Thesis
Thesis works are available !
A wide range of master's thesis works are developed in our laboratory (experimental and compilation). Thesis topics range from theoretical subjects to more applicative ones, covering several aspects of machine learning engineering, computational intelligence and pattern recognition (see projects). Some topics can be developed with companies in the sector (learn more). Thesis works are configured to maximize the candidate's learning curve and experience from a career perspective.
There are also several topics for the final exam related to assignments for computational intelligence and pattern recognition courses.
Note: on this page the various topics are organized by subjects and applications. In any case, many themes are more theoretical or transversal so they can be valid for more than one application. For example, the problem of unbalanced classes in classification tasks, the embedding of unstructured data in unconventional domains, the design of a custom dissimilarity measure, the evolutionary procedures for feature selection, unstructured data representation, etc.
Visit the Resources page for more information on how to write a thesis or a report for the final exam
Thesis works
Within the green energy transition, a series of applications are foreseen that concern the intelligent management of energy in regional and local distribution networks. Smart Grids are those networks that involve the massive use of ICT and Artificial Intelligence (in the various sub-disciplines) to optimize the transport, distribution and use of energy. At a local level, Smart Grids can be seen as a set of microgrids where there is no centralized control and production is close to the user who becomes a prosumer, through the use of renewable sources (solar, wind, etc.) and technologies of energy storage. In other words, microgrids are and will increasingly be the main driver for so-called Renewable Energy Communities. Therefore, it is extremely interesting to study, model and design distributed control techniques and algorithms that include demand-side management, peak shaving, load forecasting and control of energy flows using storage technologies both to maximize self-consumption and for energy trading. Numerous techniques based on optimization and decision-making algorithms are proposed in the technical literature, including evolutionary algorithms and fuzzy control systems. This topic is completely in line with the opportunities of the European Green Deal, the Italian PNRR and Industry 4.0.
These tesis topics are ralated with the "Intelligent control in Smart Grids" project
Optimization of energy flows in new generation intelligent electricity grids (Smart Grids) through fuzzy control systems and evolutionary algorithms.
Optimal control in the presence of renewable sources and controlled generators (e.g. microturbines). Optimization with multi-objective evolutionary algorithms.
Load forecasting grounded on standard machine learning algorithms or deep learning.
Optimization of energy flows in power train systems in the nautical sector through evolutive heuristics and Fuzzy Logic.
Optimization of energy flows in industrial machinery for auto-consumption and peak shaving with evolutionary strategies.
Capillo, A., De Santis, E., Mascioli, F. M. F., & Rizzi, A. (2024). An Online Hierarchical Energy Management System for Energy Communities, Complying with the Current Technical Legislation Framework. arXiv preprint arXiv:2402.01688.
Capillo, A., De Santis, E., Frattale Mascioli, F.M., Rizzi, A., (2022) Synthesis of an Evolutionary Fuzzy Multi-objective Energy Management System for an Electric Boat. IJCCI 2022: 199-208
De Santis, E., Rizzi, A., & Sadeghian, A. (2017). Hierarchical genetic optimization of a fuzzy logic system for energy flows management in microgrids. Applied Soft Computing, 60, 135-149.
De Santis, E., Rizzi, A., Sadeghiany, A., & Mascioli, F. M. F. (2013, June). Genetic optimization of a fuzzy control system for energy flow management in micro-grids. In 2013 Joint IFSA World Congress and NAFIPS Annual Meeting (IFSA/NAFIPS) (pp. 418-423). IEEE.
Leonori, S., De Santis, E., Rizzi, A., & Mascioli, F. F. (2016, October). Optimization of a microgrid energy management system based on a fuzzy logic controller. In IECON 2016-42nd Annual Conference of the IEEE Industrial Electronics Society (pp. 6615-6620). IEEE.
De Santis, E., Sadeghian, A., & Rizzi, A. (2017). A smoothing technique for the multifractal analysis of a medium voltage feeders electric current. International Journal of Bifurcation and Chaos, 27(14), 1750211.
Time series forecasting
In the context of innovative energy management systems - specifically in the synthesis of an Energy Management System - it is important to have efficient prediction algorithms. In particular, it is necessary to predict the time series of energy production from renewable sources, consumption by users and energy market prices. These time series can be considered stochastic processes characterized by appropriate distributions and made up of different structural elements related to time series in general. In particular, we refer to trends, seasonality and other components necessary for the characterization of the series. The research work, therefore, concerns the use of traditional algorithms and algorithms based on deep learning techniques for the analysis and synthesis of predictive models of time series.
