HoloCIM: An Ising Machine-as-a-Service Platform

Authors Details

Moysis Symeonides, Michalis Kasioulis, Joanna Georgiou, George Pallis, Marios D. Dikaiakos, Haris Volos, Jason Sakellariou, Alexis Askitopoulos, Simos Tsintzos, Kyriakos Georgiou, Emmanouil Lioudakis, Andreas Othonos

Research Unit

Laboratory for Internet Computing, Department of Computer Science, and Laboratory for Ultrafast science, Department of Physics

Description

Combinatorial optimization problems present challenges across various domains, such as logistics, network design, and decision-making, since the computation of an optimal solution requires an exponential number of trials. Traditional computers struggle with these problems, which are known to be NP-hard, due to their high computational complexity. As finding optimal solutions to such problems is computationally intractable, heuristic algorithms are typically proposed to yield suboptimal solutions. An alternative approach involves physical-inspired hardware systems that can efficiently solve specific NP-hard problems. To use these solvers, the input problem must be mapped into a configuration compatible with the hardware, and the output must then be translated into a useful representation. One of the most common physical-inspired models is the Ising model. An Ising machine model is represented by a grid, with node values of either 1 or -1 and grid edges assigned static weights. The goal is to minimize the Hamiltonian equation on the grid, which is represented by the sum of the product of node values and their respective weights. By finding the set of node values that results in the lowest possible Hamiltonian output, we effectively solve the optimization problem encoded in the grid, leading us to the optimal solution. In HoloCIM, by utilizing interconnected physical components, such as laser beam generators, lenses, and monitoring devices, in a specific deployment, we can efficiently discover the solution to an Ising problem. Additionally, we offer extra capabilities like as-a-service offerings, providing a user-friendly interface, auto-compilation of NP problems, monitoring capabilities, permission management, programming interfaces, and more. End-users can utilize our platform through the HoloCIM Python library to solve their Ising model problems or any of HoloCIM's pre-designed problems like number partitioning and bin packing.

Fogify: A Fog Computing Emulation Framework

Authors Details

Moysis Symeonides, Zacharias Georgiou, Demetris Trihinas, George Pallis, Marios D. Dikaiakos

Research Unit

Laboratory for Internet Computing, Department of Computer Science

Description

Fog computing is a decentralized computing infrastructure that brings data processing and storage closer to the data source—at the edge of the network—to reduce latency and improve efficiency compared to the Cloud, and it is emerging as the dominant paradigm bridging the computational and connectivity gaps between IoT sensing devices and latency-sensitive services. However, experimenting and evaluating IoT services is a daunting task involving the manual configuration and deployment of a mixture of geo-distributed physical and virtual infrastructure with different resource and network requirements. This results in sub-optimal, costly, and error-prone deployments due to numerous unexpected overheads not initially envisioned in the design phase and underwhelming testing conditions not resembling the end environment. For all these reasons, we introduce Fogify, an emulator that eases the modeling, deployment, and large-scale experimentation of fog and edge testbeds. Fogify provides a toolset to (i) model complex fog topologies comprised of heterogeneous resources, network capabilities, and QoS criteria; (ii) deploy the modeled configuration and services using popular containerized descriptions to a cloud or local environment; (iii) experiment, measure and evaluate the deployment by injecting faults at runtime to test different "what-if" scenarios that reveal the limitations of an IoT service. In the evaluation, IoT services with real-world workloads are introduced to show the wide applicability and benefits of Fogify's rapid prototyping. Finally, on top of the Fogify framework, we create various libraries specialized for the evaluation of different problems and environments, like 5G deployments, Big Data systems, and Federated Learning, among others.

Experimental waveform design for SWIPT using machine learning algorithms

Authors Details

Petros Stylianou, Elio Faddoul, and Ioannis Krikidis

Research Unit

IRIDA Research Centre for Communication Technologies, Department of Electrical and Computer Engineering

Description

Wireless power transfer (WPT) technology has the potential to revolutionize the powering of wireless devices, particularly in Internet of Things (IoT) applications where devices require both energy and information transmission. To this end, our project presents the design of a new waveform for simultaneous wireless information and power transfer (SWIPT) systems using software-defined radio (SDR) tools. In particular, by varying the number of multi-sinusoidal signals as well as the separation distance between the transmitter and receiver, we analyze both power transfer efficiency and the capacity to extract information from the received signal. This process allows us to develop a dataset derived from experimental measurements, which is then analyzed using machine learning algorithms. Our goal is to not only identify the received waveforms and extract information, but also evaluate the performance of different machine learning models. This approach offers a low-power solution for IoT networks, as the SWIPT receiver is composed of passive elements such as rectennas, eliminating the need for active components while still enabling efficient information and power transfer.

