Page 167 - Proceedings of the 2018 ITU Kaleidoscope

Session 4: Optimization of Data Management with Machine Learning
S4.1 A Deep Reinforcement Learning Approach for Data Migration in Multi-access Edge Computing
Fabrizio De Vita, Dario Bruneo and Antonio Puliafito (University of Messina, Italy); Giovanni

Nardini, Antonio Virdis and Giovanni Stea (University of Pisa, Italy)

5G technology promises to improve the network performance by allowing users to seamlessly
access distributed services in a powerful way. In this perspective, Multi-access Edge Computing
(MEC) is a relevant paradigm that push data and computational resources nearby users with the
final goal to reduce latencies and improve resource utilization. Such a scenario requires strong
policies in order to react to the dynamics of the environment also taking into account multiple
parameter settings. In this paper, we propose a deep reinforcement learning approach that is able
to manage data migration in MEC scenarios by learning during the system evolution. We set up a
simulation environment based on the OMNeT++/SimuLTE simulator integrated with the Keras
machine learning framework. Preliminary results showing the feasibility of the proposed approach
are discussed.

S4.2 Predicting Activities in Business Processes with LSTM Recurrent Neural Networks
Edgar Tello-Leal (Autonomous University of Tamaulipas, Mexico); Jorge Roa (National
Technological University, Santa Fe Regional Faculty, Argentina); Mariano Rubiolo (National
Technological University, Santa Fe Regional Faculty & FICH/UNL-CONICET, Argentina);
Ulises Ramírez-Alcocer (Autonomous University of Tamaulipas, Mexico)

The Long Short-Term Memory (LSTM) Recurrent Neural Networks provide a high precision in
the prediction of sequences in several application domains. In the domain of business processes it
is currently possible to exploit event logs to make predictions about the execution of cases. This
article shows that LSTM networks can also be used for the prediction of execution of cases in the
context of an event log that originates from the IoT and Industry 4.0 domain. This is a key aspect
to provide valuable input for planning and resource allocation (either physical or virtual), since
each trace associated with a case indicates the sequential execution of activities in business
processes. A methodology for the implementation of an LSTM neural network is also proposed.
An event log of the industry domain is used to train and test the proposed LSTM neural network.
Our preliminary results indicate that the prediction of the next activity is acceptable according to
the literature of the domain.

– 151 –