Dennis Noll

Abstract The development of an LHC physics analysis involves numerous investigations that require the repeated processing of terabytes of data. Thus, a rapid completion of each of these analysis cycles is central to mastering the science project. We present a solution to efficiently handle and accelerate physics analyses on small-size institute clusters. Our solution uses three key concepts: vectorized processing of collision events, the “MapReduce” paradigm for scaling out on computing clusters, and efficiently utilized SSD caching to reduce latencies in IO operations. This work focuses on the latter key concept, its underlying mechanism, and its implementation. Using simulations from a Higgs pair production physics analysis as an example, we achieve an improvement factor of 6.3 in the runtime for reading all input data after one cycle and even an overall speedup of a factor of 14.9 after 10 cycles, reducing the runtime from hours to minutes.

Abstract Variable-dependent scale factors are commonly used in HEP to improve shape agreement of data and simulation. The choice of the underlying model is of great importance, but often requires a lot of manual tuning e.g. of bin sizes or fitted functions. This can be alleviated through the use of neural networks and their inherent powerful data modeling capabilities. We present a novel and generalized method for producing scale factors using an adversarial neural network. This method is investigated in the context of the bottom-quark jet-tagging algorithms within the CMS experiment. The primary network uses the jet variables as inputs to derive the scale factor for a single jet. It is trained through the use of a second network, the adversary, which aims to differentiate between the data and rescaled simulation.

Weapons System Open Architecture (WSOA) utilizes a distributed object computing foundation as a "bridge" to enable communication between disparate real-time systems in the redirection of strike assets. This open systems "bridge" connects legacy embedded mission systems and off-board C3I (command, control, communication and information) sources and systems. WSOA will establish the potential gains in warfighting capability due to the enabling technologies of collaborative planning, information mining, and adaptive embedded resource management. This paper focuses on the experiences and lessons learned in the WSOA demonstration program with CORBA middleware, pluggable protocols and Quality of Service (QoS) enforcement, management and verification.

Abstract Many particle physics analyses are adopting the concept of vectorised computing, often making them increasingly performant and resource-efficient. While a variety of computing steps can be vectorised directly, some calculations are challenging to implement. One of these is the analytical neutrino reconstruction which involves fitting that naturally varies between events. We show a vectorised implementation of the analytical neutrino reconstruction using a graph computing model. It uses established deep learning software libraries and is natively portable to local and external hardware accelerators such as GPUs. Using the example of ttH events with a semi-leptonic final state, we present performance studies for our implementation.

Abstract Fast turnaround times for LHC physics analyses are essential for scientific success. The ability to quickly perform optimizations and consolidation studies is critical. At the same time, computing demands and complexities are rising with the upcoming data taking periods and new technologies, such as deep learning. We present a show-case of the HH→bbWW analysis at the CMS experiment, where we process 𝒪(1 − 10)TB of data on 100 threads in a few hours. This analysis is based on the columnar NanoAOD data format, makes use of the NumPy ecosystem and HEP specific tools, in particular Coffea and Dask. Data locality, especially IO latency, is optimized by employing a multi-level caching structure using local file storage and on-worker SSD caches. We process thousands of events simultaneously within a single thread, thus enabling straightforward use of vectorized operations. Resource intensive computing tasks, such as GPU accelerated DNN inference and histogram aggregation in the 𝒪(10)GB regime, are offloaded to dedicated workers. The analysis consists of hundreds of distinctly different workloads and is steered through a workflow management tool ensuring reproducibility throughout the development process up to journal publication.

VISPA (Visual Physics Analysis) is a web-platform that enables users to work on any secure shell (SSH) reachable resource using just their webbrowser. It is used successfully in research and education for HEP data analysis. The emerging JupyterLab is an ideal choice for a comprehensive, browser-based, and extensible work environment and we seek to unify it with the efforts of the VISPA-project. The primary objective is to provide the user with the freedom to access any external resources at their disposal, while maintaining a smooth integration of preconfigured ones including their access permissions. Additionally, specialized HEP tools, such as native format data browsers (ROOT, PXL), are being migrated from VISPA- to JupyterLab-extensions as well. We present these concepts and their implementation progress.

The development of an LHC physics analysis involves numerous investigations that require the repeated processing of terabytes of data. Thus, a rapid completion of each of these analysis cycles is central to mastering the science project. We present a solution to efficiently handle and accelerate physics analyses on small-size institute clusters. Our solution is based on three key concepts: Vectorized processing of collision events, the "MapReduce" paradigm for scaling out on computing clusters, and efficiently utilized SSD caching to reduce latencies in IO operations. Using simulations from a Higgs pair production physics analysis as an example, we achieve an improvement factor of $6.3$ in runtime after one cycle and even an overall speedup of a factor of $14.9$ after $10$ cycles.

