Computer Science - 110/110 cum laude
Software Systems Security - 110/110 cum laude
Do you want to know more? Visit my Linkedin .
Edison (gEneralized aDaptive contInuouS authenticatiOn eNgine) is my research project for my master's thesis in Software Systems Security and involved the development of a novel generalizable, adaptive, reliable and time-oriented multibiometric continuous authentication framework equipped with context and anomaly detection mechanisms. The generalizability of the approach lies in its independence from specific hardware or biometric architectures. In addition, I created a possible implementation of the approach through an Android mobile app written in Java and a backend written in Python and Matlab. For this implementation, I used state-of-the-art machine learning-based approaches to handle face, voice, and gait authentication.
In the course of a machine learning exam I created LUST, a predictive model that can classify the topic of a legal question. I built the dataset developing a crawler that extracted legal questions from various websites and I trained the model obtaining good results through Support Vector Machine (SVM) and a pre-trained Neural Network on which I performed fine tuning on my dataset that I balanced with text augmentation techniques.
As a project for a security exam I developed ZeroTwo, a cryptolocker ransomware written in C# effective against an up-to-date Windows system with virus protection. ZeroTwo presents itself as a simple image. In fact, once opened, nothing happens except for the image itself opening. Yet, once the file is executed, the system is infected with malicious and absolutely "silent" extraction of destructive code and infected registry keys. The lack of suspicious video output is manageable through the use of various VBS and batch scripts. Once the computer is accessed again, however, the malware activates, encrypts the user's files and extracts a reverter with instructions to pay the ransom. In addition, once the damage is done, the malware self-deletes from the system, leaving no trace. I made the malware undetectable by antivirus software through antivirus evasion techniques, such as encrypting and encoding the malicious payload. Also, thanks to some heuristics, I was able to point out some vulnerabilities specific to Windows Defender, especially with regard to avoiding its cloud analysis.
With my company, we conducted a home rehabilitation project that involved implementing a motion capturing and analysis system using a Microsoft Azure Kinect and a predictive model based on machine learning to evaluate the correctness of rehabilitation exercises. The project, based on Unity technology, has been carried out in collaboration with the Defense Veterans Center and Atlantica. Following the project, I provided theoretical and practical lessons to illustrate the basic operation of the engine to militaries. The lectures took place via videoconference and, in addition, there was a final in-person workshop at the Rome "Celio" Military Hospital.
CTR is a crawler capable of clustering Tweets by topic. For each cluster, moreover, a set of labels identifying the main topics is automatically assigned. Sentiment analysis is also performed for each tweet within the cluster. The main purpose of the project was to provide a useful tool for tourism agencies, which could easily identify trends or negative situations to better design tour packages, but obviously the approach can be considered quite generic and adaptable to any social network analysis context.
As an experimental thesis for my computer science degree, I developed using the Unity engine a video game for children with ADHD (Attention-Deficit/Hyperactivity Disorder). The project contains two challenges: a revised version of the stroop test (with a variant for children not yet able to read) and a "mind race." In the latter challenge, it is necessary to make a car move as fast as possible on a circuit and the car maintains a speed directly proportional to the child's concentration, which is measured through a mindband that can capture brain waves. During both challenges, heart rate is also continuously captured through a finger sensor that communicated with the application through an Arduino board.