Change search
Refine search result
1 - 26 of 26
CiteExportLink to result list
Permanent link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Rows per page
  • 5
  • 10
  • 20
  • 50
  • 100
  • 250
Sort
  • Standard (Relevance)
  • Author A-Ö
  • Author Ö-A
  • Title A-Ö
  • Title Ö-A
  • Publication type A-Ö
  • Publication type Ö-A
  • Issued (Oldest first)
  • Issued (Newest first)
  • Created (Oldest first)
  • Created (Newest first)
  • Last updated (Oldest first)
  • Last updated (Newest first)
  • Disputation date (earliest first)
  • Disputation date (latest first)
  • Standard (Relevance)
  • Author A-Ö
  • Author Ö-A
  • Title A-Ö
  • Title Ö-A
  • Publication type A-Ö
  • Publication type Ö-A
  • Issued (Oldest first)
  • Issued (Newest first)
  • Created (Oldest first)
  • Created (Newest first)
  • Last updated (Oldest first)
  • Last updated (Newest first)
  • Disputation date (earliest first)
  • Disputation date (latest first)
Select
The maximal number of hits you can export is 250. When you want to export more records please use the Create feeds function.
  • 1.
    Alistarh, Dan
    et al.
    IST Austria, Klosterneuburg, Austria..
    Hoefler, Torsten
    Swiss Fed Inst Technol, Zurich, Switzerland..
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Khirirat, Sarit
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Konstantinov, Nikola
    IST Austria, Klosterneuburg, Austria..
    Renggli, Cedric
    Swiss Fed Inst Technol, Zurich, Switzerland..
    The Convergence of Sparsified Gradient Methods2018In: Advances in Neural Information Processing Systems 31 (NIPS 2018) / [ed] Bengio, S Wallach, H Larochelle, H Grauman, K CesaBianchi, N Garnett, R, Neural Information Processing Systems (NIPS) , 2018, Vol. 31Conference paper (Refereed)
    Abstract [en]

    Stochastic Gradient Descent (SGD) has become the standard tool for distributed training of massive machine learning models, in particular deep neural networks. Several families of communication-reduction methods, such as quantization, large-batch methods, and gradient sparsification, have been proposed to reduce the overheads of distribution. To date, gradient sparsification methods-where each node sorts gradients by magnitude, and only communicates a subset of the components, accumulating the rest locally-are known to yield some of the largest practical gains. Such methods can reduce the amount of communication per step by up to three orders of magnitude, while preserving model accuracy. Yet, this family of methods currently has no theoretical justification. This is the question we address in this paper. We prove that, under analytic assumptions, sparsifying gradients by magnitude with local error correction provides convergence guarantees, for both convex and non-convex smooth objectives, for data-parallel SGD. The main insight is that sparsification methods implicitly maintain bounds on the maximum impact of stale updates, thanks to selection by magnitude. Our analysis also reveals that these methods do require analytical conditions to converge well, justifying and complementing existing heuristics.

  • 2.
    Aytekin, Arda
    et al.
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Exploiting serverless runtimes for large-scale optimization2019In: 2019 IEEE 12th International Conference on Cloud Computing (CLOUD), IEEE Computer Society, 2019, p. 499-501, article id 8814497Conference paper (Refereed)
    Abstract [en]

    Serverless runtimes provide efficient and cost-effective environments for scalable computations, thanks to their event-driven and elastic nature. So far, they have mostly been used for stateless, data parallel and sporadic computations. In this work, we propose exploiting serverless runtimes to solve generic, large-scale optimization problems. To this end, we implement a parallel optimization algorithm for solving a regularized logistic regression problem, and use AWS Lambda for the compute-intensive work. We show that relative speedups up to 256 workers and efficiencies above 70% up to 64 workers can be expected.

