Papers by Stojan Trajanovski
IEEE Transactions on Parallel and Distributed Systems, Apr 1, 2023
The efficacy of Network Function Virtualization (NFV) depends critically on (1) where the virtual... more The efficacy of Network Function Virtualization (NFV) depends critically on (1) where the virtual network functions (VNFs) are placed and (2) how the traffic is routed. Unfortunately, these aspects are not easily optimized, especially under time-varying network states with different QoS requirements. Given the importance of NFV, many approaches have been proposed to solve the VNF placement and Service Function Chaining (SFC) routing problem. However, those prior approaches mainly assume that the network state is static and known, disregarding dynamic network variations. To bridge that gap, we leverage Markov Decision Process (MDP) to model the dynamic network state transitions. To jointly minimize the delay and cost of NFV providers and maximize the revenue, we first devise a customized Deep Reinforcement Learning (DRL) algorithm for the VNF placement problem. The algorithm uses the attention mechanism to ascertain smooth network behavior within the general framework of network utility maximization (NUM). We then propose attention mechanism-based DRL algorithm for the SFC routing problem, which is to find the path to deliver traffic for the VNFs placed on different nodes. The simulation results show that our proposed algorithms outperform the state-of-the-art algorithms in terms of network utility, delay, cost, and acceptance ratio.
Social Science Research Network, 2022
arXiv (Cornell University), Aug 27, 2018
Smart devices of everyday use (such as smartphones and wearables) are increasingly integrated wit... more Smart devices of everyday use (such as smartphones and wearables) are increasingly integrated with sensors that provide immense amounts of information about a person's daily life such as behavior and context. The automatic and unobtrusive sensing of behavioral context can help develop solutions for assisted living, fitness tracking, sleep monitoring, and several other fields. Towards addressing this issue, we raise the question: can a machine learn to recognize a diverse set of contexts and activities in a real-life through joint learning from raw multi-modal signals (e.g. accelerometer, gyroscope and audio etc.)? In this paper, we propose a multi-stream temporal convolutional network to address the problem of multilabel behavioral context recognition. A four-stream network architecture handles learning from each modality with a contextualization module which incorporates extracted representations to infer a user's context. Our empirical evaluation suggests that a deep convolutional network trained end-to-end achieves an optimal recognition rate. Furthermore, the presented architecture can be extended to include similar sensors for performance improvements and handles missing modalities through multi-task learning without any manual feature engineering on highly imbalanced and sparsely labeled dataset.
2019 22th International Conference on Information Fusion (FUSION)
DOI to the publisher's website. • The final author version and the galley proof are versions of t... more DOI to the publisher's website. • The final author version and the galley proof are versions of the publication after peer review. • The final published version features the final layout of the paper including the volume, issue and page numbers. Link to publication General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal. If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the "Taverne" license above, please follow below link for the End User Agreement:
IEEE Transactions on Mobile Computing
In data communication networks, connection availability, which is defined as the probability that... more In data communication networks, connection availability, which is defined as the probability that the corresponding connection will be found in the operating state, is a key element of many Service Level Agreements (SLA). The path over which a connection is to be established should obey the agreed-upon availability, otherwise the service provider may face revenue loss as stipulated in the SLA. In this paper, we study the problem of establishing a connection over at most k (partially) link-disjoint paths for which the availability is no less than δ (0 < δ ≤ 1). We consider networks with and without Shared-Risk Link Groups (SRLGs). We prove that this problem, in general, cannot be approximated in polynomial time, unless P=NP. We subsequently propose a polynomial-time heuristic algorithm and an exact Integer Nonlinear Programming (INLP) formulation for availability-based path selection. Finally, the proposed algorithms and two existing heuristic algorithms are compared in terms of acceptance ratio and running time.
arXiv (Cornell University), May 6, 2010
We propose a new algorithm MARINLINGA for reverse line graph computation, i.e., constructing the ... more We propose a new algorithm MARINLINGA for reverse line graph computation, i.e., constructing the original graph from a given line graph. Based on the completely new and simpler principle of link relabeling and endnode recognition, MARINLINGA does not rely on Whitney's theorem while all previous algorithms do. MARINLINGA has a worst case complexity of O(N 2), where N denotes the number of nodes of the line graph. We demonstrate that MARINLINGA is more time-efficient compared to Roussopoulos's algorithm, which is well-known for its efficiency.
