arXiv subjects

Marco Selvi

Publications and source records attributed to Marco Selvi.

8 recordsLinked to original sources

arXiv2026

Analytical model for the photomultiplier single photoelectron response including the electron back-scattering contribution

Many models exist to describe the single photoelectron response of single photon counting photomultipliers. Generally to describe the spectral region between the fully amplified primary photoelectron peak and the electronics pedestal an ad hoc function is used (often an exponentially modified gaussian) attributing this region to `noise'. In this paper, following the physical description of back-scattered primary photoelectrons at the first dynode described in the "The Photomultiplier Handbook" by A. G. Wright published by Oxford University Press, we derive an analytical function describing these partially amplified primary photoelectron at the first dynode. This function depends only on intrinsic parameters of the photomultiplier such as the gain at the first dynode and the intrinsic resolution of the dynode chain following the first. Furthermore, analytical descriptions of the fully amplified peak and very low charge signals are derived. The model has been successfully validated with data from two different photomultipliers acquired with a low-noise amplifier.

physics.ins-det↗

arXiv2025

CWM: An Open-Weights LLM for Research on Code Generation with World Models

We release Code World Model (CWM), a 32-billion-parameter open-weights LLM, to advance research on code generation with world models. To improve code understanding beyond what can be learned from training on static code alone, we mid-train CWM on a large amount of observation-action trajectories from Python interpreter and agentic Docker environments, and perform extensive multi-task reasoning RL in verifiable coding, math, and multi-turn software engineering environments. With CWM, we provide a strong testbed for researchers to explore the opportunities world modeling affords for improving code generation with reasoning and planning in computational environments. We present first steps of how world models can benefit agentic coding, enable step-by-step simulation of Python code execution, and show early results of how reasoning can benefit from the latter. CWM is a dense, decoder-only LLM trained with a context size of up to 131k tokens. Independent of its world modeling capabilities, CWM offers strong performance on general coding and math tasks: it reaches pass@1 scores of 65.8% on SWE-bench Verified (with test-time scaling), 68.6% on LiveCodeBench, 96.6% on Math-500, and 76.0% on AIME 2024. To support further research on code world modeling, we release model checkpoints after mid-training, SFT, and RL.

cs.SE↗

arXiv2023

Nash Learning from Human Feedback

Reinforcement learning from human feedback (RLHF) has emerged as the main paradigm for aligning large language models (LLMs) with human preferences. Typically, RLHF involves the initial step of learning a reward model from human feedback, often expressed as preferences between pairs of text generations produced by a pre-trained LLM. Subsequently, the LLM's policy is fine-tuned by optimizing it to maximize the reward model through a reinforcement learning algorithm. However, an inherent limitation of current reward models is their inability to fully represent the richness of human preferences and their dependency on the sampling distribution. In this study, we introduce an alternative pipeline for the fine-tuning of LLMs using pairwise human feedback. Our approach entails the initial learning of a preference model, which is conditioned on two inputs given a prompt, followed by the pursuit of a policy that consistently generates responses preferred over those generated by any competing policy, thus defining the Nash equilibrium of this preference model. We term this approach Nash learning from human feedback (NLHF). In the context of a tabular policy representation, we present a novel algorithmic solution, Nash-MD, founded on the principles of mirror descent. This algorithm produces a sequence of policies, with the last iteration converging to the regularized Nash equilibrium. Additionally, we explore parametric representations of policies and introduce gradient descent algorithms for deep-learning architectures. To demonstrate the effectiveness of our approach, we present experimental results involving the fine-tuning of a LLM for a text summarization task. We believe NLHF offers a compelling avenue for preference learning and policy optimization with the potential of advancing the field of aligning LLMs with human preferences.

stat.ML↗

arXiv2020

The FragmentatiOn Of Target Experiment (FOOT) and its DAQ system

The FragmentatiOn Of Target (FOOT) experiment aims to provide precise nuclear cross-section measurements for two different fields: hadrontherapy and radio-protection in space. The main reason is the important role the nuclear fragmentation process plays in both fields, where the health risks caused by radiation are very similar and mainly attributable to the fragmentation process. The FOOT experiment has been developed in such a way that the experimental setup is easily movable and fits the space limitations of the experimental and treatment rooms available in hadrontherapy treatment centers, where most of the data takings are carried out. The Trigger and Data Acquisition system needs to follow the same criteria and it should work in different laboratories and in different conditions. It has been designed to acquire the largest sample size with high accuracy in a controlled and online-monitored environment. The data collected are processed in real-time for quality assessment and are available to the DAQ crew and detector experts during data taking.

