Packet erasure correcting codes for wireless sensor networks: implementation and field trial measurements

This thesis regards the Wireless Sensor Network (WSN), as one of the most important technologies for the twenty-first century and the implementation of different packet correcting erasure codes to cope with the ”bursty” nature of the transmission channel and the possibility of packet losses during the transmission. The limited battery capacity of each sensor node makes the minimization of the power consumption one of the primary concerns in WSN. Considering also the fact that in each sensor node the communication is considerably more expensive than computation, this motivates the core idea to invest computation within the network whenever possible to safe on communication costs. The goal of the research was to evaluate a parameter, for example the Packet Erasure Ratio (PER), that permit to verify the functionality and the behavior of the created network, validate the theoretical expectations and evaluate the convenience of introducing the recovery packet techniques using different types of packet erasure codes in different types of networks. Thus, considering all the constrains of energy consumption in WSN, the topic of this thesis is to try to minimize it by introducing encoding/decoding algorithms in the transmission chain in order to prevent the retransmission of the erased packets through the Packet Erasure Channel and save the energy used for each retransmitted packet. In this way it is possible extend the lifetime of entire network.

Segmentazione di immagini mammografiche con convolutional neural networks

Il tumore al seno si colloca al primo posto per livello di mortalità tra le patologie tumorali che colpiscono la popolazione femminile mondiale. Diversi studi clinici hanno dimostrato come la diagnosi da parte del radiologo possa essere aiutata e migliorata dai sistemi di Computer Aided Detection (CAD). A causa della grande variabilità di forma e dimensioni delle masse tumorali e della somiglianza di queste con i tessuti che le ospitano, la loro ricerca automatizzata è un problema estremamente complicato. Un sistema di CAD è generalmente composto da due livelli di classificazione: la detection, responsabile dell’individuazione delle regioni sospette presenti sul mammogramma (ROI) e quindi dell’eliminazione preventiva delle zone non a rischio; la classificazione vera e propria (classification) delle ROI in masse e tessuto sano. Lo scopo principale di questa tesi è lo studio di nuove metodologie di detection che possano migliorare le prestazioni ottenute con le tecniche tradizionali. Si considera la detection come un problema di apprendimento supervisionato e lo si affronta mediante le Convolutional Neural Networks (CNN), un algoritmo appartenente al deep learning, nuova branca del machine learning. Le CNN si ispirano alle scoperte di Hubel e Wiesel riguardanti due tipi base di cellule identificate nella corteccia visiva dei gatti: le cellule semplici (S), che rispondono a stimoli simili ai bordi, e le cellule complesse (C) che sono localmente invarianti all’esatta posizione dello stimolo. In analogia con la corteccia visiva, le CNN utilizzano un’architettura profonda caratterizzata da strati che eseguono sulle immagini, alternativamente, operazioni di convoluzione e subsampling. Le CNN, che hanno un input bidimensionale, vengono solitamente usate per problemi di classificazione e riconoscimento automatico di immagini quali oggetti, facce e loghi o per l’analisi di documenti.

Erasure error correcting codes applied to DTN communications.

The space environment has always been one of the most challenging for communications, both at physical and network layer. Concerning the latter, the most common challenges are the lack of continuous network connectivity, very long delays and relatively frequent losses. Because of these problems, the normal TCP/IP suite protocols are hardly applicable. Moreover, in space scenarios reliability is fundamental. In fact, it is usually not tolerable to lose important information or to receive it with a very large delay because of a challenging transmission channel. In terrestrial protocols, such as TCP, reliability is obtained by means of an ARQ (Automatic Retransmission reQuest) method, which, however, has not good performance when there are long delays on the transmission channel. At physical layer, Forward Error Correction Codes (FECs), based on the insertion of redundant information, are an alternative way to assure reliability. On binary channels, when single bits are flipped because of channel noise, redundancy bits can be exploited to recover the original information. In the presence of binary erasure channels, where bits are not flipped but lost, redundancy can still be used to recover the original information. FECs codes, designed for this purpose, are usually called Erasure Codes (ECs). It is worth noting that ECs, primarily studied for binary channels, can also be used at upper layers, i.e. applied on packets instead of bits, offering a very interesting alternative to the usual ARQ methods, especially in the presence of long delays. A protocol created to add reliability to DTN networks is the Licklider Transmission Protocol (LTP), created to obtain better performance on long delay links. The aim of this thesis is the application of ECs to LTP.

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.

Polar Codes for error correction: Analysis and Decoding Algorithms

I Polar Codes sono la prima classe di codici a correzione d’errore di cui è stato dimostrato il raggiungimento della capacità per ogni canale simmetrico, discreto e senza memoria, grazie ad un nuovo metodo introdotto recentemente, chiamato ”Channel Polarization”. In questa tesi verranno descritti in dettaglio i principali algoritmi di codifica e decodifica. In particolare verranno confrontate le prestazioni dei simulatori sviluppati per il ”Successive Cancellation Decoder” e per il ”Successive Cancellation List Decoder” rispetto ai risultati riportati in letteratura. Al fine di migliorare la distanza minima e di conseguenza le prestazioni, utilizzeremo uno schema concatenato con il polar code come codice interno ed un CRC come codice esterno. Proporremo inoltre una nuova tecnica per analizzare la channel polarization nel caso di trasmissione su canale AWGN che risulta il modello statistico più appropriato per le comunicazioni satellitari e nelle applicazioni deep space. In aggiunta, investigheremo l’importanza di una accurata approssimazione delle funzioni di polarizzazione.