| |||||||
Technical Keynote Lecture
Prof. Giuseppe Di Fatta (University of Reading)
Multi-Task Deep Learning
AbstractThis talk provides an overview of Multi-Task Learning (MTL) and to its adoption in Deep Neural Networks (DNN). MTL is based on the introduction of an inductive bias in a learning process when multiple tasks are learned simultaneously. It aims at achieving better performance in jointly learned models than learning each task independently. The underlying assumption is that leveraging on common structure, on shared domain information and, in general, on similarities of related tasks, can lead to better a generalisation level of data models, avoiding overfitting and eventually compensating the scarcity and sparsity of data on each individual task. Moreover, it has been shown that the benefits of MTL can be extended to cases with unrelated groups of tasks. MTL can be considered an approach to 'transfer' learning, where each task is learned simultaneously, whereas in other transfer learning approaches the models are learned sequentially. MTL for DNN is an interesting research direction, which may help to investigate and understand deep learning processes, and to develop more efficient and more effective ones.Professor Giuseppe Di Fatta is the Head of the Department of Computer Science at the University of Reading, UK. In 1999, he was a research fellow at the International Computer Science Institute (ICSI), Berkeley, CA, USA. From 2000 to 2004, he was with the High-Performance Computing and Networking Institute of the National Research Council, Italy. From 2004 to 2006, he was with the University of Konstanz, Germany, where he joined the initial KNIME development team until the first release of KNIME 1.0 in 2006. His research interests include data science, data mining and machine learning algorithms, distributed and parallel computing, and data-driven multidisciplinary applications. He has published over 120 articles in peer-reviewed conferences and journals, is the founder of the IEEE ICDM Workshop on Data Mining in Networks and has chaired several other international events. He is a member of the Technical Committee on Machine Learning (TC-ML) of the IEEE SMC Society.
| |||||||