physics.ins-det↗

arXiv2016

Supernova neutrino physics with xenon dark matter detectors: A timely perspective

Dark matter detectors that utilize liquid xenon have now achieved tonne-scale targets, giving them sensitivity to all flavours of supernova neutrinos via coherent elastic neutrino-nucleus scattering. Considering for the first time a realistic detector model, we simulate the expected supernova neutrino signal for different progenitor masses and nuclear equations of state in existing and upcoming dual-phase liquid xenon experiments. We show that the proportional scintillation signal (S2) of a dual-phase detector allows for a clear observation of the neutrino signal and guarantees a particularly low energy threshold, while the backgrounds are rendered negligible during the supernova burst. XENON1T (XENONnT and LZ; DARWIN) experiments will be sensitive to a supernova burst up to 25 (35; 65) kpc from Earth at a significance of more than 5 sigma, observing approximately 35 (123; 704) events from a 27 Msun supernova progenitor at 10 kpc. Moreover, it will be possible to measure the average neutrino energy of all flavours, to constrain the total explosion energy, and to reconstruct the supernova neutrino light curve. Our results suggest that a large xenon detector such as DARWIN will be competitive with dedicated neutrino telescopes, while providing complementary information that is not otherwise accessible.

astro-ph.HE↗

arXiv2015

Dark matter sensitivity of multi-ton liquid xenon detectors

We study the sensitivity of multi ton-scale time projection chambers using a liquid xenon target, e.g., the proposed DARWIN instrument, to spin-independent and spin-dependent WIMP-nucleon scattering interactions. Taking into account realistic backgrounds from the detector itself as well as from neutrinos, we examine the impact of exposure, energy threshold, background rejection efficiency and energy resolution on the dark matter sensitivity. With an exposure of 200 t x y and assuming detector parameters which have been already demonstrated experimentally, spin-independent cross sections as low as $2.5 \times 10^{-49}$ cm$^2$ can be probed for WIMP masses around 40 GeV/$c^2$. Additional improvements in terms of background rejection and exposure will further increase the sensitivity, while the ultimate WIMP science reach will be limited by neutrinos scattering coherently off the xenon nuclei.

physics.ins-det↗

arXiv2008

The LVD Core Facility: a study of LVD as muon veto and active shielding for dark matter experiments

In this study we explore the possibility of using the existing structure of a running experiment, the LVD supernova observatory at the INFN Gran Sasso National Laboratory, as an active shield and veto for the muon-induced background. In our vision LVD could become (without affecting in any way its main purpose of SN neutrino telescope) a host for a relatively compact but massive experiment looking for rare events. The empty volume that can be obtained removing 2 modules from the most internal part of the detector is 2.1m x 6.2m x 2.8m; we will call it LVD Core Facility (LVD-CF). We have evaluated the active vetoing and shielding power of LVD, with a detailed MC simulation (based on Geant4) of the detector and the rock that surrounds it. The results show that the flux of neutrons that are not associated with a visible muon in LVD is very low; it results reduced by a factor 50, equivalent to the one present in a much deeper underground laboratory, i.e. Sudbury. Moreover we present the results of on-going measurements about the gamma contamination inside the LVD-CF: it is reduced by a factor greater than 10 with respect to the one measured outside the LVD detector.

hep-ex↗

arXiv2006

LVD highlights

The Large Volume Detector (LVD) in the INFN Gran Sasso National Laboratory, Italy, is a $ν$ observatory mainly designed to study low energy neutrinos from the gravitational collapse of galactic objects. The experiment has been monitoring the Galaxy since June 1992, under increasing larger configurations: in January 2001 it has reached its final active mass M=1 kt. LVD is one of the largest liquid scintillator apparatus in the world for the detection of stellar collapses and, besides SNO, SuperKamiokande and Amanda, it is a charter member of the SNEWS network, that has become fully operational since July 1st, 2005. No gravitational core-collapse has been detected by LVD during 14 years of data acquisition; this allows to put an upper limit of 0.18 events y$^{-1}$ in our galaxy at the 90% C.L. The LVD tracking system allows the detection and the recontruction of the cosmic muon tracks in a large fraction of the whole solid angle, in particular also horizontal tracks can be reconstructed. The results of the muon depth--intensity relation and of the flux of neutrino--induced muons are presented. Moreover, during 2006, the CNGS beam will start its operation: the performances of LVD as a beam monitor are described.

hep-ex↗