Insbesondere seit Inkrafttreten der ersten Immissionsschutzregelungen zum Schutze der Gesundheit in den 1970er Jahren hat die Bedeutung der Uberwachung von gesundheitsschadlichen, toxischen Gasen in der Luft zugenommen [1]. Des Weiteren gibt es zunehmende Bestrebungen, durch die Detektion von bestimmten Gasmolekulen im menschlichen Atem Tumorerkrankungen fruhzeitig zu erkennen [2]. Damit einhergehend steht ein immer hoherer Bedarf an Gassensoren, welche die geringen Konzentrationen im parts-per-million- bis parts-per-trillion-Bereich der Schadstoffe selektiv detektieren und uberwachen konnen [3]. Fur diese Aufgabe werden derzeit zumeist Taguchi Metalloxid-Halbleitersensoren verwendet [4], welche eine Veranderung der elektrischen Leitfahigkeit unter Einfluss von bestimmten Gasen zeigen. Diese benotigen jedoch hohe Betriebstemperaturen (T > 300 ◦C ≈ 573K), um annehmbare Reaktionsgeschwindigkeiten zu erzielen und besitzen nur eine beschrankte Selektivitat [5, Kap. 18]. Um geringere Konzentrationen an Schadstoffen bei gleichbleibender oder verbesserter Reaktionsgeschwindigkeit und Selektivitat aufzulosen, werden neuartige Sensormaterialien benotigt. Einen entscheidenden Faktor, um geringe Detektionsschwellen zu erreichen, stellt das Oberflachen-zu-Volumen-Verhaltnis dar. Aus diesem Grunde stellt das im Jahre 2004 entdeckte Material Graphen [6], eine in hexagonaler Kristallstruktur zweidimensional ausgepragte Kohlenstoffschicht, einen hervorragenden Kandidaten fur zukunftige Gassensoren dar. Schedin et al. konnten so bereits im Jahre 2007 an einer mechanisch exfolierten Graphenfl ocke die Detektion von einzelnen Schadstoffmolekulen nachweisen [7]. Nichtdestotrotz stellt die mechanische Exfoliation von Graphen keinen fur die Massenanwendung tauglichen Herstellungsansatz dar. Ziel dieser Arbeit ist die Silicium (Si)-CMOS-kompatible Herstellung von Graphen-Feldeffekttransistoren (GFETs) und deren Optimierung fur Gassensor-Anwendungen. Hierfur wurde der von P.J. Wessely entwickelte Prozess der Herstellung von GFETs durch die in situ katalytisch-chemische Gasphasenabscheidung (CCVD) optimiert, wodurch selbstjustierte nanokristalline Graphen-Feldeffekttransistoren (nGFETs) mit groseren nominellen Kanallangen hergestellt werden konnen. Das nanokristalline Graphen (nG) wurde durch eine ausfuhrliche Materialcharakterisierung, bestehend aus topologischer Rasterkraftmikroskopie (AFM), Strom-Spannungs-Rasterkraftmikroskopie (CS-AFM), Raman-Spektroskopie, sowie einer vergleichenden Rontgen-Nahkanten-Absorptions-Spektroskopie (NEXAFS)-Studie analysiert. Die nGFETs zeigen ein ambipolares elektrisches Verhalten und ein von der umgebenden Atmosphare abhangiges Stromverhaltnis I_p,on / I_Dirac . Bei reduzierten Druckbedingungen wurde in einem selbstkonstruierten Vakuum-Wafertestsystem durch Messungen im passiven Betrieb der nGFETs die Veranderung des Stromflusses zwischen Drain und Source unter Exposition zu verschiedenen Gasen charakterisiert. Dabei konnten hohe Sensitivitaten der nGFETs fur Ammoniak (NH3), Stickstoffdioxid (NO2) und Kohlenmonoxid (CO) festgestellt werden. Im aktiven Betrieb der nGFETs wurden durch Messungen der Eingangskennlinien die fur die Gassensitivitat verantwortlichen physikalischen Mechanismen identifiziert. Aus den elektrischen Eigenschaften der nGFETs und den physikalischen Mechanismen der Gassensitivitat wurden auserdem mehrere Selektivitatskriterien erarbeitet, woraus sich die Moglichkeit der Differenzierung verschiedener Gasarten ergibt. Die Arbeit gliedert sich dabei wie folgt: In Kapitel 1 werden zunachst die theoretischen Grundlagen von Graphen, Halbleiterbauelementen und den verwendeten technologischen Verfahren betrachtet. Im 2. Kapitel werden die angewandten Versuchsreihen, welche zur Optimierung des in situ CCVD Graphen fuhren, beschrieben, gefolgt von einer ausfuhrlichen Behandlung der Materialcharakterisierung des nanokristallinen Graphen in Kapitel 3. Die Massenherstellung von nGFETs via in situ CCVD wird in Kapitel 4 demonstriert. Daraufhin folgen in Kapitel 5 eine elek trische Charakterisierung der nGFETs unter Vakuumbedingungen und erste Untersuchungen zur Zuverlassigkeit und Stabilitat fur Gassensor-Anwendungen. Nachfolgend wird in Kapitel 6 das Gasdetektionsvermogen der nGFETs unter Vakuumbedingungen betrachtet und eine Moglichkeit der selektiven Detektion von einzelnen Gasarten vorgestellt. Schlussendlich wird in Kapitel 7 eine Zusammenfassung mit Ausblick fur zukunftige Entwicklungsmoglichkeiten gegeben.

Abstract Deep learning architectures in particle physics are often strongly dependent on the order of their input variables. We present a two-stage deep learning architecture consisting of a network for sorting input objects and a subsequent network for data analysis. The sorting network (agent) is trained through reinforcement learning using feedback from the analysis network (environment). The optimal order depends on the environment and is learned by the agent in an unsupervised approach. Thus, the two-stage system can choose an optimal solution which is not known to the physicist in advance. We present the new approach and its application to the signal and background separation in top-quark pair associated Higgs boson production.

The VISPA (VISual Physics Analysis) project provides a streamlined work environment for physics analyses and hands-on teaching experiences with a focus on deep learning. VISPA has already been successfully used in HEP analyses and teaching and is now being further developed into an interactive deep learning platform. One specific example is to meet knowledge sharing needs in deep learning by combining paper, code and data at a central place. Additionally the possibility to run it directly from the web browser is a key feature of this development. Any SSH reachable resource can be accessed via the VISPA web interface. This enables a flexible and experiment agnostic computing experience. The user interface is based on JupyterLab and is extended with analysis specific tools, such as a parametric file browser and TensorBoard. Our VISPA instance is backed by extensive GPU resources and a rich software environment. We present the current status of the VISPA project and its upcoming new features.