-
Liquid Neural Network-based Adaptive Learning vs. Incremental Learning for Link Load Prediction amid Concept Drift due to Network Failures
Authors:
Omran Ayoub,
Davide Andreoletti,
Aleksandra Knapińska,
Róża Goścień,
Piotr Lechowicz,
Tiziano Leidi,
Silvia Giordano,
Cristina Rottondi,
Krzysztof Walkowiak
Abstract:
Adapting to concept drift is a challenging task in machine learning, which is usually tackled using incremental learning techniques that periodically re-fit a learning model leveraging newly available data. A primary limitation of these techniques is their reliance on substantial amounts of data for retraining. The necessity of acquiring fresh data introduces temporal delays prior to retraining, p…
▽ More
Adapting to concept drift is a challenging task in machine learning, which is usually tackled using incremental learning techniques that periodically re-fit a learning model leveraging newly available data. A primary limitation of these techniques is their reliance on substantial amounts of data for retraining. The necessity of acquiring fresh data introduces temporal delays prior to retraining, potentially rendering the models inaccurate if a sudden concept drift occurs in-between two consecutive retrainings. In communication networks, such issue emerges when performing traffic forecasting following a~failure event: post-failure re-routing may induce a drastic shift in distribution and pattern of traffic data, thus requiring a timely model adaptation. In this work, we address this challenge for the problem of traffic forecasting and propose an approach that exploits adaptive learning algorithms, namely, liquid neural networks, which are capable of self-adaptation to abrupt changes in data patterns without requiring any retraining. Through extensive simulations of failure scenarios, we compare the predictive performance of our proposed approach to that of a reference method based on incremental learning. Experimental results show that our proposed approach outperforms incremental learning-based methods in situations where the shifts in traffic patterns are drastic.
△ Less
Submitted 8 April, 2024;
originally announced April 2024.
-
Traffic-Aware Service Relocation in Cloud-Oriented Elastic Optical Networks
Authors:
Róża Goścień
Abstract:
In this paper, we study problem of efficient service relocation (i.e., changing assigned data center for a selected client node) in elastic optical networks (EONs) in order to increase network performance (measured by the volume of accepted traffic). To this end, we first propose novel traffic model for cloud ready transport networks. The model takes into account four flow types (i.e., city-to-cit…
▽ More
In this paper, we study problem of efficient service relocation (i.e., changing assigned data center for a selected client node) in elastic optical networks (EONs) in order to increase network performance (measured by the volume of accepted traffic). To this end, we first propose novel traffic model for cloud ready transport networks. The model takes into account four flow types (i.e., city-to-city, city-to-data center, data center-to-data center and data center-to-data center) while the flow characteristics are based on real economical and geographical parameters of the cities related to network nodes. Then, we propose dedicated flow allocation algorithm that can be supported by the service relocation process. We also introduce 21 different relocation policies, which use three types of data for decision making - network topological characteristics, rejection history and traffic prediction. Eventually, we perform extensive numerical experiments in order to: (i) tune proposed optimization approaches and (ii) evaluate and compare their efficiency and select the best one. The results of the investigation prove high efficiency of the proposed policies. The propoerly designed relocation policy allowed to allocate up to 3% more traffic (compared to the allocation without that policy). The results also reveal that the most efficient relocation policy bases its decisions on two types of data simultaneously - the rejection history and traffic prediction.
△ Less
Submitted 17 May, 2021;
originally announced May 2021.
-
On the Efficient Design of Network Resilient to Electro-Magnetic Pulse Attack -- Elastic Optical Network Case Study
Authors:
Róża Goścień
Abstract:
The telecommunication networks have become an indispensable part of our everyday life, providing support for such important areas as business, education, health care, finances, entertainment and social life. Alongside their continuous and uninterrupted operation is required while numerous new threats and attack scenarios emerge. The international security organisations warn against increasing like…
▽ More
The telecommunication networks have become an indispensable part of our everyday life, providing support for such important areas as business, education, health care, finances, entertainment and social life. Alongside their continuous and uninterrupted operation is required while numerous new threats and attack scenarios emerge. The international security organisations warn against increasing likelihood of nuclear weapon or electro-magnetic pulse (EMP) attacks, which can be extremely harmful also for transport networks. On that background, we study efficient design of network resilient to EMP attack wherein the required protection level is provided by the application of multipath routing and military grade bunkers (advanced electro-magnetic radiation resilient approaches protecting whole network node) implementation. Formally, we define and study problem of bunkers location, routing and spectrum allocation (BLRSA) in elastic optical network (EON). In the problem objective we address two criteria - network resilience (measured by the average lost flow per potential attack) and spectrum usage. For that problem we propose integer linear programming (ILP) model and two dedicated heuristics - 1S-RSA and 2S-RSA. Then, we perform extensive numerical experiments divided into three parts: (i) tuning of the proposed approaches, (ii) comparison with reference methods, (iii) realistic case study - efficient EMP-resilient network design. In the case study we analyze benefits and costs of the proposed protection scheme. Moreover, we also analyze vulnerabilities of three realistic network topologies to EMP attacks and identify their critical nodes. The investigation proves high efficiency of the proposed approaches and shows that they allow to save up to 90% of traffic lost in the case of no protection against these types of attacks.
△ Less
Submitted 30 March, 2021;
originally announced March 2021.
