| corso/attività | ore |
Deep Learning ad Structured Inference – Neural Models and Algorithms for Linguistic Recognition and Inference
Modern AI is growingly faced with complex problems, characterized by heterogeneous forms of structured evidence in input and complex decisions. In medicine historical data, biological phenomena or images manifest through streams of structured data, usually digitally represented into sequences, trees or graphs. Machine Learning methods for structured learning have been studied whereas some mathematical paradigms (such as dimensionality reduction, structured kernels or neural embedding) have been proposed as modeling tools. In Natural Language Processing, Machine Translation and other Natural Language Inference (NLI) tasks, such as Question Answering or Textual Entailment, have been approached via kernels or neural models of the input representation. These achieved accurate state-of-the-art classification and prediction capabilities by enabling the exploration of huge spaces of possible solutions (e.g. target sequences or decisions). In this way, they correspond to both enabling technologies and software tools as well as to models of investigation able to systematically select hypotheses and validate controversial theories about linguistic phenomena. The application of these empirical methodologies to other areas like biology, medicine and medical robotics is more than promising, given the similar complexity of the domains targeted by AI and Life Sciences. The course will try to promote this interesting research perspective in Deep Learning to PhD students with a specific focus, but not limited to, Life Science phenomena.
Tipologia: altro
Tipo corso: internazionale
Macrosettore: open science
Area: scientifica
Elevata formazione: SI
Verifica finale: SI
Lingua: ITALIANO/INGLESE
Modalità: riconducibile al progetto formativo del Dottorando
Ciclo di Seminari
| 12 |
An introduction to score-based generative models
In simple words, generative modeling consists in learning a map capable of generating new data instances that resemble a given set of observations, starting from a simple prior distribution, most often a standard Gaussian distribution. This course aims at providing a mathematical introduction to generative models and in particular to Score-based Generative Models (SGM). SGMs have gained prominence for their ability to generate realistic data across diverse domains, making them a popular tool for researchers and practitioners in machine learning. Participants will learn about the methodological and theoretical foundations, as well as some practical applications associated with these models. The first two lectures motivate the use of generative models, introduce their formalism and present two simple though relevant examples: energy-based models and Generative Adversarial Networks. In the third and fourth lecture we present score-based diffusion models and explain how they provide an algorithmic framework to the basic idea that sampling from the time-reversal of a diffusion process converts noise into new data instances. We shall do so following two different approaches: a first elementary one that only relies on discrete transition probabilities, and a second one based on stochastic calculus. After this introduction, we derive sharp theoretical guarantees of convergence for score-based diffusion models assembling together ideas coming from stochastic control, functional inequalities and regularity theory for HamiltonJacobi-Bellman equations. The course ends with an overview of some of the most recent and sophisticated algorithms such as flow matching and diffusion Sch¨odinger bridges (DSB), which bring an (entropic) optimal transport insight into generative modeling.
Tipologia: scuole di formazione dedicate
Tipo corso: internazionale
Macrosettore: open science
Area: umanistica
Elevata formazione: SI
Verifica finale: SI
Lingua: INGLESE
Modalità: riconducibile al progetto formativo del Dottorando
Lecturers: Proff. Giovanni Conforti (Università di Padova) and Alain Durmus (École Polytechnique, Parigi)
| 12 |
Quantile regression
The main techniques of quantile regression, an alternative to classical linear regression, will be introduced. As an example, consider a regression model in which we estimate the association between Equivalised Disposable Income of a sample of households and various predictors, including an exogenous treatment. Using quantile regression, it is possible to estimate the effect of treatment on the entire distribution of households, resulting in a potentially different estimated effect at each quantile. Indeed, the treatment could be positive for the income of rich households (high quantiles) and negative for the income of poor households (low quantiles). Similarly, the association of predictors with median income can be evaluated, avoiding the need to assume that the response is Gaussian (symmetric, homoschedastic) and that there are no outliers. If time permits, principles of robust statistics will also be discussed, including linear regression techniques and robust prediction.
Background: Use of software R. Undergraduate Courses in Statistical Inference and Linear Models
Tipologia: scuole di formazione dedicate
Tipo corso: internazionale
Macrosettore: open science
Area: scientifica
Elevata formazione: SI
Verifica finale: SI
Lingua: ITALIANO/INGLESE
Modalità: riconducibile al progetto formativo del Dottorando
Prof. Alessio Farcomeni (alessio.farcomeni@uniroma2.it)
| 0 |
Simulation-based Predictive Process Mining
The course introduces the essential elements of process mining (PM) and simulation. These approaches are initially proposed as tools for analyzing processes from different perspectives, to achieve different objectives. While PM aims to extract knowledge by analyzing a log that records data on past process executions, simulation provides predictions on future or alternative behaviors of the same process. Then, an innovative point of view is proposed in which PM and simulation are seen as complementary tools whose joint adoption leads to an effective analysis paradigm. The first part of the course introduces basic concepts on simulation: simulation modeling, discrete event simulation, local and distributed simulation. The implementation of a Java-based discrete event simulator is also discussed. In the second part, principles, methods, and tools for PM are provided. Finally, the course introduces “Predictive Process Mining” as an innovative paradigm based on the joint use of the two approaches. It is outlined how the knowledge extracted from the log analysis through PM techniques can be used to guide the development of a simulation model, whose execution provides further insights into the system under study. In this context, the most relevant research challenges, opportunities and open issues are illustrated.
Background: Basic skills in software development and knowledge of at least one object-oriented programming language (Java recommended).
Tipologia: scuole di formazione dedicate
Tipo corso: internazionale
Macrosettore: open science
Area: scientifica
Elevata formazione: SI
Verifica finale: SI
Lingua: ITALIANO/INGLESE
Modalità: riconducibile al progetto formativo del Dottorando
Lecturer: dott. Paolo Bocciarelli (paolo.bocciarelli@uniroma2.it)
| 12 |