Statistical analysis of time series in the energy context.
Application of traditional techniques (ARIMA) for the synthesis of predictive models of time series related to energy production, consumption and prices.
Application of deep learning techniques for the synthesis of predictive models of time series related to energy production, consumption and prices.
Transformers for the synthesis of predictive models of time series related to energy production, consumption and prices
Rastkar, S.; Zendehdel, D.; De Santis, E. and Rizzi, A. (2023). A Comparison Between Seasonal and Non-Seasonal Forecasting Techniques for Energy Demand Time Series in Smart Grids. In Proceedings of the 15th International Joint Conference on Computational Intelligence - NCTA; ISBN 978-989-758-674-3; ISSN 2184-3236, SciTePress, pages 459-467. DOI: 10.5220/0012265900003595
Bianchi, F. M., De Santis, E., Rizzi, A., & Sadeghian, A. (2015). Short-term electric load forecasting using echo state networks and PCA decomposition. Ieee Access, 3, 1931-1943.
Artificial Intelligence for Predictive Manteinance
The ability to perform predictive maintenance, as one of the main assets of Industry 4.0, is known to help improve downtime, costs, control and production quality. Modern predictive maintenance programs involve machine learning techniques, within the AI umbrella, that work in a data-driven fashion. This is true in all machinery where, through intelligent sensors, it is possible to collect data to be processed to detect faults or carry out anomaly detection activities. This is possible also thanks to Information and Communication Technologies (ICT), which -- as enabling factors -- allow not only the collection of a huge amount of data on systems but also transporting them on reliable communication networks allowing also the elaboration on powerful hardware in the cloud.
Predictive maintenance techniques in the context of Smart Grids: intelligent modeling and real-time fault prediction
Predictive maintenance of pressure signals in railways pipelines: the problem of unbalanced data using standard machine learning techniques or deep learning
De Santis, E., & Rizzi, (2024) A, Antonello, Modelling Failures in Smart Grids by a Bilinear Logistic Regression Approach. Available at SSRN: https://ssrn.com/abstract=4600241 or http://dx.doi.org/10.2139/ssrn.4600241
De Santis, E., Arnò, F., Martino, A., & Rizzi, A. (2022, July). A statistical framework for labeling unlabelled data: a case study on anomaly detection in pressurization systems for high-speed railway trains. In 2022 International Joint Conference on Neural Networks (IJCNN) (pp. 1-8). IEEE.
De Santis, E., Arnò, F., & Rizzi, A. (2022). Estimation of fault probability in medium voltage feeders through calibration techniques in classification models. Soft Computing, 26(15), 7175-7193.
De Santis, E., Capillo, A., Mascioli, F. M. F., & Rizzi, A. (2020). Classification and Calibration Techniques in Predictive Maintenance: A Comparison between GMM and a Custom One-Class Classifier. In IJCCI (pp. 503-511).
Giampieri, M., De Santis, E., Rizzi, A., & Mascioli, F. M. F. (2018, July). A supervised classification system based on evolutive multi-agent clustering for smart grids faults prediction. In 2018 International Joint Conference on Neural Networks (IJCNN) (pp. 1-8). IEEE.
De Santis, E., Rizzi, A., & Sadeghian, A. (2018). A cluster-based dissimilarity learning approach for localized fault classification in Smart Grids. Swarm and evolutionary computation, 39, 267-278.
De Santis, E., Rizzi, A., & Sadeghian, A. (2017, June). A learning intelligent system for classification and characterization of localized faults in smart grids. In 2017 IEEE Congress on Evolutionary Computation (CEC) (pp. 2669-2676). IEEE.
De Santis, E., Livi, L., Sadeghian, A., & Rizzi, A. (2015). Modeling and recognition of smart grid faults by a combined approach of dissimilarity learning and one-class classification. Neurocomputing, 170, 368-383.
Pruning methodologies of deep neural networks for video surveillance
In the context of artificial neural networks, pruning is the practice of removing parameters (which may entail removing individual parameters, or parameters in groups such as by neurons) from an existing network. The goal of this process is to maintain the accuracy of the network while increasing its efficiency. This can be done to reduce the computational resources required to run the neural network and it is important in applications that need low computational and memory burdens as in embedded systems (e.g. video surveillance). The pruning is performed in pre-trained networks for general recognition tasks and the challenge here is to teach the network to learn a small number of very informative features pruning some connections (or filters in CNNs) maintaining a trade-off between memory footprint and accuracy of recognition for the problem at hand. From a scientific point of view, this topic allows also us to study how information is represented in a distributed manner (within a deep neural network architecture) and how to reduce redundancy.
