« Back

Rappos Efstratios

Adjoint scientifique

Main skills

Distributed and Parallel Computing

Adjoint scientifique

Phone: +41 24 557 71 89

Desktop: C0-15.2

Haute école d'Ingénierie et de Gestion du Canton de Vaud
Route de Cheseaux 1, 1400 Yverdon-les-Bains, CH

Institute
IICT - Institut des Technologies de l'Information et de la Communication

BSc en Informatique et systèmes de communication - Haute école d'Ingénierie et de Gestion du Canton de Vaud

machine learning

Completed

Advanced analytics for fraud detection

AGP

Role: Collaborator

Requérant(e)s: IICT, Robert Stephan, IICT

Financement: CTI

Description du projet : Fraud detection and operational risk management are two main challenges faced by banks and financial institutions worldwide. They use several technological tools to analyse their data and detect fraudulent patterns and protect their assets. Most of these tools are rule-based and rely on human input. This project aims to enhance the product provided by NetGuardians SA to include a scientifically-backed component for fraud detection which is based on machine learning.

Research team within HES-SO: Rappos Efstratios , Robert Stephan , Houssou Zounyon Régis

Partenaires académiques: IICT; Robert Stephan, IICT

Durée du projet: 14.02.2017 - 15.05.2019

Montant global du projet: 348'410 CHF

Statut: Completed

CacHing Optimization In Cloud infrastructurE

AGP

Role: Collaborator

Requérant(e)s: hepia inIT, Abdennadher Nabil, hepia inIT

Financement: HES-SO Rectorat

Description du projet : Nous proposons une nouvelle méthode pour améliorer la performance du caching dans une infrastructure Cloud en faisant usage d'algorithmes prédictifs et adaptatifs où le contenu de la mémoire cache est déterminé par la solution d'un modèle d'optimisation stochastique. La méthode utilise les informations des requêtes effectuées et les coûts spécifiques du cloud. L'objectif est de produire un algorithme qui conduit à l'optimisation globale du coût des opérations.

Research team within HES-SO: Rappos Efstratios , Robert Stephan , Abdennadher Nabil

Partenaires académiques: IICT; hepia inIT; Abdennadher Nabil, hepia inIT

Durée du projet: 02.01.2013 - 31.07.2014

Montant global du projet: 154'000 CHF

Statut: Completed

Stockage intelligent de séries temporelles à l'aide de caching prédictif

AGP

Role: Collaborator

Requérant(e)s: IICT, Robert Stephan, IICT

Financement: HES-SO Rectorat

Description du projet : Nous proposons une nouvelle méthode, appelée PICASSO (PredICtive cAching framework for faSt temporal StOrage), pour augmenter les performances des solutions de stockage au niveau matériel et logiciel, pour de très grands volumes de séries temporelles. Nos algorithmes utilisent des modèles qui prévoient l'accès aux données et qui peuvent ainsi diminuer les coûts de stockage de manière drastique. La réduction des coûts va faire profiter à la fois les entreprises et les scientifiques qui ont besoin d'accéder à des séries temporelles de manière efficace.

Research team within HES-SO: Rappos Efstratios , Robert Stephan , Riedi Rudolf

Partenaires académiques: IICT; FR - EIA - Institut iSIS; Robert Stephan, IICT

Durée du projet: 01.09.2011 - 30.04.2013

Montant global du projet: 167'300 CHF

Statut: Completed

2024

Chain-structured neural architecture search for financial time series forecasting

Scientific paper ArODES

Denis Levchenko, Efstratios Rappos, Shabnam Ataee, Biagio Nigro, Stephan Robert-Nicoud

International Journal of Data Science and Analytics, 2024

Link to the publication

Summary:

