Eleftherios Mylonas

Mr. Eleftherios Mylonas

Department of Electrical and Computer Engineering, University of Patras
orcid

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
9 February 2020 in Electronics
Verified
Journal Article
Open access
PDF file
EMEleftherios Mylonas
Eleftherios Mylonas
NTNikolaos Tzanis
Nikolaos Tzanis
MBMichael Birbas
Michael Birbas
Alexios BirbasAlexios Birbas
Alexios Birbas
Smart grid technology is the next step to the evolution of classical power grids, providing robustness, reliability, and security throughout the network, enabling real-time management and control. To achieve these goals, distributed computing (microgrid concept) and intelligent control algorithms, tailored to the nature and needs of the network under study, are necessary. To deal with the vast diversity of power grids, being able to capture the dynamics of any given network, and create tools for network analysis, apparatus testing, and power grid management, an automatic design framework for real-time power system simulators is needed. In this article, a prototype of this approach is presented, employing Field Programmable Gate Array (FPGA) platforms due to their reconfigurability that enables low-power, low-latency, and high-performance designs, as a first attempt towards an open source platform, compatible with the majority of hardware design suites. It comprises two major parts: (i) a user-oriented section, built in Matlab/Simulink; and (ii) a hardware-oriented section, written in Matlab and Very High Speed Integrated Circuit (VHSIC)-Hardware Description Language (VHDL) code. To verify its functionality, two test power networks were given in a schematic format, analyzed through Matlab code and turned into dedicated hardware simulators with the aid of the VHDL template. Then, simulation results from Simulink and the prototype were compared for error estimation. The results show the prototype’s successful implementation with minimal resources utilization, high performance and low latency in the order of nanoseconds in Xilinx 6- and 7-series FPGAs, therefore proving its modularity and efficient use in many different scenarios, meeting low-latency/real-time requirements while enabling further smart grid research.

28 January 2022 in IEEE Access
Verified
Journal Article
Open access
GGGerasimos Gerogiannis
Gerasimos Gerogiannis
MBMichael Birbas
Michael Birbas
ALAimilios Leftheriotis
Aimilios Leftheriotis
EMEleftherios Mylonas
Eleftherios Mylonas
NTNikolaos Tzanis
Nikolaos Tzanis
Alexios BirbasAlexios Birbas
Alexios Birbas
In this work, we present the design and implementation of an ultra-low latency Deep Reinforcement Learning (DRL) FPGA based accelerator for addressing hard real-time Mixed Integer Programming problems. The accelerator exhibits ultra-low latency performance for both training and inference operations, enabled by training-inference parallelism, pipelined training, on-chip weights and replay memory, multi-level replication-based parallelism and DRL algorithmic modifications such as distribution of training over time. The design principles can be extended to support hardware acceleration for other relevant DRL algorithms (embedding the experience replay technique) with hard real time constraints. We evaluate the accuracy of the accelerator in a task offloading and resource allocation problem stemming from a Mobile Edge Computing (MEC/5G) scenario. The design has been implemented on a Xilinx Zynq Ultrascale+ MPSoC ZCU104 evaluation kit using High Level Synthesis. The accelerator achieves near optimal performance and exhibits a 10-fold decrease in training-inference execution latency when compared to a high-end CPU-based implementation.

2 February 2023 in IEEE Transactions on Cloud Computing
Verified
Journal Article
Closed access
NTNikolaos Tzanis
Nikolaos Tzanis
EMEleftherios Mylonas
Eleftherios Mylonas
PPPanagiotis Papaioannou
Panagiotis Papaioannou
MBMichael Birbas
Michael Birbas
Alexios BirbasAlexios Birbas
Alexios Birbas
APAlex Papalexopoulos
Alex Papalexopoulos
Edge Cloud is providing unprecedented opportunities for IoT and WAMS (Wide Area Monitoring Systems) in electrical grid operation. It is an orchestrated environment able to address low latency events through appropriate edge-cloud computing configurations.Transient State Estimation (TSE) is a key monitoring tool for capturing a reliable knowledge of the Smart Grid status in real-time, given the impediments introduced by the increasing penetration of Distributed Energy Resources in the energy mix. Frequency response anomalies, large scale transients, and voltage swings can be captured by TSE for real time or post failure data analytics. This work presents a cloud edge framework for the efficient calculation of TSE which, albeit its benefits, demands high computational resources at the edge (close to the measurement units) along with ultra low latency communications. The framework enables TSE as a service through the coordination of Virtual Machines (VMs) running on virtualized infrastructure and other non-virtualized physical nodes. In order to support the stringent time requirements, part of the TSE is offloaded to dedicated hardware acceleration units (FPGA). The proposed TSE framework is validated using an IEEE 30-bus, and the results show a significant superiority in terms of total latency compared to conventional cloud and edge deployments.

17 January 2026 in Electronics
Verified
Journal Article
Open access
PDF file
EMEleftherios Mylonas
Eleftherios Mylonas
CFChrisanthi Filippou
Chrisanthi Filippou
SKSotirios Kontraros
Sotirios Kontraros
MBMichael Birbas
Michael Birbas
Alexios BirbasAlexios Birbas
Alexios Birbas
The ever-increasing need for energy-efficient implementation of AI algorithms has driven the research community towards the development of many hardware architectures and frameworks for AI. A lot of work has been presented around FPGAs, while more sophisticated architectures like CGRAs have also been at the center. However, AI ecosystems are isolated and fragmented, with no standardized way to compare different frameworks with detailed Power–Performance–Area (PPA) analysis. This paper bridges the gap by presenting a unified, fully open-source hardware-aware AI acceleration pipeline that enables seamless deployment of neural networks on both FPGA and CGRA architectures. Built around the Brevitas quantization framework, it supports two distinct backend flows: FINN for high-performance dataflow accelerators and CGRA4ML for low-power coarse-grained reconfigurable designs. To facilitate this, a model translation layer from QONNX to QKeras is also introduced. To demonstrate its effectiveness, we use an autoencoder model for anomaly detection in wind turbines. We deploy our accelerated models on the AMD’s ZCU104 and benchmark it against a Raspberry Pi. Evaluation on a realistic cyber–physical testbed shows that the hardware-accelerated solutions achieve substantial performance and energy-efficiency gains—up to 10× and 37× faster inference per flow and over 11× higher efficiency—while maintaining acceptable reconstruction accuracy.

11 June 2025
Verified
Conference Paper
Closed access
DBDimitrios Brodimas
Dimitrios Brodimas
DKDimitrios Kapolos
Dimitrios Kapolos
EMEleftherios Mylonas
Eleftherios Mylonas
Alexios BirbasAlexios Birbas
Alexios Birbas
MBMichael Birbas
Michael Birbas
The continuous expansion and transformation of next-gen networks presents a substantial challenge for management and orchestration (MANO) frameworks to efficiently handle the increasing computational demand, particularly at the network's edge. To enhance MANO decision-making processes, it is essential to proactively identify both periodic and aperiodic patterns that may influence network's behavior. This paper presents a computational load forecasting approach designed to provide MANO frameworks with predictive insights into infrastructure conditions. Forecasting leverages machine learning models such as LSTMs and Bi-LSTMs, employing the federated learning family of algorithms, FedOpt, to enable distributed training. Experimental results demonstrate that the proposed federated learning strategy achieves performance comparable to centralized learning methods while significantly reducing the amount of data exchanged between edge nodes and the cloud.