Adaptive Agents and Multi-Agents Systems, May 5, 2014
Shared effort games model people's contribution to projects and sharing the obtained profits. Tho... more Shared effort games model people's contribution to projects and sharing the obtained profits. Those games generalize both public projects like writing for Wikipedia, where everybody shares the resulting benefits, and all-pay auctions such as contests and political campaigns, where only the winner obtains a profit. In θ-equal sharing (effort) games, a threshold for effort defines which contributors win and then receive their (equal) share. (For public projects θ = 0 and for all-pay auctions θ = 1.) Thresholds between 0 and 1 can model games such as paper co-authorship and shared homework assignments. First, we fully characterize the conditions for the existence of a pure-strategy Nash equilibrium for two-player shared effort games with close budgets and project value functions that are linear on the received contribution and prove some efficiency results. Second, since the theory does not work for more players, fictitious play simulations are used to show when such an equilibrium exists and what its efficiency is. The results about existence and efficiency of these equilibria provide the likely strategy profiles and the socially preferred strategies to use in real life situations of contribution to public projects.
IEEE Transactions on Biomedical Engineering, 2021
The utilization of hyperspectral imaging (HSI) in real-time tumor segmentation during a surgery h... more The utilization of hyperspectral imaging (HSI) in real-time tumor segmentation during a surgery have recently received much attention, but it remains a very challenging task. Methods: In this work, we propose semantic segmentation methods and compare them with other relevant deep learning algorithms for tongue tumor segmentation. To the best of our knowledge, this is the first work using deep learning semantic segmentation for tumor detection in HSI data using channel selection and accounting for more spatial tissue context and global comparison between the prediction map and the annotation per sample. Results and Conclusion: On a clinical data set with tongue squamous cell carcinoma, our best method obtains very strong results of average dice coefficient and area under the ROC-curve of 0.891 ± 0.053 and 0.924 ± 0.036, respectively on the original spatial image size. The results show that a very good performance can be achieved even with a limited amount of data. We demonstrate that important information regarding tumor decision is encoded in various channels, but some channel selection and filtering is beneficial over the full spectra. Moreover, we use both visual (VIS) and near-infrared (NIR) spectrum, rather than commonly used only VIS spectrum; although VIS spectrum is generally of higher significance, we demonstrate NIR spectrum is crucial for tumor capturing in some cases. Significance: The HSI technology augmented with accurate deep learning algorithms has a huge potential to be a promising alternative to digital pathology or a doctors' supportive tool in real-time surgeries.
Optical Fiber Communication Conference, 2017
Secure Autonomous Response NETworks (SARNET) is a framework for automated response against attack... more Secure Autonomous Response NETworks (SARNET) is a framework for automated response against attacks on computer network infrastructures. The framework addresses several cyber-security problems at three crucial levels: strategic, tactical and operational.
Multilevel Strategic Interaction Game Models for Complex Networks, 2019
We have so far concentrated on networks, which do not change over time. In reality, a network may... more We have so far concentrated on networks, which do not change over time. In reality, a network may change over time in an independent process from the epidemic spread. Such networks, where the topology changes according to some rule or pattern, are known as evolving networks. The epidemic threshold in evolving networks has been studied in the past [214, 280]. Adaptive networks possess more complex properties than evolving networks, such that the topology is modified based on epidemic processes. An adaptive model over the standard SIS model has been considered by Gross et al. [116]. This model is based on fixed probability of an infected nodes to infect and incident susceptible node and a fixed recovery probability of an infected node. Similarly, a link between a susceptible and an infected node is broken with a fixed probability and subsequently, a connection is established between the susceptible node and another susceptible node at random, which is an example of a rewiring process....