FIMIScope: Software-supported Participatory Analysis of Disinformation

Authors Details

Marios Dikaiakos Team: Katerina Ioannidou, Demetris Paschalides, Kyriakos Nikos, Dimos Stefanidis, George Pallis

Research Unit

Laboratory for Internet Computing, Department of Computer Science

Description

The proliferation of fake news in online social media has been identified as one of the main problems in social media platforms. Many seminal studies have explored this phenomenon, documenting the potential impact of fake news on modern society, and the threat they pose to the democratic process and social cohesion. Recent incident analyses have provided evidence that fake news, and the underlying mechanisms that make them viral in online settings, are weaponized by state and non-state geopolitical actors who launch computational propaganda campaigns as part of psychological warfare seeking to achieve short-term objectives and/or long-term strategic goals. The potency of industrialized misinformation, threatens to undermine the democratic fabric of EU countries. As a result, combatting disinformation is one of the three main pillars of the European Democracy Action Plan, and the European Parliament advocates for the development of “sound, robust and interlinked, systems to detect, analyze, track and map” incidents of online disinformation. In response to these concerns and initiatives, several research groups, NGOs, governmental and intergovernmental organizations, including the European External Actions Service, the Carnegie Endowment for International Peace, the Brookings Institute, and NATO, have proposed and developed methodological frameworks and tools to analyze and cope with “Foreign Information Manipulation and Interference” (FIMI) campaigns. Following a review of several such proposals and inspired by Design Thinking and Business Model Analysis methodologies, we have designed a methodology for the participatory analysis of FIMI campaigns. The methodology entails an analysis process and two “FIMI canvases,” which facilitate the analysis of FIMI observables, incidents, and campaigns. The canvases can be printed, posted on a board, and used by teams of analysts in participatory, iterative analysis exercises of FIMI cases. To enhance the capturing of semi-structured, annotated data identified during the execution of such analyses, and to relieve expert analysts from the tedious work of encoding, collecting, and storing metadata about FIMI cases under examination, we designed and implemented FIMIScope, a software toolset prototype that implements the FIMI canvases as a graphical user interface running on a browser and supported by a back-end software system which provides services for: i) storing, managing, and exporting data and meta-data collected during the analysis process, in a simple JSON format; ii) supporting remote collaboration among analysts collaboratively analyzing cases, and iii) keeping snapshots of different stages of an analysis exercise and versioning. FIMIScope is enhanced with a Large-Language Module tool, which retrieves JSON data collected and creates narratives describing captured FIMI incidents and campaigns, and allows analysts to interact with the collected data through a natural language, dialogue interface.

Revolutionizing Film Analysis and Cultural Discovery Through AI

Authors Details

Landry Digeon

Research Unit

French and European Studies Department, Digital Humanities, Film, and Intercultural Communication

Description

The Möbius Trip Replay is an advanced AI-driven multimodal deep-learning toolkit developed through the collaboration of a Humanities scholar and an AI engineer. This innovative solution automatically manages extensive datasets from movies and TV shows, enabling comprehensive digital analysis of cultural representations. Addressing the challenge of objectively deconstructing and measuring cinematic and cultural elements, Möbius Trip Replay combines cultural analytics, reverse-engineering, and distant reading methodologies to uncover intricate patterns and trends within audiovisual content. Technically, the platform is built on the Google Cloud Platform, utilizing custom configurations of GPU instances, specialized Cloud Run Functions running text, image, and audio algorithms under optimal parallel batch processing. We develop and use custom models to improve character tracking, camera angle and shot scale recognition, color detection and various additional advanced visual metrics, optimized through Bit Pair Encoding (BPE) for improved recognition accuracy. The user interface, developed with Looker Studio, offers interactive visualisations and customisable filters, facilitating both granular and broad-spectrum analyses. Scientifically, Möbius Trip Replay integrates theoretical frameworks from film theories, intercultural communication models, and multimodality, bridging digital humanities and artificial intelligence. This interdisciplinary approach allows for the validation and exploration of cultural models through large-scale data analysis. From a business perspective, The Möbius Trip LLC aims to serve diverse markets including academia, the film industry, social media influencers, and everyday users. By providing a scalable and comprehensive tool for film analysis, Möbius Trip Replay seeks to revolutionise how cultural and cinematic studies are conducted, fostering deeper insights and informed decision-making across various sectors.