francescoferretto Goto Github PK

I'm a fervid Data Science student enthusiastic about Statistics, Mathematics, Programming, all Machine Learning's flavours and in Classical Music. You can find me on

• Bayesian Statistics and Machine Learning
• Networks
• Physics Informed Neural Networks
• Active Learning and Optimal Experimental Design
• Reinforcement Learning
• Neural Networks
• Time Series
• A personal website

• Algorithmic Methods @ here
• Neural Networks and Deep Learning @ here

• Keyword-Spotting-Recognition-through-state-of-art-neural-architectures Repo

  The project focuses on models for speech recognition of a pre-defined set of keywords (Automatic Speech Recognition System) - Under Development.

• Investigating-relationships-between-Twitter-latent-topics-and-political-orientation Repo

  The aim of this project is, after a preliminary phase of manual data collection and labeling, in finding topics by analysing the cluster of tweets' words from people having different political parties. Two approaches where adopted for topic recognition:

  • a syntactic approach (TF-IDF index) and
  • a semantic one (with Doc2Vec).

  To detect clusters of words (topics) the k-means algorithm was used (under the assumption of non-overlapping clusters of words) and applied w.r.t. time in order to analyse the progress.

• Analysis-of-Rain-Behaviour-in-Australia Repo

  The project analyses 140'000 daily meteorological measurements of 24 variables from 49 different Australian cities between the 2008 and 2017. One approach is to adapt a multiple linear regression model to the meteorological conditions (pressure, air direction, irradiation, ...) on day i wrt the quantity of fallen rain of day i+1, to overcome serious residuals' correlation the Cochrane-Orcutt procedure was applied. Another approach is to predict whether it could rain or not using logistic regression combined with dimensionality reduction techniques (e.g. PCA).

• Community-Detection-Graph-Optimization-Methods-based-on-Modularity-and-Balanced-Cuts Repo

  Optimization project that tackles the problem of community detection through:

  • a Modularity based approach
  • a Graph Cut based approach

  applied on the General Relativity and Quantum Cosmology collaboration network dataset, and other 3 from SNAP. The first optimization problem was faced with the usage of the FAST-ATVO algorithm, a software for community detection in undirected graphs with non-negative weights. The second one was addressed through Radio DCA and Fista.

• MLProjectDS-2020 Repo

  In this work we explain, discuss and solve an image recognition problem in Machine Learning in a different number of ways. We start with simpler methods like Logistic Regression or k-NN, then made some experiments with decision trees and random forests and finally use some more complex models like Support Vector Machines and Convolutional Neural Networks. We also report our results, and try to explain and interpret them. All the presented models were implemented using scikit-learn and keras modules for Python.

• Biological-Data-Project-Characterization-of-the-Phosphomethylpyrimidine-kinase-protein-domain

  This project is about the characterization of a single protein domain. As starting point, each group is provided with a representative domain sequence and the corresponding Pfam identifier. The objective of the first part of the project is to build a sequence model starting from the assigned sequence. In the second part, the actual domain family characterization is performed analyzing different aspects: structure, taxonomy and functions.