Gloria Desideri

About Me

Ciao! I am Gloria a PhD Student under the ELLIS Programme. I am supervised by Claire Vernade @ Uneversity of Nuremberg and Marlos C. Machado @ Univeristy of Alberta. My primary research interest is in reinforcement learning theory. Prior to that, I obtained my BSc and MSc at Politecnico di Milano, Italy. I worked with Prof. Francesco Trovò and Prof. Alberto Maria Metelli in AIrLab during my master thesis. Previously I was a data science intern working on applied Reinforcement Learning at MLCube.

In my free time I enjoy bouldering🧗🏻‍♀️, running🏃‍♀️, playing guitar🎸 and I am a big fan of the magic AS Roma🐺.

Latest News

Projects

Tennis Match Analysis using Computer Vision

  • Detected ball trajectory, players, and court from tennis videos using Tracknet and numerical methods.
  • Implemented trajectory smoothing using the RDP algorithm.
  • Extracted 2D and 3D ball bounce coordinates using geometrical and analytical techniques.

View on GitHub

Claim Veracity Prediction using Sentence Bert

  • Performed advanced text data cleaning and converted JSON data into CSV format.
  • Applied clustering and tSNE for nonlinear dimensionality reduction and data exploration.
  • Trained a custom classifier to efficiently embed question–answer pairs.
  • Developed an ensemble model for claim classification achieving 0.83 accuracy.
  • Built a pipeline to generate justifications for claim veracity using a T5 model.

View on GitHub

Simulation of Dynamic Pricing and Online Bidding in Stationary and Non-Stationary Environments

  • Modeled user purchase behavior and auction dynamics in both stationary and non-stationary environments achieving sub-linear regret with algorithms like GP-UCB, Thompson Sampling, and EXP3.
  • Designed and tested bidding strategies for truthful and non-truthful auctions, comparing Multiplicative Pacing and UCB under varying budgets.
  • Applied change detection methods (UCB1-CUSUM) to adapt to market shifts.
  • Analyzed multi-item auctions and pricing strategies to optimize sales dynamics.

View on GitHub

Prediction of Parkinson Development using Protein Data

  • Led a team from problem formulation to poster presentation.
  • Analyzed protein abundance data from cerebrospinal fluid samples alongside clinical data (e.g., UPDRS scores, medication usage).
  • Conducted t-tests to identify key proteins with significant expression differences.
  • Developed a regression model to capture shifts across selected proteins.
  • Presented findings in a poster session at a scientific showcase.

Publications

Awards