  • 3.
    Biel, Martin
    et al.
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Decision and Control Systems (Automatic Control).
    Aytekin, Arda
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Decision and Control Systems (Automatic Control).
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Decision and Control Systems (Automatic Control).
    POLO.Jl: Policy-based optimization algorithms in Julia2019In: Advances in Engineering Software, ISSN 0965-9978, E-ISSN 1873-5339, Vol. 136, article id 102695Article in journal (Refereed)
    Abstract [en]

    We present POLO. j1- a Julia package that helps algorithm developers and machine-learning practitioners design and use state-of-the-art parallel optimization algorithms in a flexible and efficient way. POLO. j1 extends our C+ + library POLO, which has been designed and implemented with the same intentions. POLO. j1 not only wraps selected algorithms in POLO and provides an easy mechanism to use data manipulation facilities and loss function definitions in Julia together with the underlying compiled C+ + library, but it also uses the policy-based design technique in a Julian way to help users prototype optimization algorithms from their own building blocks. In our experiments, we observe that there is little overhead when using the compiled C+ + code directly within Julia. We also notice that the performance of algorithms implemented in pure Julia is comparable with that of their C+ + counterparts. Both libraries are hosted on GitHub(1)under the free MIT license, and can be used easily by pulling the pre-built 64-bit architecture Docker images.(2)

  • 4. Carlsson, Mats
    et al.
    Johansson, Mikael
    KTH, School of Electrical Engineering (EES), Automatic Control.
    Larson, Jeffrey
    Scheduling double round-robin tournaments with divisional play using constraint programming2017In: European Journal of Operational Research, ISSN 0377-2217, E-ISSN 1872-6860, Vol. 259, no 3, p. 1180-1190Article in journal (Refereed)
    Abstract [en]

    We study a tournament format that extends a traditional double round-robin format with divisional single round-robin tournaments. Elitserien, the top Swedish handball league, uses such a format for its league schedule. We present a constraint programming model that characterizes the general double round-robin plus divisional single round-robin format. This integrated model allows scheduling to be performed in a single step, as opposed to common multistep approaches that decompose scheduling into smaller problems and possibly miss optimal solutions. In addition to general constraints, we introduce Elitserien-specific requirements for its tournament. These general and league-specific constraints allow us to identify implicit and symmetry-breaking properties that reduce the time to solution from hours to seconds. A scalability study of the number of teams shows that our approach is reasonably fast for even larger league sizes. The experimental evaluation of the integrated approach takes considerably less computational effort to schedule Elitserien than does the previous decomposed approach.

  • 5.
    Charalambous, Themistoklis
    et al.
    Aalto Univ, Dept Elect Engn & Automat, Espoo, Finland..
    Kim, Su Min
    Korea Polytech Univ, Dept Elect Engn, Shihung, South Korea..
    Nomikos, Nikolaos
    Univ Aegean, Dept Informat & Commun Syst Engn, Samos, Greece..
    Bengtsson, Mats
    KTH, School of Electrical Engineering and Computer Science (EECS), Information Science and Engineering.
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Relay-pair selection in buffer-aided successive opportunistic relaying using a multi-antenna source2019In: Ad hoc networks, ISSN 1570-8705, E-ISSN 1570-8713, Vol. 84, p. 29-41Article in journal (Refereed)
    Abstract [en]

    We study a cooperative network with a buffer-aided multi-antenna source, multiple half-duplex (HD) buffer-aided relays and a single destination. Such a setup could represent a cellular downlink scenario, in which the source can be a more powerful wireless device with a buffer and multiple antennas, while a set of intermediate less powerful devices are used as relays to reach the destination. The main target is to recover the multiplexing loss of the network by having the source and a relay to simultaneously transmit their information to another relay and the destination, respectively. Successive transmissions in such a cooperative network, however, cause inter-relay interference (IRI). First, by assuming global channel state information (CSI), we show that the detrimental effect of IRI can be alleviated by precoding at the source, mitigating or even fully cancelling the interference. A cooperative relaying policy is proposed that employs a joint precoding design and relay-pair selection. Note that both fixed rate and adaptive rate transmissions can be considered. For the case when channel state information is only available at the receiver side (CSIR), we propose a relay selection policy that employs a phase alignment technique to reduce the IRI. The performance of the two proposed relay pair selection policies are evaluated and compared with other state-of-the-art relaying schemes in terms of outage and throughput. The results show that the use of a powerful source can provide considerable performance improvements.