Todor Minchev (one of following 3 papers or combined any of them) 1.1 A. Carzaniga, D.S. Rosenblu... more Todor Minchev (one of following 3 papers or combined any of them) 1.1 A. Carzaniga, D.S. Rosenblum, A.L. Wolf: Achieving scalability and expressiveness in an internet-scale event notification service, PODC, 2001. 1.2. A. Carzaniga, M.J. Rutherford, A.L. Wolf: A Routing Scheme for Content-Based Networking, INFOCOM, 2004. 1.3. A. Carzaniga, A.L. Wolf: Forwarding in a content-based network, SIGCOMM, 2003.
Consider mitigating the effects of denial of service or of malicious traffic in networks by delet... more Consider mitigating the effects of denial of service or of malicious traffic in networks by deleting edges. Edge deletion reduces the DoS or the number of the malicious flows, but it also inadvertently removes some of the desired flows. To model this important problem, we formulate two problems: (1) remove all the undesirable flows while minimizing the damage to the desirable ones and (2) balance removing the undesirable flows and not removing too many of the desirable flows. We prove these problems are equivalent to important theoretical problems, thereby being important not only practically but also theoretically, and very hard to approximate in a general network. We employ reductions to nonetheless approximate the problem and also provide a greedy approximation. When the network is a tree, the problems are still MAX SNP-hard, but we provide a greedy-based 2l-approximation algorithm, where l is the longest desirable flow. We also provide an algorithm, approximating the first and t...
We study the problem of fully mitigating the effects of denial of service by filtering the minimu... more We study the problem of fully mitigating the effects of denial of service by filtering the minimum necessary set of the undesirable flows. First, we model this problem and then we concentrate on a subproblem where every good flow has a bottleneck. We prove that unless \(\text {P}= \text {NP}\), this subproblem is inapproximable within factor \(2^{\log ^{1 - 1/\log \log ^c (n)}(n)}\), for \(n = \left| E \right| + \left| GF \right| \) and any \(c < 0.5\). We provide a \(b (k + 1)\)-factor polynomial approximation, where k bounds the number of the desirable flows that a desirable flow intersects, and b bounds the number of the undesirable flows that can intersect a desirable one at a given edge. Our algorithm uses the local ratio technique.
ArXiv, 2017
Stress can be seen as a physiological response to everyday emotional, mental and physical challen... more Stress can be seen as a physiological response to everyday emotional, mental and physical challenges. A long-term exposure to stressful situations can have negative health consequences, such as increased risk of cardiovascular diseases and immune system disorder. Therefore, a timely stress detection can lead to systems for better management and prevention in future circumstances. In this paper, we suggest a multi-task learning based neural network approach (with hard parameter sharing of mutual representation and task-specific layers) for personalized stress recognition using skin conductance and heart rate from wearable devices. The proposed method is tested on multi-modal physiological responses collected during real-world and simulator driving tasks.
Real-time feedback based on hyperspectral images (HSI) to a surgeon can lead to a higher precisio... more Real-time feedback based on hyperspectral images (HSI) to a surgeon can lead to a higher precision and additional insights compared to the standard techniques. To the best of our knowledge, deep learning with semantic segmentation utilizing both visual (VIS) and infrared channels (NIR) has never been exploited with the HSI data with human tumors. We propose using channels selection with U-Net deep neural network for tumor segmentation in hyperspectral images. The proposed method, based on bigger patches, accounts for bigger spatial context and achieves better results (average dice coefficient 0.89 ± 0.07 and area under the ROC-curve AUC 0.93 ± 0.04) than pixel-level spectral and structural approaches in a clinical data set with tongue squamous cell carcinoma. The importance of VIS channel for the performance is higher, but NIR contribution is non-negligible.
