Deep learning methods have gained popularity in high energy physics for fast modeling of particle showers in detectors. Detailed simulation frameworks such as the gold standard \textsc{Geant4} are computationally intensive, and current deep generative architectures work on discretized, lower resolution versions of the detailed simulation. The development of models that work at higher spatial resolutions is currently hindered by the complexity of the full simulation data, and by the lack of simpler, more interpretable benchmarks. Our contribution is \textsc{SUPA}, the SUrrogate PArticle propagation simulator, an algorithm and software package for generating data by simulating simplified particle propagation, scattering and shower development in matter. The generation is extremely fast and easy to use compared to \textsc{Geant4}, but still exhibits the key characteristics and challenges of the detailed simulation. The proposed simulator generates thousands of particle showers per second on a desktop machine, a speed up of up to 6 orders of magnitudes over \textsc{Geant4}, and stores detailed geometric information about the shower propagation. \textsc{\textsc{SUPA}} provides much greater flexibility for setting initial conditions and defining multiple benchmarks for the development of models. Moreover, interpreting particle showers as point clouds creates a connection to geometric machine learning and provides challenging and fundamentally new datasets for the field.

The two key characteristics of a normalizing flow is that it is invertible (in particular, dimension preserving) and that it monitors the amount by which it changes the likelihood of data points as samples are propagated along the network. Recently, multiple generalizations of normalizing flows have been introduced that relax these two conditions \citep{nielsen2020survae,huang2020augmented}. On the other hand, neural networks only perform a forward pass on the input, there is neither a notion of an inverse of a neural network nor is there one of its likelihood contribution. In this paper we argue that certain neural network architectures can be enriched with a stochastic inverse pass and that their likelihood contribution can be monitored in a way that they fall under the generalized notion of a normalizing flow mentioned above. We term this enrichment \emph{flowification}. We prove that neural networks only containing linear and convolutional layers and invertible activations such as LeakyReLU can be flowified and evaluate them in the generative setting on image datasets.

The tracking of charged particles produced in high energy collisions is particularly challenging. The combinatorics approach currently used to track tens of thousands of particles becomes inadequate as the number of simultaneous collisions increase at the High Luminosity Large Hadron Collider (HLLHC). We propose to reduce the complexity of tracking in such dense environments with the use of similarity hashing. We use hashing techniques to separate the detector space into buckets. The particle purity of these buckets is increased using Approximate Nearest Neighbors search. The bucket size is sufficiently small to significantly reduce the complexity of track reconstruction within the buckets. We demonstrate the use of the proposed approach on a public dataset of simulated collisions. The performance evaluation shows a significant speed improvement over the current technique and a further understanding of charged particles structure.

To attain its ultimate discovery goals, the luminosity of the Large Hadron Collider at CERN will increase so the amount of additional collisions will reach a level of 200 interaction per bunch crossing, a factor 7 w.r.t the current (2017) luminosity. This will be a challenge for the ATLAS and CMS experiments, in particular for track reconstruction algorithms. In terms of software, the increased combinatorial complexity will have to harnessed without any increase in budget. To engage the Computer Science community to contribute new ideas, we organized a Tracking Machine Learning challenge (TrackML) running on the Kaggle platform from March to June 2018, building on the experience of the successful Higgs Machine Learning challenge in 2014. The data were generated using [ACTS], an open source accurate tracking simulator, featuring a typical all silicon LHC tracking detector, with 10 layers of cylinders and disks. Simulated physics events (Pythia ttbar) overlaid with 200 additional collisions yield typically 10000 tracks (100000 hits) per event. The first lessons from the Accuracy phase of the challenge will be discussed.

Detectors of High Energy Physics experiments, such as the ATLAS dectector [1] at the Large Hadron Collider [2], serve as cameras that take pictures of the particles produced in the collision events. One of the key detector technologies used for measuring the energy of particles are calorimeters. Particles will lose their energy in a cascade (called a shower) of electromagnetic and hadronic interactions with a dense absorbing material. The number of the particles produced in this showering process is subsequently measured across the sampling layers of the calorimeter. The deposition of energy in the calorimeter due to a developing shower is a stochastic process that can not be described from first principles and rather relies on a precise simulation of the detector response. It requires the modeling of particles interactions with matter at the microscopic level as implemented using the Geant4 toolkit [3]. This simulation process is inherently slow and thus presents a bottleneck in the ATLAS simulation pipeline [4]. The current work addresses this limitation. To meet the growing analysis demands, ATLAS already relies strongly on fast calorimeter simulation techniques based on thousands of individual parametrizations of the calorimeter response [5]. The algorithms currently employed for physics analyses by the ATLAS collaboration achieve a significant speedup over the full simulation of the detector response at the cost of accuracy. Current developments [6] [7] aim at improving the modeling of taus, jet-substructure-based boosted objects or wrongly identified objects in the calorimeter and will benefit from an improved detector description following data taking and a more detailed forward calorimeter geometry. Deep Learning techniques have been improving state of the art results in various science areas such as: astrophysics [8], cosmology [9] and medical imaging [10]. These techniques are able to describe complex data structures and scale well with highdimensionality problems. Generative models are powerful deep learning algorithms to map complex distributions into a lower dimensional space, to generate samples of higher dimensionality and to approximate the underlying probability densities. Among the most promising approaches are Variational Auto-Encoders [11] [12] and Generative Adversarial Networks [13]. In this context, the talk presents the first application of such models to the fast simulation of the calorimeter response in the ATLAS detector. This work [14] demonstrates the feasibility of using such algorithms for large scale high energy physics experiments in the future, and opens the possibility to complement current techniques.