  • 6.
    Della Penda, Demia
    et al.
    Ericsson AB, S-16480 Kista, Sweden..
    Abrardo, Andrea
    Univ Siena, Dept Informat Engn, I-53100 Siena, Italy.;Consorzio Nazl Interuniv Telecomunicaz, I-43124 Parma, Italy..
    Moretti, Marco
    Consorzio Nazl Interuniv Telecomunicaz, I-43124 Parma, Italy.;Univ Pisa, Dept Informat Engn, I-56122 Pisa, Italy..
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Distributed Channel Allocation for D2D-Enabled 5G Networks Using Potential Games2019In: IEEE Access, E-ISSN 2169-3536, Vol. 7, p. 11195-11208Article in journal (Refereed)
    Abstract [en]

    Frequency channel allocation is a key technique for improving the performance of cellular networks. In this paper, we address the channel allocation problem for a 5G multi-cell system. We consider a heterogeneous network in which cellular users, micro-cell users, and device-to-device (D2D) communications coexist within the radio footprint of the macro cell. We maximize the aggregate transmission rate, exploiting channel diversity and managing both the inter-cell interference, typical of cellular networks and the intra-cell interference generated by the nonorthogonal transmissions of the small-cell and D2D users. By modeling the allocation problem as a potential game, whose Nash equilibria correspond to the local optima of the objective function, we propose a new decentralized solution. The convergence of our scheme is enforced by using a better response dynamic based on a message passing approach. The simulation results assess the validity of the proposed scheme in terms of convergence time and achievable rate under different settings.

  • 7.
    Della Penda, Demia
    et al.
    KTH, School of Electrical Engineering (EES), Automatic Control.
    Nomikos, Nikolaos
    Charalambous, Themistoklis
    Johansson, Mikael
    KTH, School of Electrical Engineering (EES), Automatic Control.
    Minimum Power Scheduling under Rician Fading in Full-Duplex Relay-Assisted D2D Communication2017In: 2017 IEEE Globecom Workshops, GC Wkshps 2017 - Proceedings, IEEE , 2017Conference paper (Refereed)
    Abstract [en]

    In cellular systems, the combination of Device-to-Device (D2D) communication and relaying is an efficient means for improving network coverage and transmissions quality without additional infrastructure deployment. It enables communication between user pairs in situations when both their direct D2D transmission and the traditional communication via the base station experience poor channel quality. In this paper, we propose a joint relaying-operation selection and power-allocation scheme, herein called HyD2D, for relay-assisted D2D communication in Rician fading environment. The target is to choose the set of communication links that minimizes the power consumption, while ensuring a minimum success probability. To overcome the nonconvexity of the outage probability constraints under Rician fading, we use the concept of coherent-measure-of-risk from the field of finance. We therefore obtain a linear programming formulation that we can efficiently solve. Simulations show that HyD2D selects the most energy-efficient relaying operation that satisfies the success probability requirement, while leveraging only statistical channel state information.

  • 8.
    Della Penda, Demia
    et al.
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Decision and Control Systems (Automatic Control). Ericsson AB, Stockholm, 164 83, Sweden.
    Wichman, Risto
    Aalto Univ, Sch Elect Engn, FI-00076 Aalto, Finland..
    Charalambous, Themistoklis
    Aalto Univ, Sch Elect Engn, FI-00076 Aalto, Finland..
    Fodor, Gabor
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Decision and Control Systems (Automatic Control). Ericsson AB, Stockholm, 164 83, Sweden.
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Decision and Control Systems (Automatic Control).
    A Distributed Mode Selection Scheme for Full-Duplex Device-to-Device Communication2019In: IEEE Transactions on Vehicular Technology, ISSN 0018-9545, E-ISSN 1939-9359, Vol. 68, no 10, p. 10267-10271Article in journal (Refereed)
    Abstract [en]

    Networks with device-to-device(D2D) technology allow for two possible communication modes: traditional communication via the base station, and direct communication between the users. Recent studies show that in-band full-duplex(IBFD) operations can be advantageously combined with D2D communication to improve the spectral efficiency. However, no algorithms for selecting the communication mode of mobile users in IBFD networks have yet appeared in the literature. In this paper, we design a distributed mode selection scheme for users in D2D-enabled IBFD networks. The proposed scheme maximizes the users prob-ability of successful communication by leveraging only existing signaling mechanisms.

