TeachingDidattica
I teach a doctoral course on the statistical physics of learning, lead two postgraduate modules in the Department of Computer Science at UCL, and supervise industry projects through UCL IXN.Tengo un corso di dottorato sulla fisica statistica dell’apprendimento, sono responsabile di due moduli magistrali al Department of Computer Science di UCL e seguo progetti con l’industria tramite UCL IXN.
I corsi si svolgono in inglese, quindi le schede dei corsi qui sotto sono in inglese.
PhD courseCorso di dottorato
Statistical Physics for Machine Learning
Phase transitions, inference and high-dimensional landscapes
A doctoral course delivered at the Department of Computer, Control and Management Engineering (DIAG) of Sapienza, my alma mater. It treats learning as a problem in statistical mechanics: starting from entropy, free energy and Bayesian inference, maximum likelihood, regularisation and posterior inference become instances of the same physics, and phase transitions explain when learning becomes hard. Each day mixes lectures, whiteboard derivations, exercises and open problems.
Programme
- Foundations — entropy, free energy and Bayesian inference; learning as statistical mechanics
- Disordered systems — random-field Ising model, Sherrington–Kirkpatrick spin glass, p-spin model, random k-SAT
- Methods — replica, cavity and mean-field approaches
- Random matrix theory — semicircle law, Stieltjes transform, BBP transition
- Applications — PCA and community detection, Hopfield networks, loss-landscape Hessians and the edge of stability, diffusion models
- Open problems — overparameterised networks, implicit regularisation, the limits of high-dimensional generation
Background expected
Linear algebra, probability and basic calculus. No prior knowledge of statistical physics or spin glasses is needed.
Listing in the Sapienza PhD course catalogue →
Master’s programmesMaster
My modules are options on four one-year MSc programmes at UCL, so each class brings together students from computer science, quantitative finance, risk and data science.I miei moduli sono opzionali in quattro MSc annuali di UCL, quindi ogni classe riunisce studenti di informatica, finanza quantitativa, rischio e data science.
UCL modulesModuli UCL
Both modules are 15-credit, Level 7 (postgraduate) modules taught in person in Term 2, and are assessed half through in-class activity and half through coursework.Entrambi i moduli valgono 15 crediti, sono di livello 7 (magistrale), si tengono in presenza nel secondo trimestre e sono valutati per metà con attività in aula e per metà con progetti.
Advanced Machine Learning in Finance
Modern machine-learning methods applied to the problems that matter in finance: derivatives pricing and model calibration, portfolio allocation and hedging, investment decisions, credit scoring, fraud detection and risk management. The module also introduces text mining as a source of alternative data. It is hands-on throughout: students implement models in Python with Keras/TensorFlow on real and simulated financial data, with an emphasis on noisy, non-stationary datasets.
Topics
- Neural network fundamentals, layer types and architectures
- Optimisation: loss functions, regularisation, stochastic gradient descent, overfitting and generalisation
- Supervised deep learning: CNNs, RNNs and LSTMs
- Unsupervised learning: PCA and autoencoders
- Reinforcement learning for dynamic control problems in finance
- Graph neural networks and learning on financial networks
- Text mining and sentiment analysis; time-series prediction
Example projects
- Network filtering and anomaly detection
- Link prediction in temporal networks
- Autoencoders for payment-network analysis
- Return forecasting with recurrent networks
- Sentiment analysis of financial text
Learning outcomes
- Describe the main advanced algorithms (CNNs, RNNs, reinforcement learning) and choose the right one for a financial problem
- Speed up pricing and calibration, and formulate dynamic control problems in finance
- Build flexible models for financial prediction and analyse textual data algorithmically
- Write and adapt machine-learning code in Python with Keras and TensorFlow
Background expected
Probability, linear algebra and multivariate calculus; programming in a numerical language; and Machine Learning with Applications in Finance (COMP0050 or COMP0198).
Full description in the UCL module catalogue →
Algorithmic Trading
An introduction to algorithmic trading and risk-premia strategies: why they work, what their statistical properties are, and how to design, implement and evaluate them. The module surveys the main families of strategies used in industry, points to open areas of research including recent machine-learning approaches, and gives students the tools to build strategies and to test their known and provable properties rigorously.
Topics
- The trading industry, market structure and data sources
- Families of trading and risk-premia strategies
- Limit order book dynamics
- Portfolio theory and portfolio construction
- Statistical analysis and evaluation of strategies
- Sharpe ratio and other performance metrics
- Backtest overfitting, multiple hypothesis testing and model validation
Learning outcomes
- Analyse trading strategies statistically
- Research, design and develop new strategies
Background expected
Probability and statistics, mathematical analysis, and fluency in a scientific programming language (Python is used in class).
Full description in the UCL module catalogue →
Course previewsAnteprime dei corsi
UCL IXN industry projects
The UCL Industry Exchange Network (IXN), running in Computer Science since 2011, embeds real industry problems into the degree. Each project pairs a UCL academic supervisor with a technical mentor from the partner organisation, so students work on a live problem while partners prototype ideas at low risk.
- Master's students spend three months full-time on an IXN dissertation or team project over the summer (June–September). Confidential and closed-source projects are possible under NDA.
- Undergraduates spend part of each year on industry projects during term, from team projects to final-year dissertations.
- Partners range from large technology firms and international banks to insurers, start-ups, the NHS and charities.
I supervise IXN projects in quantitative finance, financial networks and machine learning. Organisations interested in proposing a project for next summer can get in touch; proposals for master's projects are usually collected between November and January.
SupervisionSupervisione
I supervise PhD and MSc projects on complex networks, random matrix theory and physics-informed machine learning, often with applications in finance. Prospective students are welcome toSeguo tesi di dottorato e di master su reti complesse, teoria delle matrici aleatorie e machine learning informato dalla fisica, spesso con applicazioni in finanza. Chi è interessato può get in touchscrivermi with a short description of their interests.con una breve descrizione dei propri interessi.