2018 IEEE 5th International Conference on Data Science and Advanced Analytics (DSAA), 2018
Stress and accompanying physiological responses can occur when everyday emotional, mental and phy... more Stress and accompanying physiological responses can occur when everyday emotional, mental and physical challenges exceed one's ability to cope. A long-term exposure to stressful situations can have negative health consequences, such as increased risk of cardiovascular diseases and immune system disorder. It is also shown to adversely affect productivity, well-being, and self-confidence, which can lead to social and economic inequality. Hence, a timely stress recognition can contribute to better strategies for its management and prevention in the future. Stress can be detected from multimodal physiological signals (e.g. skin conductance and heart rate) using well-trained models. However, these models need to be adapted to a new target domain and personalized for each test subject. In this paper, we propose a deep reconstruction classification network and multi-task learning (MTL) for domain adaption and personalization of stress recognition models. The domain adaption is achieved via a hybrid model consisting of temporal convolutional and recurrent layers that perform shared feature extraction through supervised source label predictions and unsupervised target data reconstruction. Furthermore, MTL based neural network approach with hard parameter sharing of mutual representation and task-specific layers is utilized to acquire personalized models. The proposed methods are tested on multimodal physiological time-series data collected during driving tasks, in both real-world and driving simulator settings.
2017 IEEE International Conference on Data Mining Workshops (ICDMW), 2017
Driving is an activity that requires considerable alertness. Insufficient attention, imperfect pe... more Driving is an activity that requires considerable alertness. Insufficient attention, imperfect perception, inadequate information processing, and sub-optimal arousal are possible causes of poor human performance. Understanding of these causes and the implementation of effective remedies is of key importance to increase traffic safety and improve driver's wellbeing. For this purpose, we used deep learning algorithms to detect arousal level, namely, under-aroused, normal and overaroused for professional truck drivers in a simulated environment. The physiological signals are collected from 11 participants by wrist wearable devices. We presented a cost effective groundtruth generation scheme for arousal based on a subjective measure of sleepiness and score of stress stimuli. On this dataset, we evaluated a range of deep neural network models for representation learning as an alternative to handcrafted feature extraction. Our results show that a 7-layers convolutional neural network trained on raw physiological signals (such as heart rate, skin conductance and skin temperature) outperforms a baseline neural network and denoising autoencoder models with weighted F-score of 0.82 vs. 0.75 and Kappa of 0.64 vs. 0.53, respectively. The proposed convolutional model not only improves the overall results but also enhances the detection rate for every driver in the dataset as determined by leave-one-subject-out cross-validation.
Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Industry Papers, 2021
Email and chat communication tools are increasingly important for completing daily tasks. Accurat... more Email and chat communication tools are increasingly important for completing daily tasks. Accurate real-time phrase completion can save time and bolster productivity. Modern text prediction algorithms are based on large language models which typically rely on the prior words in a message to predict a completion. We examine how additional contextual signals (from previous messages, time, and subject) affect the performance of a commercial text prediction model. We compare contextual text prediction in chat and email messages from two of the largest commercial platforms Microsoft Teams and Outlook, finding that contextual signals contribute to performance differently between these scenarios. On emails, time context is most beneficial with small relative gains of 2% over baseline. Whereas, in chat scenarios, using a tailored set of previous messages as context yields relative improvements over the baseline between 9.3% and 18.6% across various critical serviceoriented text prediction metrics.
International Journal of Forecasting, 2020
Researchers from various scientific disciplines have attempted to forecast the spread of coronavi... more Researchers from various scientific disciplines have attempted to forecast the spread of coronavirus disease 2019 (COVID-19). The proposed epidemic prediction methods range from basic curve fitting methods and traffic interaction models to machine-learning approaches. If we combine all these approaches, we obtain the Network Inferencebased Prediction Algorithm (NIPA). In this paper, we analyse a diverse set of COVID-19 forecast algorithms, including several modifications of NIPA. Among the algorithms that we evaluated, the original NIPA performed best at forecasting the spread of COVID-19 in Hubei, China and in the Netherlands. In particular, we show that network-based forecasting is superior to any other forecasting algorithm.
Uploads
Papers by Stojan Trajanovski