Research project title
Computational methods and frameworks for unsupervised machine learning
Education level
Master or doctorate
Director/co-director
Director: Daniel Aloise
End of display
September 5, 2024
Areas of expertise
Primary sphere of excellence in research
Industry of the Future and Digital Society
Unit(s) and department(s)
Department of Computer Engineering and Software Engineering
Detailed description
Data is present everywhere in all sectors. Many organizations have faced the problem of having a lot of data without properly exploiting it. Machine learning models are widely used in practice to extract knowledge from data. Whether it's predicting the effects of global warming or helping detect fake news, this rapidly developing field often yields promising results. Machine learning models can be classified into two paradigms: supervised learning, where learning is assisted by data with known classes, and unsupervised learning, where no prior knowledge is available, and the goal is typically set to capture the best underlying description for the available data. Although supervised learning techniques have been successfully applied in many analytical tasks, labeled data is often scarce and expensive to obtain, especially given the speed at which data is generated in contemporary applications. Thus, the ability of unsupervised learning techniques to automatically identify data structures from raw data collections makes them very attractive.
The goal of the project will be to contribute to the practice of data mining for unsupervised learning tasks by providing efficient and effective software. Its results should promote data-driven decision-making for the era of the Internet of Things, where vast amounts of unlabeled data will be generated in real-time on an unprecedented scale. This goal will be supported by the achievement of the following short-term objectives:
A. Develop global optimization algorithms for clustering problems.
B. Explore the use of secondary information for data embedding methods.
C. Develop an anomaly detection framework using evolving embeddings for data stream clustering and novelty detection.
The software and algorithms developed in this research program should serve as new benchmarks in the field, allowing for the exploration of larger datasets and leading to new perspectives in business, industry, and academia.
Financing possibility
Scholarships available.