  • 9.
    Demirel, Burak
    et al.
    KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    Aytekin, Arda
    KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    Quevedo, Daniel E.
    Univ Newcastle, Sch Elect Engn & Comp Sci, Callaghan, NSW 2308, Australia..
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control. KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    To wait or to drop: on the optimal number of retransmissions in wireless control2015In: 2015 EUROPEAN CONTROL CONFERENCE (ECC), IEEE , 2015, p. 962-968Conference paper (Refereed)
    Abstract [en]

    The dimensioning of wireless communication protocols for networked control involves a non-trivial trade-off between reliability and delay. Due to the lossy nature of wireless communications, there is a risk that sensor messages will be dropped. The end-to-end reliability can be improved by retransmitting dropped messages, but this comes at the expense of additional delays. In this work, we determine the number of retransmissions that strikes the optimal balance between communication reliability and delay, in the sense that it achieves the minimal expected linear-quadratic loss of the closed-loop system. An important feature of our setup is that it accounts for the random delays and possible losses that occur when unreliable communication is combatted with retransmissions. The resulting controller dynamically switches among a set of infinite-horizon linear-quadratic regulators, and is simple to implement. Numerical simulations are carried out to highlight the trade-off between reliability and delay.

  • 10.
    Demirel, Burak
    et al.
    KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre. University of Paderborn, Germany.
    Gupta, V.
    Quevedo, D. E.
    Johansson, Mikael
    KTH, School of Electrical Engineering (EES), Automatic Control. KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    On the Trade-Off between Communication and Control Cost in Event-Triggered Dead-Beat Control2016In: IEEE Transactions on Automatic Control, ISSN 0018-9286, E-ISSN 1558-2523, Vol. PP, no 99, article id 7562376Article in journal (Refereed)
    Abstract [en]

    We consider a stochastic system where the communication between the controller and the actuator is triggered by a thresholdbased rule. The communication is performed across an unreliable link that stochastically erases transmitted packets. To decrease the communication burden, and as a partial protection against dropped packets, the controller sends a sequence of control commands to the actuator in each packet. These commands are stored in a buffer and applied sequentially until the next control packet arrives. In this context, we study dead-beat control laws and compute the expected linear-quadratic loss of the closed-loop system for any given event-threshold. Furthermore, we provide analytical expressions that quantify the trade-off between the communication cost and the control performance of event-triggered control systems. Numerical examples demonstrate the effectiveness of the proposed technique.

  • 11.
    Di Benedetto, Maria Domenica
    et al.
    Univ Aquila, Coll Engn, Ctr Excellence DEWS, I-67100 Laquila, Italy..
    Johansson, Karl H.
    KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    Johansson, Mikael
    KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    Santucci, Fortunato
    Univ Aquila, Coll Engn, Ctr Excellence DEWS, I-67100 Laquila, Italy..
    Industrial control over wireless networks2010In: International Journal of Robust and Nonlinear Control, ISSN 1049-8923, E-ISSN 1099-1239, Vol. 20, no 2, p. 119-122Article in journal (Other academic)
  • 12.
    Ghadimi, Euhanna
    et al.
    KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    Teixeira, André
    KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    Rabbat, Michael G.
    McGill Univ, Dept Elect & Comp Engn, Montreal, PQ, Canada..
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control. KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    The ADMM algorithm for distributed averaging: Convergence rates and optimal parameter selection2014In: CONFERENCE RECORD OF THE 2014 FORTY-EIGHTH ASILOMAR CONFERENCE ON SIGNALS, SYSTEMS & COMPUTERS / [ed] Matthews, M B, IEEE COMPUTER SOC , 2014, p. 783-787Conference paper (Refereed)
    Abstract [en]

    We derive the optimal step-size and over-relaxation parameter that minimizes the convergence time of two ADMM-based algorithms for distributed averaging. Our study shows that the convergence times for given step-size and over-relaxation parameters depend on the spectral properties of the normalized Laplacian of the underlying communication graph. Motivated by this, we optimize the edge-weights of the communication graph to improve the convergence speed even further. The performance of the ADMM algorithms with our parameter selection are compared with alternatives from the literature in extensive numerical simulations on random graphs.

  • 13.
    Gunnar, Anders
    et al.
    KTH, School of Electrical Engineering (EES).
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Robust load-balancing under statistical uncertainty: models and polynomial-time algorithms2009In: 2009 NEXT GENERATION INTERNET NETWORKS, IEEE , 2009, p. 99-+Conference paper (Refereed)
    Abstract [en]

    We study the problem of guaranteed-performance routing under statistical traffic uncertainty. Relevant traffic models are presented and a polynomial-time algorithm for solving the associated robust routing problem is given. We demonstrate how our techniques, in combination with fundamental limitations on the accuracy of estimated traffic matrices, enable us to compute bounds on the achievable performance of OSPF-routing optimized using only topology information and link count data. We discuss extensions to other types of traffic uncertainties and describe an alternative, more memory efficient, algorithm based on combined constraint and column generation. The proposed techniques are evaluated in several numerical examples to highlight the features of our approach.