Neural architecture search (NAS) emerged as a way to automatically optimize neural networks for a specific task and dataset. Despite an abundance of research on NAS for images and natural language applications, similar studies for time series data are lacking. Among NAS search spaces, chain-structured are the simplest and most applicable to small datasets like time series. We compare three popular NAS strategies on chain-structured search spaces: Bayesian optimization (specifically Tree-structured Parzen Estimator), the hyperband method, and reinforcement learning in the context of financial time series forecasting. These strategies were employed to optimize simple well-understood neural architectures like the MLP, 1D CNN, and RNN, with more complex temporal fusion transformers (TFT) and their own optimizers included for comparison. We find Bayesian optimization and the hyperband method performing best among the strategies, and RNN and 1D CNN best among the architectures, but all methods were very close to each other with a high variance due to the difficulty of working with financial datasets. We discuss our approach to overcome the variance and provide implementation recommendations for future users and researchers.

2022

A mixed-integer programming approach for solving university course timetabling problems

Scientific paper ArODES

Efstratios Rappos, Eric Thiémard, Stephan Robert, Jean-François Hêche

Journal of Scheduling, 2022, vol. 25, pp. 391–404

Link to the publication

Summary:

This article presents a mixed-integer programming model for solving the university timetabling problem which considers the allocation of students to classes and the assignment of rooms and time periods to each class. The model was developed as part of our participation in the International Timetabling Competition 2019 and produced a ranking of second place at the competition. Modeling a timetabling problem as a mixed-integer program is not new. Our contribution rests on a number of innovative features adapted to this problem which allow for a reduction in the number of variables and constraints of the mixed-integer program to manageable levels achieving a reasonable computational performance. The proposed algorithm consists of a first-stage method to obtain an initial feasible solution and a second-stage local search procedure to iteratively improve the solution value, both of which involve the optimization of mixed-integer programming problems.

A mixed-integer programming approach for solving university course timetabling problems

Scientific paper

Rappos Efstratios

Journal of Scheduling, 2022

Link to the publication

Summary:

2018

Bit.ly/practice :

Scientific paper ArODES

uncovering content publishing and sharing through URL shortening services

Daejin Choi, Jinyoung Han, Selin Chun, Efstratios Rappos, Stephan Robert, Ted Taekyoung Kwon

Telematics and Informatics, 2018, 35, 5, 1310-1323

Link to the publication

Summary:

It becomes the norm for people to share online content such as images, videos, and news over various channels including online social networks, news media, or online communities. One of the popular ways to publish and share online content is using a URL shortening service, which provides a short equivalent URL that is redirected to an original URL of content. This paper comprehensively analyze the practice of using short URLs from their creations to publishing to sharing, using a large scale dataset that contains 4.2 B requests for 80 M URLs created through Bit.ly, one of the most popular URL shortening services. We find that content URLs are mosunknown.

2016

Analyse prédictive de séries temporelles :

Professional paper ArODES

prédiction étendue à l'aide de l'apprentissage automatique

Mohamed Bibimoune, Serge Rigori, Lanpeng Ji, Efstratios Rappos, Stephan Robert

bulletin.ch = Fachzeitschrift und Verbandsinformationen von Electrosuisse und VSE = Bulletin SEV/AES : revue spécialisée et informations des associations Electrosuisse et AES, 2016, no. 10, pp. 41-44

Link to the publication

Summary:

La recherche dans le domaine de l'exploitation des données est un sujet qui gagne de l'importance. En effet, un nombre de plus en plus considérable de données sont à disposition, et ce, notamment à cause de l'initiative des données ouvertes (open data) et de l'avènement de l'Internet des objets. La manière de traiter ces données dans un but de prédiction est esquissée dans cet article.

