Analysis and Evaluation of Tiny Machine Learning applications

The aim of TinyML is to bring the capability of Machine Learning to ultra-low-power devices, typically under a milliwatt, and with this it breaks the traditional power barrier that prevents the widely distributed machine intelligence. TinyML allows greater reactivity and privacy by conducting inference on the computer and near-sensor while avoiding the energy cost associated with wireless communication, which is far higher at this scale than that of computing. In addition, TinyML’s efficiency makes a class of smart, battery-powered, always-on applications that can revolutionize the collection and processing of data in real time. This emerging field, which is the end of a lot of innovation, is ready to speed up its growth in the coming years. In this thesis, we deploy three model on a microcontroller. For the model, datasets are retrieved from an online repository and are preprocessed as per our requirement. The model is then trained on the split of preprocessed data at its best to get the most accuracy out of it. Later the trained model is converted to C language to make it possible to deploy on the microcontroller. Finally, we take step towards incorporating the model into the microcontroller by implementing and evaluating an interface for the user to utilize the microcontroller’s sensors. In our thesis, we will have 4 chapters. The first will give us an introduction of TinyML. The second chapter will help setup the TinyML Environment. The third chapter will be about a major use of TinyML in Wake Word Detection. The final chapter will deal with Gesture Recognition in TinyML.

Design of a GNSS receiver for the ESEO mission

This thesis was carried out inside the ESA's ESEO mission and focus in the design of one of the secondary payloads carried on board the spacecraft: a GNSS receiver for orbit determination. The purpose of this project is to test the technology of the orbit determination in real time applications by using commercial components. The architecture of the receiver includes a custom part, the navigation computer, and a commercial part, the front-end, from Novatel, with COCOM limitation removed, and a GNSS antenna. This choice is motivated by the goal of demonstrating the correct operations in orbit, enabling a widespread use of this technology while lowering the cost and time of the device’s assembly. The commercial front-end performs GNSS signal acquisition, tracking and data demodulation and provides raw GNSS data to the custom computer. This computer processes this raw observables, that will be both transferred to the On-Board Computer and then transmitted to Earth and provided as input to the recursive estimation filter on-board, in order to obtain an accurate positioning of the spacecraft, using the dynamic model. The main purpose of this thesis, is the detailed design and development of the mentioned GNSS receiver up to the ESEO project Critical Design Review, including requirements definition, hardware design and breadboard preliminary test phase design.

Computer aided design and finite element modeling to predict bulk mechanical properties of bone scaffolds using triply periodic minimal surfaces (progettazione assistita al calcolatore ed analisi con il metodo degli elementi finiti per predire il comportamento meccanico di scaffolds ossei creati con l'uso di superfici minime periodiche in tre direzioni)

The aim of Tissue Engineering is to develop biological substitutes that will restore lost morphological and functional features of diseased or damaged portions of organs. Recently computer-aided technology has received considerable attention in the area of tissue engineering and the advance of additive manufacture (AM) techniques has significantly improved control over the pore network architecture of tissue engineering scaffolds. To regenerate tissues more efficiently, an ideal scaffold should have appropriate porosity and pore structure. More sophisticated porous configurations with higher architectures of the pore network and scaffolding structures that mimic the intricate architecture and complexity of native organs and tissues are then required. This study adopts a macro-structural shape design approach to the production of open porous materials (Titanium foams), which utilizes spatial periodicity as a simple way to generate the models. From among various pore architectures which have been studied, this work simulated pore structure by triply-periodic minimal surfaces (TPMS) for the construction of tissue engineering scaffolds. TPMS are shown to be a versatile source of biomorphic scaffold design. A set of tissue scaffolds using the TPMS-based unit cell libraries was designed. TPMS-based Titanium foams were meant to be printed three dimensional with the relative predicted geometry, microstructure and consequently mechanical properties. Trough a finite element analysis (FEA) the mechanical properties of the designed scaffolds were determined in compression and analyzed in terms of their porosity and assemblies of unit cells. The purpose of this work was to investigate the mechanical performance of TPMS models trying to understand the best compromise between mechanical and geometrical requirements of the scaffolds. The intention was to predict the structural modulus in open porous materials via structural design of interconnected three-dimensional lattices, hence optimising geometrical properties. With the aid of FEA results, it is expected that the effective mechanical properties for the TPMS-based scaffold units can be used to design optimized scaffolds for tissue engineering applications. Regardless of the influence of fabrication method, it is desirable to calculate scaffold properties so that the effect of these properties on tissue regeneration may be better understood.

Deep learning for computer vision: a comparison between convolutional neural networks and hierarchical temporal memories on object recognition tasks

In recent years, Deep Learning techniques have shown to perform well on a large variety of problems both in Computer Vision and Natural Language Processing, reaching and often surpassing the state of the art on many tasks. The rise of deep learning is also revolutionizing the entire field of Machine Learning and Pattern Recognition pushing forward the concepts of automatic feature extraction and unsupervised learning in general. However, despite the strong success both in science and business, deep learning has its own limitations. It is often questioned if such techniques are only some kind of brute-force statistical approaches and if they can only work in the context of High Performance Computing with tons of data. Another important question is whether they are really biologically inspired, as claimed in certain cases, and if they can scale well in terms of "intelligence". The dissertation is focused on trying to answer these key questions in the context of Computer Vision and, in particular, Object Recognition, a task that has been heavily revolutionized by recent advances in the field. Practically speaking, these answers are based on an exhaustive comparison between two, very different, deep learning techniques on the aforementioned task: Convolutional Neural Network (CNN) and Hierarchical Temporal memory (HTM). They stand for two different approaches and points of view within the big hat of deep learning and are the best choices to understand and point out strengths and weaknesses of each of them. CNN is considered one of the most classic and powerful supervised methods used today in machine learning and pattern recognition, especially in object recognition. CNNs are well received and accepted by the scientific community and are already deployed in large corporation like Google and Facebook for solving face recognition and image auto-tagging problems. HTM, on the other hand, is known as a new emerging paradigm and a new meanly-unsupervised method, that is more biologically inspired. It tries to gain more insights from the computational neuroscience community in order to incorporate concepts like time, context and attention during the learning process which are typical of the human brain. In the end, the thesis is supposed to prove that in certain cases, with a lower quantity of data, HTM can outperform CNN.