  • 14.
    Khirirat, Sarit
    et al.
    KTH, School of Electrical Engineering (EES), Automatic Control. KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    Feyzmahdavian, Hamid Reza
    Johansson, Mikael
    KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre. KTH, School of Electrical Engineering (EES), Automatic Control.
    Mini-batch gradient descent: faster convergence under data sparsity2017In: 2017 IEEE 56TH ANNUAL CONFERENCE ON DECISION AND CONTROL (CDC), IEEE , 2017Conference paper (Refereed)
    Abstract [en]

    The practical performance of stochastic gradient descent on large-scale machine learning tasks is often much better than what current theoretical tools can guarantee. This indicates that there is an inherent structure in these problems that could be exploited to strengthen the analysis. In this paper, we argue that data sparsity is such a property. We derive explicit expressions for how data sparsity affects the range of admissible step-sizes and the convergence factors of mini-batch gradient descent. Our theoretical results are validated by solving least-squares support vector machine problems on both synthetic and real-life data sets. The experimental results demonstrate improved performance of our update rules compared to the traditional mini-batch gradient descent algorithm.

  • 15.
    Khirirat, Sarit
    et al.
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control. KTH, School of Electrical Engineering and Computer Science (EECS), Centres, ACCESS Linnaeus Centre.
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control. KTH, School of Electrical Engineering and Computer Science (EECS), Centres, ACCESS Linnaeus Centre.
    Alistarh, Dan
    IST Austria, Vienna, Austria..
    Gradient compression for communication-limited convex optimization2018In: 2018 IEEE Conference on Decision and Control (CDC), Institute of Electrical and Electronics Engineers (IEEE), 2018, p. 166-171, article id 8619625Conference paper (Refereed)
    Abstract [en]

    Data-rich applications in machine-learning and control have motivated an intense research on large-scale optimization. Novel algorithms have been proposed and shown to have optimal convergence rates in terms of iteration counts. However, their practical performance is severely degraded by the cost of exchanging high-dimensional gradient vectors between computing nodes. Several gradient compression heuristics have recently been proposed to reduce communications, but few theoretical results exist that quantify how they impact algorithm convergence. This paper establishes and strengthens the convergence guarantees for gradient descent under a family of gradient compression techniques. For convex optimization problems, we derive admissible step sizes and quantify both the number of iterations and the number of bits that need to be exchanged to reach a target accuracy. Finally, we validate the performance of different gradient compression techniques in simulations. The numerical results highlight the properties of different gradient compression algorithms and confirm that fast convergence with limited information exchange is possible.

  • 16.
    Khirirat, Sarit
    et al.
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Magnusson, Sindri
    Harvard Univ, Sch Engn & Appl Sci, 33 Oxford St, Cambridge, MA 02138 USA..
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    CONVERGENCE BOUNDS FOR COMPRESSED GRADIENT METHODS WITH MEMORY BASED ERROR COMPENSATION2019In: 2019 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP) / [ed] RTSEKAS D. P., 2011, Optimization for Machine Learning, V2010, P1 ngni J., 2017, arXiv preprint arXiv: 1710. 09854, u Shengyu, 2016, 2016 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING PROCEEDINGSIEEE International Conference on Acoustics, Speech, and Signal Processing, MAR 20-25, 2016, Shanghai, PEOPLES R CHINA, P4134, IEEE , 2019, p. 2857-2861Conference paper (Refereed)
    Abstract [en]

    The veritable scale of modern data necessitates information compression in parallel/distributed big-data optimization. Compression schemes using memory-based error compensation have displayed superior performance in practice, however, to date there are no theoretical explanations for these observed advantages. This paper provides the first theoretical support for why such compression schemes yields higher accuracy solutions in optimization. Our results cover both gradient and incremental gradient algorithms for quadratic optimization. Unlike previous works, our theoretical results explicitly quantify the accuracy gains from error compensation, especially for ill-conditioned problems. Finally, the numerical results on linear least-squares problems validate the benefit of error compensation and demonstrate tightness of our convergence guarantees.

