arXiv subjects

David Walter

Publications and source records attributed to David Walter.

7 recordsLinked to original sources

arXiv2026

Efficient binned profile likelihood minimization for precision measurements with RABBIT

Precision measurements at the LHC increasingly rely on binned profile maximum likelihood fits with thousands of bins and nuisance parameters, and the High-Luminosity LHC will push these numbers further. Fast and robust minimization of such likelihoods is crucial for timely analysis development and accurate inference. We present Rabbit (Rapid Automatic Bin-Based Inference Tool), a Python framework that exploits differentiable programming in TensorFlow 2 to perform this task on CPUs and GPUs. Automatic differentiation provides exact gradients and Hessian-vector products for a trust-region minimizer operating in Krylov subspaces, and just-intime compilation yields near-C++ execution speed. Rabbit implements flexible statistical models with analytic treatments where possible, supports symmetrization options that establish Gaussian approximations and a linearized likelihood formulation with deterministic solutions, and focuses on measuring physical observables through differentiable transformations of the model, including unfolded differential cross sections. Benchmarks on synthetic models demonstrate excellent scaling with the number of bins and parameters, outperforming established tools in challenging regimes where these fail to converge within reasonable time.

hep-ex↗

arXiv2025

SubMIT: A Physics Analysis Facility at MIT

The recently completed SubMIT platform is a small set of servers that provide interactive access to substantial data samples at high speeds, enabling sophisticated data analyses with very fast turnaround times. Additionally, it seamlessly integrates massive processing resources for large-scale tasks by connecting to a set of powerful batch processing systems. It serves as an ideal prototype for an Analysis Facility tailored to meet the demanding data and computational requirements anticipated during the High-Luminosity phase of the Large Hadron Collider. The key features that make this facility so powerful include highly optimized data access with a minimum of 100Gbps networking per server, a large managed NVMe storage system, and a substantial spinning-disk Ceph file system. The platform integrates a diverse set of high multicore CPU machines for tasks benefiting from the multithreading and GPU resources for example for neural network training. SubMIT also provides and supports a flexible environment for users to manage their own software needs for example by using containers. This article describes the facility, its users, and a few complementary, generic and real-life analyses that are used to benchmark its various capabilities.

cs.DC↗

arXiv2025

Annealing behaviour of charge collection of neutron irradiated diodes from 8-inch p-type silicon wafers

To face the higher levels of radiation due to the 10-fold increase in integrated luminosity during the High-Luminosity LHC, the CMS detector will replace the current Calorimeter Endcap (CE) using the High-Granularity Calorimeter (HGCAL) concept. The electromagnetic section as well as the high-radiation regions of the hadronic section of the CE will be equipped with silicon pad sensors, covering a total area of 620 $\rm m^2$. Fluences up to $\rm1.0\cdot10^{16}~n_{eq}/cm^{2}$ and doses up to 2 MGy are expected considering an integrated luminosity of 3 $\rm ab^{-1}$. The whole CE will normally operate at -35{\deg}C in order to mitigate the effects of radiation damage. The silicon sensors are processed on novel 8-inch p-type wafers with an active thickness of 300 $\mu m$, 200 $\mu m$ and 120 $\mu m$ and cut into hexagonal shapes for optimal use of the wafer area and tiling. With each main sensor several small sized test structures (e.g pad diodes) are hosted on the wafers, used for quality assurance and radiation hardness tests. In order to investigate the radiation-induced bulk damage, these diodes have been irradiated with reactor neutrons at the TRIGA reactor in JSI (Jo\v{z}ef Stefan Institute, Ljubljana) to 13 fluences between $\rm6.5\cdot10^{14}~n_{eq}/cm^{2}$ and $\rm1.5\cdot10^{16}~n_{eq}/cm^{2}$. The charge collection of the irradiated silicon diodes was determined through transient current technique (TCT) measurements. The study focuses on the isothermal annealing behaviour of the bulk material at 60{\deg}C. The results have been used to extend the usage of thicker silicon sensors in regions expecting higher fluences and are being used to estimate the expected annealing effects of the silicon sensors during year-end technical stops and long HL-LHC shutdowns currently foreseen with a temperature around 0{\deg}C.

physics.ins-det↗

arXiv2023

Inclusive and differential results of top quark pair production from the ATLAS and CMS experiments

This report summarizes recent results of inclusive and differential $\mathrm{t\bar{t}}$ cross section measurements from the ATLAS and CMS Collaborations at the LHC. Measurements at $\sqrt{s}=7,\ 8,\ 13$, and $13.6\,$TeV are compared to state-of-the-art theory predictions, using different PDF sets, matrix element calculations, or parton shower models. No significant disagreement of a single inclusive measurement is found, with an overall trend towards lower values. For the differential measurements, no theory model is able to describe the data across all bins.

hep-ex↗

arXiv2022

Adaptive Block Floating-Point for Analog Deep Learning Hardware

Analog mixed-signal (AMS) devices promise faster, more energy-efficient deep neural network (DNN) inference than their digital counterparts. However, recent studies show that DNNs on AMS devices with fixed-point numbers can incur an accuracy penalty because of precision loss. To mitigate this penalty, we present a novel AMS-compatible adaptive block floating-point (ABFP) number representation. We also introduce amplification (or gain) as a method for increasing the accuracy of the number representation without increasing the bit precision of the output. We evaluate the effectiveness of ABFP on the DNNs in the MLPerf datacenter inference benchmark -- realizing less than $1\%$ loss in accuracy compared to FLOAT32. We also propose a novel method of finetuning for AMS devices, Differential Noise Finetuning (DNF), which samples device noise to speed up finetuning compared to conventional Quantization-Aware Training.

cs.LG↗

arXiv2022

Measurement of top-quark electroweak couplings in associated top quark production with vector bosons at the ATLAS and CMS experiments

Recent analyses of top quark production in association with vector bosons are summarized, representing the most precise inclusive and differential cross section measurements of these processes to date. Proton-proton collision data at a center-of-mass energy of $\sqrt{s}=13$ TeV, recorded by the ATLAS and CMS detectors at the LHC are analyzed, corresponding to an integrated luminosity of up to 139 fb$^{-1}$ for each experiment. Comparisons with theory calculations are performed and overall good agreement with standard model predictions are obtained.

hep-ex↗

arXiv2022

Rare top quark production in CMS

Latest results from the CMS experiment at the LHC on top quark production in association with a Z boson or a photon, and the production of four top quarks, are summarized. Proton-proton collision data corresponding to an integrated luminosity of up to 138 fb$^{-1}$, collected at 13 TeV center-of-mass energy are used. Among the measured results, most precise inclusive cross sections, most stringent limits on models beyond the standard model and first differential measurements are presented. Overall, good agreement with standard model predictions is observed.

hep-ex↗