Highlights and recent results from BESIII

by Dr Gang LI (Institute of high energy physics)

Tuesday 28 Mar 2017, 12:30 → 13:30 US/Pacific

Madrone (SLAC)

A core problem in experimental high-energy physics is the correct categorization of particle interactions recorded in our detectors as signal or background. This characterization is commonly done by reconstructing high-level components such as clusters, tracks, showers, jets, and rings within the event topology recorded by the detector and summarizing the energy, directions, and shape of these objects with a small number of quantities. These are often then fed into multi-variate algorithms to produce a final selector. However, these techniques are highly sensitive to the quality of the reconstruction of the high-level features and our imagination in developing them. Recent advances in deep learning have created new classes of machine learning algorithms which are capable of learning features from raw data, avoiding the pitfalls associated with high level reconstruction. In particular, convolutional neural networks, which automatically learn features from raw data, have been very successful in both the computer vision and natural language processing communities. Due to the spatial segmentation of sampling calorimeters, many high energy experiments can take advantage of CNN technology to identify events. In particular, I will discuss the specific application to the NOvA neutrino detector. This algorithm, known as CVN (Convolutional Visual Network), identifies neutrino events based on their topology without the need for detailed event reconstruction and outperforms algorithms currently in use by the experiment.