  • 17.
    Mai, Vien V.
    et al.
    KTH, School of Electrical Engineering (EES), Automatic Control. KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    Johansson, Mikael
    KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre. KTH, School of Electrical Engineering (EES), Automatic Control.
    Lock-Free Incremental Coordinate Descent2017In: 2017 IEEE 56th Annual Conference on Decision and Control, CDC 2017, Institute of Electrical and Electronics Engineers (IEEE), 2017Conference paper (Refereed)
    Abstract [en]

    We study a flexible algorithm for minimizing a sum of component functions, each of which depends on a large number of decision variables. The algorithm combines aspects of incremental gradient method with that of coordinate descent. In contrast to earlier algorithms of this kind, our algorithm is lock-free and does not require synchronization of access to the shared memory. We prove convergence of the algorithm under asynchronous operation and provide explicit bounds on how the solution times depend on the degree of asynchrony. Numerical experiments confirm our theoretical results.

  • 18.
    Mai, Vien V.
    et al.
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Noisy Accelerated Power Method for Eigenproblems With Applications2019In: IEEE Transactions on Signal Processing, ISSN 1053-587X, E-ISSN 1941-0476, Vol. 67, no 12, p. 3287-3299Article in journal (Refereed)
    Abstract [en]

    This paper introduces an efficient algorithm for finding the dominant generalized eigenvectors of a pair of symmetric matrices. Combining tools from approximation theory and convex optimization, we develop a simple scalable algorithm with strong theoretical performance guarantees. More precisely, the algorithm retains the simplicity of the well-knownpower method but enjoys the asymptotic iteration complexity of the powerful Lanczos method. Unlike these classic techniques, our algorithm is designed to decompose the overall problem into a series of subproblems that only need to be solved approximately. The combination of good initializations, fast iterative solvers, and appropriate error control in solving the subproblems lead to a linear running time in the input sizes compared to the superlinear time for the traditional methods. The improved running time immediately offers acceleration for several applications. As an example, we demonstrate how the proposed algorithm can be used to accelerate canonical correlation analysis, which is a fundamental statistical tool for learning of a low-dimensional representation of high-dimensional objects. Numerical experiments on real-world datasets confirm that our approach yields significant improvements over the current state of the art.

  • 19.
    Mai, Vien V.
    et al.
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    NONLINEAR ACCELERATION OF CONSTRAINED OPTIMIZATION ALGORITHMS2019In: 2019 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP), IEEE , 2019, p. 4903-4907Conference paper (Refereed)
    Abstract [en]

    This paper introduces a novel technique for nonlinear acceleration of first-order methods for constrained convex optimization. Previous studies of nonlinear acceleration have only been able to provide convergence guarantees for unconstrained convex optimization. In contrast, our method is able to avoid infeasibility of the accelerated iterates and retains the theoretical performance guarantees of the unconstrained case. We focus on Anderson acceleration of the classical projected gradient descent (PGD) method, but our techniques can easily be extended to more sophisticated algorithms, such as mirror descent. Due to the presence of a constraint set, the relevant fixed-point mapping for PGD is not differentiable. However, we show that the convergence results for Anderson acceleration of smooth fixed-point iterations can be extended to the non-smooth case under certain technical conditions.

  • 20.
    Ohnishi, Motoya
    et al.
    Keio Univ, KTH, RIKEN, Tokyo, Japan..
    Yukawa, Masahiro
    Keio Univ, RIKEN, Tokyo, Japan..
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Sugiyama, Masashi
    Univ Tokyo, RIKEN, Tokyo, Japan..
    Continuous-time Value Function Approximation in Reproducing Kernel Hilbert Spaces2018In: Advances in Neural Information Processing Systems 31 (NIPS 2018) / [ed] Bengio, S Wallach, H Larochelle, H Grauman, K CesaBianchi, N Garnett, R, Neural Information Processing Systems (NIPS) , 2018, Vol. 31Conference paper (Refereed)
    Abstract [en]

    Motivated by the success of reinforcement learning (RL) for discrete-time tasks such as AlphaGo and Atari games, there has been a recent surge of interest in using RL for continuous-time control of physical systems (cf. many challenging tasks in OpenAI Gym and DeepMind Control Suite). Since discretization of time is susceptible to error, it is methodologically more desirable to handle the system dynamics directly in continuous time. However, very few techniques exist for continuous-time RL and they lack flexibility in value function approximation. In this paper, we propose a novel framework for model-based continuous-time value function approximation in reproducing kernel Hilbert spaces. The resulting framework is so flexible that it can accommodate any kind of kernel-based approach, such as Gaussian processes and kernel adaptive filters, and it allows us to handle uncertainties and nonstationarity without prior knowledge about the environment or what basis functions to employ. We demonstrate the validity of the presented framework through experiments.