2024

Optimized automated university timetabling with Covid-19 social distancing restrictions

Conference ArODES

Efstratios Rappos, Pier Donini

Proceedings of the 14th International Conference on the Practice and Theory of Automated Timetabling (PATAT 2024)

Link to the conference

Summary:

This extended abstract article describes the impact of the COVID 19 health restrictions in the teaching activities at the HEIG-VD university in Switzerland and the way that teaching activities could be maintained by using a custom timetabling algorithm incorporating the necessary social distancing health measures. The modified algorithm, using a mixed-integer program model, produced a new timetable and at the same time automatically selected the optimal mode of delivery for each lecture: remotely by video conference or physically in the classroom. The optimization model ensured that physical contact among students was minimized, while at the same time guaranteed that the courses or laboratories requiring physical presence were still able to take place

2018

A force-directed approach for offline GPS trajectory map matching

Conference ArODES

Efstratios Rappos, Stephan Robert, Philippe Cudré-Mauroux

Proceedings of the 26th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, 6-9 November 2018, Seattle, Washington, USA

Link to the conference

Summary:

We present a novel algorithm to match GPS trajectories onto maps offline (in batch mode) using techniques borrowed from the field of force-directed graph drawing. We consider a simulated physical system where each GPS trajectory is attracted or repelled by the underlying road network via electrical-like forces. We let the system evolve under the action of these physical forces such that individual trajectories are attracted towards candidate roads to obtain a map matching path. Our approach has several advantages compared to traditional, routing-based, algorithms for map matching, including the ability to account for noise and to avoid large detours due to outliers in the data whilst taking into account the underlying topological restrictions (such as one-way roads). Our empirical evaluation using real GPS traces shows that our method produces better map matching results compared to alternative offline map matching algorithms on average, especially for routes in dense, urban areas.

2013

A cloud data center optimization approach using dynamic data interchanges

Conference ArODES

Efstratios Rappos, Stephan Robert, Rudolf H. Riedi

Proceedings of 2013 IEEE 2nd International Conference on Cloud Networking (CloudNet), 11-13 November 2013, San Francisco, CA, USA

Link to the conference

Summary:

Distributed data center architectures have been recently developed for a more efficient and economical storage of data. In many models of distributed storage, the aim is to store the data in such a way so that the storage costs are minimized and increased redundancy requirements are maintained. However, many approaches do not fully consider issues relating to delivering the data to the end user and the associated costs that this creates. We present an integer programming optimization model for determining the optimal allocation of data components among a network of Cloud data servers in such a way that the total costs of additional storage, estimated data retrieval costs and network delay penalties is minimized. The method is suitable for periodic dynamic reconfiguration of the Cloud data servers, so that the when localized data request spikes occur the data can be moved to a closer or cheaper data server for cost reduction and increased efficiency.

Predictive caching in computer grids

Conference ArODES

Efstratios Rappos, Stephan Robert

Proceedings of 2013 13th IEEE/ACM International Symposium on Cluster, Cloud, and Grid Computing, 13-16 May 2013, Delft, Netherlands

Link to the conference

Summary:

We present a model for predictive caching where a shared cache is used to improve performance across a grid. Unlike local caching mechanisms, shared, grid or cloud-based caches incur high costs or latency associated with the additional data transfer. Our proposed caching model, which is dynamically optimized and constantly updated over time, determines the optimal allocation of objects into the shared cache, in such a way that the total cost or latency is minimized. This is achieved by including in the caching algorithm design measures of grid latency, data retrieval costs and a predictive component based on the probability of cached objects being requested in the near future.

Achievements

PEOPLE@HES-SO
Directory and Skills inventory

Rappos Efstratios

Adjoint scientifique

Main skills

Data Science

Data Analysis

Data Engineering

Distributed and Parallel Computing

Combinatorial Optimization

Machine Learning

Artificial Intelligence (AI)

Adjoint scientifique

PEOPLE@HES-SO Directory and Skills inventory

Rappos Efstratios

Adjoint scientifique

Main skills

Data Science

Data Analysis

Data Engineering

Distributed and Parallel Computing

Combinatorial Optimization

Machine Learning

Artificial Intelligence (AI)

Adjoint scientifique

PEOPLE@HES-SO
Directory and Skills inventory