-theory calculations, and precision particle interactions simulation.
MY SKILLS
Physicists have become interested in the potential of deep learning for fundamental research, and theoretical physicists no exception. Machine learning, especially deep learning, has been used to calculate the Feynman loop integrals, estimate the uncertainty of the combined parton distribution function (PDF) from a single PDF, and learn the mapping between the new physics model’s parameter space and the experimental physical observables. I am interested in contributing to this researches based on my understanding of deep learning.
The beauty and clearness of the trained deep neural networks, which assert particle interactions category and energy, are obscured by two clouds. The first came into existence with the interactions and detector simulation and was dealt with Monte Carlo generators, such as Geant4. It involved the question, how could we accurately simulate the particle interactions? The second is the long-neglected uncertainty issue, which comes from different deep neural network architectures. I am committed to resolving these issues by developing physics-informed and physics-aware deep learning algorithms.
Theoretical Calculations
Community Building
Recent Researches
For these advances to happen, we need to build a high energy physics dedicateddeep learning research community, aimed to bridge the gap between the two disciplines. SinceNOvA released the first deep neural networks based particle identification (PID), the group quicklygrew to includes most of the neutrino experiments and LHC experiments. More dedicated trainingefforts in deep learning are emerging, including the Yandex machine learning school and Dark Ma-chines. I am actively involved in these groups and help to plan and organize events and conferences..
DL Algorithms