  • 21.
    Poulimeneas, Dimitrios
    et al.
    KTH, School of Electrical Engineering (EES), Automatic Control.
    Charalambous, T.
    Nomikos, N.
    Krikidis, I.
    Vouyioukas, D.
    Johansson, Mikael
    KTH, School of Electrical Engineering (EES), Automatic Control.
    A delay-aware hybrid relay selection policy2016In: 2016 23rd International Conference on Telecommunications, ICT 2016, IEEE conference proceedings, 2016Conference paper (Refereed)
    Abstract [en]

    In this paper, we propose a novel relay selection policy based on the Hybrid Relay Selection (HRS) relay selection protocol that takes into account the state of the buffers and aims at reducing the average packet delays in the network. The proposed protocol, called the Delay-Aware HRS (DA - HRS) protocol is analyzed by means of Markov Chains and expressions for the outage probability, throughput and delay are derived. The distributed implementation of the protocol is also discussed. The performance of our proposed protocol is demonstrated via extensive simulations and comparisons with the classical HRS.

  • 22. Shi, Guodong
    et al.
    Li, Bo
    Johansson, Mikael
    KTH, School of Electrical Engineering (EES), Automatic Control. KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    Johansson, Karl Henrik
    KTH, School of Electrical Engineering (EES), Automatic Control. KTH, School of Electrical Engineering (EES), Centres, ACCESS Linnaeus Centre.
    Finite-Time Convergent Gossiping2016In: IEEE/ACM Transactions on Networking, ISSN 1063-6692, E-ISSN 1558-2566, Vol. 24, no 5, p. 2814-2826Article in journal (Refereed)
    Abstract [en]

    Gossip algorithms are widely used in modern distributed systems, with applications ranging from sensor networks and peer-to-peer networks to mobile vehicle networks and social networks. A tremendous research effort has been devoted to analyzing and improving the asymptotic rate of convergence for gossip algorithms. In this work we study finite-time convergence of deterministic gossiping. We show that there exists a symmetric gossip algorithm that converges in finite time if and only if the number of network nodes is a power of two, while there always exists an asymmetric gossip algorithm with finite-time convergence, independent of the number of nodes. For n = 2(m) nodes, we prove that a fastest convergence can be reached in nm = n log(2) n node updates via symmetric gossiping. On the other hand, under asymmetric gossip among n = 2(m) + rnodes with 0 <= r <= 2(m), it takes at least mn + 2r node updates for achieving finite-time convergence. It is also shown that the existence of finite-time convergent gossiping often imposes strong structural requirements on the underlying interaction graph. Finally, we apply our results to gossip algorithms in quantum networks, where the goal is to control the state of a quantum system via pairwise interactions. We show that finite-time convergence is never possible for such systems.

  • 23.
    Simonetto, Andrea
    et al.
    Delft Univ Technol, Delft Ctr Syst & Control, Mekelweg 2, NL-2628 CD Delft, Netherlands..
    Keviczky, Tamas
    Delft Univ Technol, Delft Ctr Syst & Control, Mekelweg 2, NL-2628 CD Delft, Netherlands..
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control. Royal Inst Technol KTH, Sch Elect Engn, ACCESS Linnaeus Ctr, S-10044 Stockholm, Sweden..
    A Regularized Saddle-Point Algorithm for Networked Optimization with Resource Allocation Constraints2012In: 2012 IEEE 51ST ANNUAL CONFERENCE ON DECISION AND CONTROL (CDC), IEEE , 2012, p. 7476-7481Conference paper (Refereed)
    Abstract [en]

    We propose a regularized saddle-point algorithm for convex networked optimization problems with resource allocation constraints. Standard distributed gradient methods suffer from slow convergence and require excessive communication when applied to problems of this type. Our approach offers an alternative way to address these problems, and ensures that each iterative update step satisfies the resource allocation constraints. We derive step-size conditions under which the distributed algorithm converges geometrically to the regularized optimal value, and show how these conditions are affected by the underlying network topology. We illustrate our method on a robotic network application example where a group of mobile agents strive to maintain a moving target in the barycenter of their positions.

