To Train or Not to Train: Balancing Efficiency and Training Cost in Deep Reinforcement Learning for Mobile Edge Computing
Authors:
Maddalena Boscaro,
Federico Mason,
Federico Chiariotti,
Andrea Zanella
Abstract:
Artificial Intelligence (AI) is a key component of 6G networks, as it enables communication and computing services to adapt to end users' requirements and demand patterns. The management of Mobile Edge Computing (MEC) is a meaningful example of AI application: computational resources available at the network edge need to be carefully allocated to users, whose jobs may have different priorities and…
▽ More
Artificial Intelligence (AI) is a key component of 6G networks, as it enables communication and computing services to adapt to end users' requirements and demand patterns. The management of Mobile Edge Computing (MEC) is a meaningful example of AI application: computational resources available at the network edge need to be carefully allocated to users, whose jobs may have different priorities and latency requirements. The research community has developed several AI algorithms to perform this resource allocation, but it has neglected a key aspect: learning is itself a computationally demanding task, and considering free training results in idealized conditions and performance in simulations. In this work, we consider a more realistic case in which the cost of learning is specifically accounted for, presenting a new algorithm to dynamically select when to train a Deep Reinforcement Learning (DRL) agent that allocates resources. Our method is highly general, as it can be directly applied to any scenario involving a training overhead, and it can approach the same performance as an ideal learning agent even under realistic training conditions.
△ Less
Submitted 11 November, 2024;
originally announced November 2024.
Analysis of geospatial behaviour of visitors of urban gardens: is positioning via smartphones a valid solution?
Authors:
Francesco Pirotti,
Alberto Guarnieri,
Marco Piragnolo,
Marco Boscaro,
Raffaele Cavalli
Abstract:
Tracking locations is practical and speditive with smartphones, as they are omnipresent devices, relatively cheap, and have the necessary sensors for positioning and networking integrated in the same box. Nowadays recent models have GNSS antennas capable of receiving multiple constellations. In the proposed work we test the hypothesis that GNSS positions directly recorded by smartphones can be a v…
▽ More
Tracking locations is practical and speditive with smartphones, as they are omnipresent devices, relatively cheap, and have the necessary sensors for positioning and networking integrated in the same box. Nowadays recent models have GNSS antennas capable of receiving multiple constellations. In the proposed work we test the hypothesis that GNSS positions directly recorded by smartphones can be a valid solution for spatial analysis of people's behaviour in an urban garden. Particular behaviours can be linked to therapeutic spots that promote health and well-being of visitors. Three parts are reported: (i) assessment of the accuracy of the positions relative to a reference track, (ii) implementation of a framework for automating transmission and processing of the location information, (iii) analysis of preferred spots via spatial analytics. Different devices were used to survey at different times and with different methods, i.e. in the pocket of the owner or on a rigid frame. Accuracy was estimated using distance of each located point to the reference track, and precision was estimated with static multiple measures. A chat-bot through the Telegram application was implemented to allow users to send their data to a centralized computing environment thus automating the spatial analysis. Results report a horizontal accuracy below ~2.3 m at 95% confidence level, without significant difference between surveys, and very little differences between devices. GNSS-only and assisted navigation with telephone cells also did not show significant difference. Autocorrelation of the residuals over time and space showed strong consistency of the residuals, thus proving a valid solution for spatial analysis of walking behaviour.
△ Less
Submitted 20 June, 2021;
originally announced July 2021.
