Pulse shape studies of neutral particles with a liquid scintillator

The hadrontherapy exploits beams of charged particles against deep cancers. These ions have a depth-dose profile in which there is a little release of energy at the beginning of their path, whereas there is a sharp maximum, the Bragg Peak, near its end path. However, if heavy ions are used, the fragmentation of the projectile can happen and the fragments can release some dose outside the treatment volume beyond the Bragg peak. The fragmentation process takes place also when the Galactic Cosmic Rays at high energy hit the spaceship during space missions. In both cases some neutrons can be produced and if they interact with the absorbing materials nuclei some secondary particles are generated which can release energy. For this reason, studies about the cross section measurements of the fragments generated during the collisions of heavy ions against the tissues nuclei are very important. In this context, the FragmentatiOn Of Target (FOOT) experiment was born, and aims at measuring the differential and double differential fragmentation cross sections for different kinetic energies relevant to hadrontherapy and space radioprotection with high accuracy. Since during fragmentation processes also neutrons are produced, tests of a neutron detection system are ongoing. In particular, recently a neutron detector made up of a liquid organic scintillator, BC-501A with neutrons/gammas discrimination capability was studied, and it represents the core of this thesis. More in details, an analysis of the data collected at the GSI laboratory, in Darmstadt, Germany, is effectuated which consists in discriminating neutral and charged particles and then to separate neutrons from gammas. From this analysis, a preliminary energy-differential reaction cross-section for the production of neutrons in the 16O + (C_2H_4)_(n) and 16O + C reactions was estimated.

Embedding Video Object Detection Capabilities on Low-Power Nano-Drones

Our objective for this thesis work was the deployment of a Neural Network based approach for video object detection on board a nano-drone. Furthermore, we have studied some possible extensions to exploit the temporal nature of videos to improve the detection capabilities of our algorithm. For our project, we have utilized the Mobilenetv2/v3SSDLite due to their limited computational and memory requirements. We have trained our networks on the IMAGENET VID 2015 dataset and to deploy it onto the nano-drone we have used the NNtool and Autotiler tools by GreenWaves. To exploit the temporal nature of video data we have tried different approaches: the introduction of an LSTM based convolutional layer in our architecture, the introduction of a Kalman filter based tracker as a postprocessing step to augment the results of our base architecture. We have obtain a total improvement in our performances of about 2.5 mAP with the Kalman filter based method(BYTE). Our detector run on a microcontroller class processor on board the nano-drone at 1.63 fps.

Valutazione sperimentale di approcci di TinyML per problemi di classificazione: un caso di studio con Arduino Nano 33 BLE

Il TinyMachineLearning (TinyML) è un campo di ricerca nato recentemente che si inserisce nel contesto dell’Internet delle cose (IoT). Mentre l’idea tradizionale dell’IoT era che i dati venissero inviati da un dispositivo locale a delle infrastrutture cloud per l’elaborazione, il paradigma TinyML d’altra parte, propone di integrare meccanismi basati sul Machine Learning direttamente all’interno di piccoli oggetti alimentati da microcontrollori (MCU ). Ciò apre la strada allo sviluppo di nuove applicazioni e servizi che non richiedono quindi l’onnipresente supporto di elaborazione dal cloud, che, come comporta nella maggior parte dei casi, consumi elevati di energia e rischi legati alla sicurezza dei dati e alla privacy. In questo lavoro sono stati svolti diversi esperimenti cercando di identificare le sfide e le opportunità correlate al TinyML. Nello specifico, vengono valutate e analizzate le prestazioni di alcuni algoritmi di ML integrati in una scheda Arduino Nano 33 BLE Sense, attraverso un framework TinyML. Queste valutazioni sono state effettuate conducendo cinque diversi macro esperimenti, ovvero riconoscimento di Colori, di Frequenze, di Vibrazioni, di Parole chiave e di Gesti. In ogni esperimento, oltre a valutare le metriche relative alla bontà dei classificatori, sono stati analizzati l’occupazione di memoria e il tasso di inferenza (tempo di predizione). I dati utilizzati per addestrare i classificatori sono stati raccolti direttamente con i sensori di Arduino Nano. I risultati mostrano che il TinyML può essere assolutamente utilizzato per discriminare correttamente tra diverse gamme di suoni, colori, modelli di vibrazioni, parole chiave e gesti aprendo la strada allo sviluppo di nuove promettenti applicazioni sostenibili.

Design of a UHF-band transceiver module for nano-satellite applications

The thesis presents the UHF band transceiver project carried out under the lead of Spacemind company. In particular reports the outcome of the first phase of the project encompassing management tasks, requirements definition and the first electrical design. Then follows the study of the UHF band antenna which develops in parallel with the transceiver. The antenna plus the transceiver will be sold together as a complete UHF telecommunication system for cubesats made by Spacemind. As a main result, this work contributed to the design and manufacturing of the first transceiver prototype.

Phenomenology of top-philic axion-like particles

Axion like particles (ALPs), i.e., pseudo-scalar bosons interacting via derivative couplings, are a generic feature of many new physics scenarios, including those addressing the strong-CP problem and/or the existence of dark matter. Their phenomenology is very rich, with a wide range of scales and interactions being directly probed at very different experiments, from accelerators to observatories. In this thesis, we explore the possibility that ALPs might indirectly affect precision collider observables. In particular, we consider an ALPs that preferably couple to the top quark (top-philic) and we study new-physics 1- loop corrections to processes involving top quarks in the final state. Our study stems from the simple, yet non-trivial observation that 1-loop corrections are infrared finite even in the case of negligible ALP masses and therefore can be considered on their own. We compute the 1-loop corrections of new physics analytically in key cases involving top quark pair production and then implement and validate a fully general next-to-leading-order model in MadGraph5_aMC@NLO that allows to compute virtual effects for any process of interest. A detailed study of the expected sensitivity to virtual ALPs in ttbar production at the LHC is performed.

Continual learning on nano drones

Miniaturized flying robotic platforms, called nano-drones, have the potential to revolutionize the autonomous robots industry sector thanks to their very small form factor. The nano-drones’ limited payload only allows for a sub-100mW microcontroller unit for the on-board computations. Therefore, traditional computer vision and control algorithms are too computationally expensive to be executed on board these palm-sized robots, and we are forced to rely on artificial intelligence to trade off accuracy in favor of lightweight pipelines for autonomous tasks. However, relying on deep learning exposes us to the problem of generalization since the deployment scenario of a convolutional neural network (CNN) is often composed by different visual cues and different features from those learned during training, leading to poor inference performances. Our objective is to develop and deploy and adaptation algorithm, based on the concept of latent replays, that would allow us to fine-tune a CNN to work in new and diverse deployment scenarios. To do so we start from an existing model for visual human pose estimation, called PULPFrontnet, which is used to identify the pose of a human subject in space through its 4 output variables, and we present the design of our novel adaptation algorithm, which features automatic data gathering and labeling and on-device deployment. We therefore showcase the ability of our algorithm to adapt PULP-Frontnet to new deployment scenarios, improving the R2 scores of the four network outputs, with respect to an unknown environment, from approximately [−0.2, 0.4, 0.0,−0.7] to [0.25, 0.45, 0.2, 0.1]. Finally we demonstrate how it is possible to fine-tune our neural network in real time (i.e., under 76 seconds), using the target parallel ultra-low power GAP 8 System-on-Chip on board the nano-drone, and we show how all adaptation operations can take place using less than 2mWh of energy, a small fraction of the available battery power.