Cybersecurity student passionate about ML/AI. My greatest passion is video games, and I aspire to one day harness the power of AI to push the boundaries of gaming. My goal is to develop increasingly complex and immersive experiences that redefine player engagement and interactivity
In today's rapidly evolving digital landscape, the widespread use of
encryption in key protocols poses an unprecedented challenge for corporate
network security. The increased interconnection of networks and digital
technology expansion have rendered traditional firewalls less effective,
as they struggle to provide robust defense. Deep packet inspection (DPI)
has proven inconsistent and often inadequate.
This growing challenge demands a sophisticated yet lightweight solution
that can penetrate the encryption barrier to protect networks. As cyber
threats become more advanced, there is a pressing need for innovative
approaches that stay ahead of attackers while safeguarding sensitive data.
In this context, this thesis introduces an innovative tool for securing
encrypted network traffic. It provides precise network analysis,
efficiently identifying and blocking suspicious activities while
maintaining data confidentiality and overall system security.
The thesis aims to define and implement assurance checks for artificial intelligence models. These checks will focus on various aspects that impact the final model and its application, such as the training dataset, the training process, and the resulting model. The evaluations will assess several non-functional properties, including robustness to specific attacks and fairness.
This project focuses on Human Activity Recognition (HAR) using LSTM-based neural networks. The goal is to classify different human activities based on motion sensor data.
This project lays the groundwork for developing intelligent, context-aware NPCs using Large Language Models. It demonstrates how to structure their memory, personality, and dialogue logic, setting the stage for smarter, more immersive in-game interactions.
IcaDevops is used to automate dataset quality assessment in MLOps pipelines. It helps determine whether a dataset is suitable for machine learning by extracting independent components using ICA and evaluating their non-Gaussianity with kurtosis.
This project implements a method to analyze and visualize risk levels in a dataset using Support Vector Machines (SVMs) with linear and polynomial kernels. The analysis identifies "at-risk" data points, assigns them to specific risk levels, and visualizes the results in risk maps and histograms.
The goal is to identify potential biases in data distributions and decision-making processes. The project offers a structured way to analyze disparities in different groups, such as gender, race, or socio-economic status, through various fairness metrics and visualization tools.
This probe evaluates ONNX models to determine their vulnerability to membership inference attacks, a type of privacy threat. It analyzes the model's behavior probabilistically to assess the risk of sensitive data exposure based on its predictions.
This project analyzes datasets to identify the most important and potentially sensitive variables using mutual information and feature importance techniques. It helps ensure fairness in machine learning models by detecting attributes that could introduce bias or disproportionately affect model outcomes.
This project focuses on Human Activity Recognition (HAR) using LSTM-based neural networks. The goal is to classify different human activities based on motion sensor data.
This project analyzes datasets to identify the most important and potentially sensitive variables using mutual information and feature importance techniques. It helps ensure fairness in machine learning models by detecting attributes that could introduce bias or disproportionately affect model outcomes.
IcaDevops is used to automate dataset quality assessment in MLOps pipelines. It helps determine whether a dataset is suitable for machine learning by extracting independent components using ICA and evaluating their non-Gaussianity with kurtosis.
This project implements a method to analyze and visualize risk levels in a dataset using Support Vector Machines (SVMs) with linear and polynomial kernels. The analysis identifies "at-risk" data points, assigns them to specific risk levels, and visualizes the results in risk maps and histograms.
The goal is to identify potential biases in data distributions and decision-making processes. The project offers a structured way to analyze disparities in different groups, such as gender, race, or socio-economic status, through various fairness metrics and visualization tools.
This probe evaluates ONNX models to determine their vulnerability to membership inference attacks, a type of privacy threat. It analyzes the model's behavior probabilistically to assess the risk of sensitive data exposure based on its predictions.
Description of project 7.
Description of project 8.