Convolutional Neural Network pruning: a survey on methodologies and algorithms
Genetic pruning for Convolutional Neural Networks
Pruning Convolutional Neural Networks, information representation and white-box models (Explainable AI)
Natuaral Languege Processing and text Mining
Natural Language Processing and text mining are awesome and topical disciplines that adopt machine learning and pattern recognition techniques in a lot of real-world applications, especially with the advent of the Internet, the World Wide Web and Social Media. Natural Language Processing means studying and researching a new way of machine understanding and production of natural language. One of the main interesting topics is the representation of text excerpts in a suitable algebraic space - embedding - trying to preserve as much as possible the syntactic-semantic relationships in the text. This topic is a specialization of the problem of information representation in the pattern recognition discipline. Text mining is, in turn, a specialization of data mining and the main aim here is investigating new and performant ways of extracting useful information from texts allowing also to build gray or white-box models in line with the Explainable AI paradigm. Moreover, in our research lab, natural language is conceived as a complex system, that is a product of the most complex system in the universe: the brain. Accordingly, machine learning and pattern recognition can be merged with granular computing and complex system modeling to understand how meaning in texts is represented. This allows us to model also some cognitive aspects of human activities. In general, applications span from supervised and unsupervised learning techniques to solve problems in social media analysis, document categorization and natural language production.
Textual embedding methodologies using transformers (deep learning) for long texts
Modeling and analyzing artificial texts generated by deep learning architectures through complex systems.
Topic Tracking for social media analysis; hot keywords/emergent topics extraction with statistical methods
(INDVAL)Hate speech detection in social media
Deep learning for language modeling through Transformer architectures
Text representation through new embedding techniques mediated from granular computing
Modeling and analyzing artificial texts generated by deep learning architectures through the study of complex systems.
De Santis, E., Martino, A., & Rizzi, A. (2024). Human versus Machine Intelligence: Assessing Natural Language Generation Models through Complex Systems Theory. IEEE Transactions on Pattern Analysis and Machine Intelligence.
De Santis, E., Martino, A., Ronci, F., & Rizzi, A. (2023, June). A Comparison of Neural Word Embedding Language Models for Classifying Social Media Users in the Healthcare Context. In 2023 International Joint Conference on Neural Networks (IJCNN) (pp. 1-9). IEEE.
De Santis, E., & Rizzi, A. (2023). Prototype Theory Meets Word Embedding: A Novel Approach for Text Categorization via Granular Computing. Cognitive Computation, 15(3), 976-997.
De Santis, E., Martino, A., Ronci, F., & Rizzi, A. (2023). An Unsupervised Graph-Based Approach for Detecting Relevant Topics: A Case Study on the Italian Twitter Cohort during the Russia–Ukraine Conflict. Information, 14(6), 330.
De Santis, E., De Santis, G., & Rizzi, A. (2023). Multifractal Characterization of Texts for Pattern Recognition: on the Complexity of Morphological Structures in Modern and Ancient Languages. IEEE Transactions on Pattern Analysis and Machine Intelligence.
De Santis, E., Capillo, A., Ferrandino, E., Mascioli, F.M.F., Rizzi, A. (2023). An Information Granulation Approach Through m-Grams for Text Classification. In: Garibaldi, J., et al. Computational Intelligence. IJCCI 2021. Studies in Computational Intelligence, vol 1119. Springer, Cham. https://doi.org/10.1007/978-3-031-46221-4_4
De Santis, E., Martino, A., & Rizzi, A. (2020). An infoveillance system for detecting and tracking relevant topics from Italian tweets during the COVID-19 event. Ieee Access, 8, 132527-132538.
Martino, A., De Santis, E., & Rizzi, A. (2020, July). An ecology-based index for text embedding and classification. In 2020 International Joint Conference on Neural Networks (IJCNN) (pp. 1-8). IEEE.
Capillo, A., De Santis, E., Mascioli, F. M. F., & Rizzi, A. (2020). Mining M-Grams by a Granular Computing Approach for Text Classification. In IJCCI (pp. 350-360).
