Browsing by Author "Wiebel, Alexander"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Saliency Clouds: Visual Analysis of Point Cloud-oriented Deep Neural Networks in DeepRL for Particle Physics
    (The Eurographics Association, 2022) Mulawade, Raju Ningappa; Garth, Christoph; Wiebel, Alexander; Archambault, Daniel; Nabney, Ian; Peltonen, Jaakko
    We develop and describe saliency clouds, that is, visualization methods employing explainable AI methods to analyze and interpret deep reinforcement learning (DeepRL) agents working on point cloud-based data. The agent in our application case is tasked to track particles in high energy physics and is still under development. The point clouds contain properties of particle hits on layers of a detector as the input to reconstruct the trajectories of the particles. Through visualization of the influence of different points, their possible connections in an implicit graph, and other features on the decisions of the policy network of the DeepRL agent, we aim to explain the decision making of the agent in tracking particles and thus support its development. In particular, we adapt gradient-based saliency mapping methods to work on these point clouds. We show how the properties of the methods, which were developed for image data, translate to the structurally different point cloud data. Finally, we present visual representations of saliency clouds supporting visual analysis and interpretation of the RL agent's policy network.
  • Thumbnail Image
    Item
    Semi-Automatic Particle Tracking for and Visualization of Particle Detector Data
    (The Eurographics Association, 2022) Eschbach, Robin; Messerschmidt, Kai; Keidel, Ralf; Wiebel, Alexander; Bender, Jan; Botsch, Mario; Keim, Daniel A.
    In high energy physics, tracking particles in point-based particle detector data is important to reconstruct the particles' trajectories. As the numbers of particles can be in the hundreds or over a thousand, automatic algorithmic tracking is usually preferred over manual tracking. However, in some cases, manual tracking is needed as a baseline to assess the quality of the algorithmic tracking. Tracking particle locations manually is time-consuming and challenging when dealing with thousands of particles in the same data frame. In this paper, we describe Semi-Automatic Particle Tracking (SAPT), a collection of methods that aid manual particle tracking and visualization. These methods aim to make related particle hits easier to recognize by stretching the data and hiding likely irrelevant hits based on an angle criterion. They also help with finding the most likely track among a set of intuitively selected detector hits. These methods, together with a prediction of the most probable continuation of a track that can simply be accepted by the human user, accelerate the manual tracking process tremendously. We demonstrate the usefulness and efficiency of our methods by applying them to simulation data of a detector for proton computed tomography (pCT).