Brief intro into the project

The aim of this page is to present information concerning the 2-year (Apr.2022 - Mar. 2024) granted project, entitled: "Deep Learning and its Applications in Big Data Networks Performance Improvement", under the 3rd Call of PhD Fellowships (Fellowship Number: 5631), supported by the Hellenic Foundation for Research and Innovation (HFRI). Fragkou Evangelia, a member of the DANA Lab, and under the supervision of associate professor Dr.Dimitrios Katsaros, has earned this grant in order to carry our her PhD studies. Specifically, in this page we present the articles, being accepted in journals and conferences and they are also found in various bibliographic databases: Google Scholar, DBLP, ACM, Scopus.

Articles in Journals

[ J2] Fragkou E., Koultouki M., Katsaros D., "Model Reduction of Feed Forward Neural Networks for Resource-Constrained Devices", Applied Intelligence (Springer), 2022, https://doi.org/10.1007/s10489-022-04195-8. [ link ]

[ J1] Fragkou E., Papakostas D., Kasidakis T., Katsaros D., "Multilayer Backbones for Internet of Battlefield Things", Future Internet (MDPI), vol. 14, no. 6, art. 186, 2022. [ link ]

Articles in Conferences

[ C3] Fragkou E., Katsaros D., "Transfer Deep Learning for TinyML", Poster, 5th Summit on Gender Equality in Computing, (GEC'23), Athens, June 27, 2023.

[ C2] Fragkou E., Lygnos V., Katsaros D., "Transfer Learning for Convolutional Neural Networks in Tiny Deep Learning Environments", PCI2022, DOI: 10.1145/3575879.3575984. [ link ]

[ C1] Fragkou E., Katsaros D., "Memory Reduction and Training Acceleration of Neural Networks for Tiny Machine Learning", Poster, 4th Summit on Gender Equality in Computing, (GEC'22), Hybrid Event, June 16-17, 2022.

Book Chapters

[ BC1] Fragkou E., Katsaros D., "Non-Static TinyML for Ad hoc Networked Devices" in TinyML for Edge Intelligence in IoT and LPWAN Networks, 1st edition, ISBN:9780443222030, Elsevier, (to be published, June 1, 2024). [ link ]

Main goals

• Neural model reduction. The aim is to sparsify a deep neural network by cutting of connections of neurons among the network that don't contribute to the neural network training, during the training phase. In particular, we employ specific rule-oriented concepts, developed in the realm of network science (We base our motivation on observations in real neural networks whose actual topology is scale-free (or smallworld)),in order to sparsify the neural network with the aim of keeping the most significant linkages, which are responsible for better information distribution in the network. Thus, we reduce, drastically both the number of trainable variables and the size of the model, which leads to training acceleration and therefore our algorithms are applicable to resource-constrained (TinyML) devices.


• Federated learning over ad hoc networks. The aim is to drastically reduce communication in purely distributed ad hoc network, by allowing only a percentage of nodes (not every node), belonging to the network, to cooperatively train a global model, in Federated Learning techniques. So, we focus on finding smart node participation protocols (like following data similarity criteria) in order to exclude specific nodes of the training procedure and therefore reduce the data, needed to be transmitted among the nodes in the network.


• Deep learning in caching and in distributed indices. Published papers in the literature focus on learned indexes and generalization of B-trees (R -tree) concerning centralized data, so far. It would be of great interest to study and develop deep learning based techniques, with the aim of finding/matching distributed data.

Contact

>> Fragkou Evangelia, email: efragkou@uth.gr