Evolutionary Multi Angent-based clustering and classification algorithms
Evolutionary Agent-Based Clustering Classifier (E-ABC2) algorithm
In the Big Data era, designing powerful data mining and machine learning algorithms able to extract useful information from huge amounts of data is of paramount importance. Furthermore, such big datasets are often noisy and/or with plenty of redundant information, making the data mining task and the interpretability of the learning algorithm output rather difficult. Recently, multi-agent systems emerged as powerful strategies to solve data mining problems. Broadly speaking, multi-agent systems are composed of atomic units (agents) that somehow cooperate in order to reach a common goal, where the agent's definition is vague and algorithm-dependent. Multi-agent approaches can be used for local graph clustering. For example, each agent runs a different clustering algorithm in order to return the best one for the data set at hand. Hence, it is interesting both investigating the possibility of representing data locally (local metric learning) and representing unstructured data such as graphs, sequences, images etc. The last task can be explored by exploiting the so-called symbolic histogram approach widely investigated in our research lab.
Investigating new consensus mechanisms
Investigating new evolutionary strategies
Applications in Smart Grids: faults modeling and recognitions in a real-world power grid dataset
De Santis, E., Granato, G., Rizzi, A. (2023). Facing Graph Classification Problems by a Multi-agent Information Granulation Approach. In: Garibaldi, J., et al. Computational Intelligence. IJCCI 2021. Studies in Computational Intelligence, vol 1119. Springer, Cham. https://doi.org/10.1007/978-3-031-46221-4_9
Giampieri, M., Baldini, L., De Santis, E., & Rizzi, A. (2020, July). Facing big data by an agent-based multimodal evolutionary approach to classification. In 2020 International Joint Conference on Neural Networks (IJCNN) (pp. 1-8). IEEE.
Complex Systems, unstructured data representation, embeddings and Bioinformatics
In real-world applications, phenomena and objects can have a complex structure. In data-driven applications, therefore, they are represented through unstructured data such as images, graphs, sequences, time series, categorical data, histograms, n-tuple of real numbers, etc. In order to construct an automatic representation of such complex objects so that it can be automatically processed by a machine for a machine learning system, it is necessary to have an embedding procedure. This means that the unstructured object is represented as a mathematical entity in an appropriate (hopefully algebraic) space with specific properties, that is, with a norm from which it is possible to derive a measure of dissimilarity. The challenge is to synthesize a measure of dissimilarity and an appropriate embedding technique to represent such complex objects. The techniques developed apply to numerous fields such as bioinformatics, predictive maintenance, Natural Language Processing and time series prediction. The developed models can then be conceived through the Explainable AI paradigm, that is, as explainable and interpretable models. The following topic is of a more theoretical and transversal nature. Consequently, from an application point of view, it covers most of the topics presented on this page.
Unstructured data representation in non-conventional domains
Graphs embedding techniques
Protein function prediction in dissimilarity space representation
Non-euclidean data representations and custom dissimilarity measure
The problem of imbalanced classes: discriminative vs generative learning
Estimating the classification probabilities and calibrating classifiers
De Santis, E., Naraei, P., Martino, A., Sadeghian, A., & Rizzi, A. (2022). Multifractal Characterization and Modeling of Blood Pressure Signals. Algorithms, 15(8), 259.
De Santis, E., Martino, A., & Rizzi, A. (2022). On component-wise dissimilarity measures and metric properties in pattern recognition. PeerJ Computer Science, 8, e1106.
Martino, A., De Santis, E., & Rizzi, A. (2023). On Information Granulation via Data Filtering for Granular Computing-Based Pattern Recognition: A Graph Embedding Case Study. SN Computer Science, 4(3), 314.
De Santis, E., Martino, A., Rizzi, A., & Mascioli, F. M. F. (2018, July). Dissimilarity space representations and automatic feature selection for protein function prediction. In 2018 International joint conference on neural networks (IJCNN) (pp. 1-8). IEEE.
Martino, A., De Santis, E., Giuliani, A., & Rizzi, A. (2020). Modelling and recognition of protein contact networks by multiple kernel learning and dissimilarity representations. Entropy, 22(7), 794.
Martino, A., De Santis, E., Baldini, L., & Rizzi, A. (2019, September). Calibration Techniques for Binary Classification Problems: A Comparative Analysis. In IJCCI (pp. 487-495).
Ferrandino, E., Capillo, A., De Santis, E., Mascioli, F. M., & Rizzi, A. (2021). A Modular Autonomous Driving System for Electric Boats based on Fuzzy Controllers and Q-Learning.
Project works for the Exam
The projects for the final exam (homeworks) cover part of the topics covered in this page and are agreed with the professor.
PhD
What does it mean for us PhDs?
property of CIPAR TEAMS © - 2024