PhD Program in Computer and Control Engineering
Elena Baralis – email@example.com
Benoit Huet – firstname.lastname@example.org
Phd Student: Andrea Pasini
Context of the research activity
Due to the increased processing capability provided by big data architectures, deep learning techniques are gaining momentum for classification and prediction problems. Deep learning algorithms build complex models, which provide high quality classification performance, but are hardly interpretable by human beings. Exploiting semantics both in the learning and interpretation process may enhance effectiveness and flexibility in the learning approach, and increase model interpretability.
To this aim, intelligent systems may mimic the characteristics of the human cognitive process, which encompasses (i) induction, i.e., the process that leads from data to abstract representations, corresponding to general rules and concepts, and (ii) deduction, i.e., the application of abstract concepts to derive new information or new abstractions. Effective data summarization and pattern representation techniques will be explored to represent abstractions, which may then be used to drive the generation of models based both on available data and on patterns, possibly generated by different data collections (like, e.g., in transfer learning). The candidate will focus her/his research activities on the design and development of novel data mining algorithms and techniques to (a) discover summarization patterns that effectively describe the main features of the considered classes (e.g., for concept discovery in images) and (b) exploit semantic information (e.g. ontologies, generated patterns) to improve the classification models and prediction processes. The new approaches will contribute to a paradigm-shift in distributed data mining, by also addressing issues raised by the characteristics of Big Data such as data sparsity, horizontal scaling, and parallel computation.
The research activity fits in the SmartData@PoliTo interdepartmental center, that brings together competences from different fields, ranging from modeling to computer programming, from communications to statistics. The candidate will join this interdisciplinary team of experts and collaborate with them.
The research activity will be co-advised by Prof. Benoit Huet (Eurecom Data Science Department).
The objective of the research activity is the definition of big data analytics approaches capable of extracting and managing knowledge of heterogeneous types (e.g., structured data, textualinformation, images). The following steps are envisioned:
- Abstract pattern extraction. The definition of new semantic patterns to represent abstractions on data. Different abstract representations (e.g., correlations, semantic relationships) will be evaluated on datasets characterized by diverse data types and data distributions.
- Heterogeneous pattern storage and integration. Extracted patterns will be stored in an abstract pattern lake, from which relevantpatterns will be efficiently retrieved for the needed classification task.
- Augmented classification algorithms design and development. State-of-the-art deep learning algorithms will be augmented with semantic pattern to improve prediction capability and interpretability of the result.
Skills and competencies for the development of the activity
The candidate should have excellent programming skills, programming experience in the Hadoop/Spark ecosystem, good knowledge of machine learning algorithms.
Further information about the PhD program at Politecnico can be found here
Back to the list of PhD positions