  • 24.
    Talebi Mazraeh Shahi, Mohammad Sadegh
    et al.
    KTH, School of Electrical Engineering and Computer Science (EECS), Centres, ACCESS Linnaeus Centre.
    Zou, Zhenhua
    Ericsson Res, SE-16483 Stockholm, Sweden..
    Combes, Richard
    Cent Supelec L2S, Telecommun Dept, F-91192 Gif Sur Yvette, France..
    Proutiere, Alexandre
    KTH, School of Electrical Engineering and Computer Science (EECS), Centres, ACCESS Linnaeus Centre.
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Centres, ACCESS Linnaeus Centre.
    Stochastic Online Shortest Path Routing: The Value of Feedback2018In: IEEE Transactions on Automatic Control, ISSN 0018-9286, E-ISSN 1558-2523, Vol. 63, no 4, p. 915-930Article in journal (Refereed)
    Abstract [en]

    This paper studies online shortest path routing over multihop networks. Link costs or delays are time varying and modeled by independent and identically distributed random processes, whose parameters are initially unknown. The parameters, and hence the optimal path, can only be estimated by routing packets through the network and observing the realized delays. Our aim is to find a routing policy that minimizes the regret (the cumulative difference of expected delay) between the path chosen by the policy and the unknown optimal path. We formulate the problem as a combinatorial bandit optimization problem and consider several scenarios that differ in where routing decisions are made and in the information available when making the decisions. For each scenario, we derive a tight asymptotic lower bound on the regret that has to be satisfied by any online routing policy. Three algorithms, with a tradeoff between computational complexity and performance, are proposed. The regret upper bounds of these algorithms improve over those of the existing algorithms. We also assess numerically the performance of the proposed algorithms and compare it to that of existing algorithms.

  • 25. Tzortzis, Ioannis
    et al.
    Charalambous, Charalambos D.
    Charalambous, Themistoklis
    Hadjicostis, Christoforos N.
    Johansson, Mikael
    KTH, School of Electrical Engineering (EES), Automatic Control.
    Approximation of Markov Processes by Lower Dimensional Processes via Total Variation Metrics2017In: IEEE Transactions on Automatic Control, ISSN 0018-9286, E-ISSN 1558-2523, Vol. 62, no 3, p. 1030-1045Article in journal (Refereed)
    Abstract [en]

    The aim of this paper is to approximate a Finite-State Markov (FSM) process by another process defined on a lower dimensional state space, called the approximating process, with respect to a total variation distance fidelity criterion. The approximation problem is formulated as an optimization problem using two different approaches. The first approach is based on approximating the transition probability matrix of the FSM process by a lower-dimensional transition probability matrix, resulting in an approximating process which is a Finite-State Hidden Markov (FSHM) process. The second approach is based on approximating the invariant probability vector of the original FSM process by another invariant probability vector defined on a lower-dimensional state space. Going a step further, a method is proposed based on optimizing a Kullback-Leibler divergence to approximate the FSHM processes by FSM processes. The solutions of these optimization problems are described by optimal partition functions which aggregate the states of the FSM process via a corresponding water-filling solution, resulting in lower-dimensional approximating processes which are FSHM or FSM processes. Throughout the paper, the theoretical results are justified by illustrative examples that demonstrate our proposed methodology.

  • 26.
    Åstrand, Max
    et al.
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control. ABB Corporate Research Center, Västerås, Sweden.
    Johansson, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Zanarini, A.
    Fleet scheduling in underground mines using constraint programming2018In: 15th International Conference on Integration of Constraint Programming, Artificial Intelligence, and Operations Research, CPAIOR 2018, Springer, 2018, Vol. 10848, p. 605-613Conference paper (Refereed)
    Abstract [en]

    The profitability of an underground mine is greatly affected by the scheduling of the mobile production fleet. Today, most mine operations are scheduled manually, which is a tedious and error-prone activity. In this contribution, we present and formalize the underground mine scheduling problem, and propose a CP-based model for solving it. The model is evaluated on instances generated from real data. The results are promising and show a potential for further extensions.

1 - 26 of 26
CiteExportLink to result list
Permanent link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf