Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626836
Jakub Matousek, J. Duník, M. Brandner, V. Elvira
This paper deals with the state estimation of non-linear stochastic dynamic systems, both continuous and discrete in time, with an emphasis on a numerical solution to the Bayesian relations by the point-mass filters. The filters for discrete-discrete and continuous-discrete state-space models are reviewed and a new highly accurate and fast active flux method is introduced and adapted for a continuous filter design. A wide set of the point-mass filters is compared in a numerical study together with a set of particle filters.
{"title":"Comparison of Discrete and Continuous State Estimation with Focus on Active Flux Scheme","authors":"Jakub Matousek, J. Duník, M. Brandner, V. Elvira","doi":"10.23919/fusion49465.2021.9626836","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626836","url":null,"abstract":"This paper deals with the state estimation of non-linear stochastic dynamic systems, both continuous and discrete in time, with an emphasis on a numerical solution to the Bayesian relations by the point-mass filters. The filters for discrete-discrete and continuous-discrete state-space models are reviewed and a new highly accurate and fast active flux method is introduced and adapted for a continuous filter design. A wide set of the point-mass filters is compared in a numerical study together with a set of particle filters.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125489367","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626881
Lino Antoni Giefer, J. Clemens
State estimation of objects plays an important role in various kinds of applications in the fields of robotics and autonomous vehicles. With the continuous advancement of sensors with high spatial resolution, especially light detection and ranging (LiDAR), the interest in accurate and reliable extended object trackers has grown over the last years. Classical state estimation approaches assume static and symmetric shapes, such as rectangles or ellipses, or compositions of those. The disadvantage of that assumption is obvious: deformations, as in the case of articulated vehicles driving along curves, cannot be captured appropriately. In this paper, we tackle this problem by proposing a novel approach to state estimation employing deformed superellipses. This allows a closed-form mathematical description of an articulated object’s state in the Euclidean plane consisting of its pose and shape. Two additional state parameters are introduced capturing the deformation angle and the joint’s position. We evaluate the proposed approach to state estimation of articulated objects employing a model fitting algorithm of simulated LiDAR measurements and show the improvements compared to classical shape assumptions. Furthermore, we discuss the use of our approach in a tracking algorithm.
{"title":"State Estimation of Articulated Vehicles Using Deformed Superellipses","authors":"Lino Antoni Giefer, J. Clemens","doi":"10.23919/fusion49465.2021.9626881","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626881","url":null,"abstract":"State estimation of objects plays an important role in various kinds of applications in the fields of robotics and autonomous vehicles. With the continuous advancement of sensors with high spatial resolution, especially light detection and ranging (LiDAR), the interest in accurate and reliable extended object trackers has grown over the last years. Classical state estimation approaches assume static and symmetric shapes, such as rectangles or ellipses, or compositions of those. The disadvantage of that assumption is obvious: deformations, as in the case of articulated vehicles driving along curves, cannot be captured appropriately. In this paper, we tackle this problem by proposing a novel approach to state estimation employing deformed superellipses. This allows a closed-form mathematical description of an articulated object’s state in the Euclidean plane consisting of its pose and shape. Two additional state parameters are introduced capturing the deformation angle and the joint’s position. We evaluate the proposed approach to state estimation of articulated objects employing a model fitting algorithm of simulated LiDAR measurements and show the improvements compared to classical shape assumptions. Furthermore, we discuss the use of our approach in a tracking algorithm.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121542246","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626890
Gang Yao, P. Wang, K. Berntorp, Hassan Mansour, P. Boufounos, P. Orlik
This paper considers extended object tracking (EOT) using high-resolution automotive radar measurements with online spatial model adaptation. This is motivated by the fact that offline learned spatial models may be over-smoothed due to coarsely labeled training data and can be mismatched to onboard radar sensors due to different specifications. To refine the offline learned spatial representation in an online setting, we first apply the unscented Rauch-Tung-Striebel (RTS) smoother that explicitly accounts for the predicted and filtered states based on the offline learned model (i.e., the B-spline chained ellipses model). The smoothed state estimates are then used to create an online batch of state-decoupled training data that are subsequently utilized by an expectation-maximization algorithm to update the spatial model parameters. Numerical validation with synthetic automotive radar measurements is provided to verify the effectiveness of the proposed online model adaptation scheme.
{"title":"Extended Object Tracking with Spatial Model Adaptation Using Automotive Radar","authors":"Gang Yao, P. Wang, K. Berntorp, Hassan Mansour, P. Boufounos, P. Orlik","doi":"10.23919/fusion49465.2021.9626890","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626890","url":null,"abstract":"This paper considers extended object tracking (EOT) using high-resolution automotive radar measurements with online spatial model adaptation. This is motivated by the fact that offline learned spatial models may be over-smoothed due to coarsely labeled training data and can be mismatched to onboard radar sensors due to different specifications. To refine the offline learned spatial representation in an online setting, we first apply the unscented Rauch-Tung-Striebel (RTS) smoother that explicitly accounts for the predicted and filtered states based on the offline learned model (i.e., the B-spline chained ellipses model). The smoothed state estimates are then used to create an online batch of state-decoupled training data that are subsequently utilized by an expectation-maximization algorithm to update the spatial model parameters. Numerical validation with synthetic automotive radar measurements is provided to verify the effectiveness of the proposed online model adaptation scheme.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127794762","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9627029
D. Grimmett, D. Abraham, Ricki Alberto
In this paper, a "cognitive" active sonar tracking algorithm is described, and results of its application to data from the LCAS’15 sea trial are shown. A key factor in tracker performance is the scheme used for track initiation and termination. A very common track initiation algorithm (TIA) is the sliding M-of-N processor, however, the tuning of its parameters can be difficult. It is often heuristic and sub-optimum, in achieving both good tracking performance of true targets as well as controlling the false track rate (FTR) for a desired sonar Pd/Pfa operating point. This is of particular concern for clutter-rich reverberation-limited undersea acoustic environments, where the false-alarm rates are high. The algorithm utilizes available in-situ data to estimate the statistics of the encountered clutter, and then optimizes tracker performance to meet specified operational levels. The adaptive algorithm is shown to effectively control the false track rate. The algorithm has potential to cognitively self-tune its operations for optimum performance.
{"title":"Cognitive Active Sonar Tracking for Optimum Performance in Clutter","authors":"D. Grimmett, D. Abraham, Ricki Alberto","doi":"10.23919/fusion49465.2021.9627029","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9627029","url":null,"abstract":"In this paper, a \"cognitive\" active sonar tracking algorithm is described, and results of its application to data from the LCAS’15 sea trial are shown. A key factor in tracker performance is the scheme used for track initiation and termination. A very common track initiation algorithm (TIA) is the sliding M-of-N processor, however, the tuning of its parameters can be difficult. It is often heuristic and sub-optimum, in achieving both good tracking performance of true targets as well as controlling the false track rate (FTR) for a desired sonar Pd/Pfa operating point. This is of particular concern for clutter-rich reverberation-limited undersea acoustic environments, where the false-alarm rates are high. The algorithm utilizes available in-situ data to estimate the statistics of the encountered clutter, and then optimizes tracker performance to meet specified operational levels. The adaptive algorithm is shown to effectively control the false track rate. The algorithm has potential to cognitively self-tune its operations for optimum performance.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134339776","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626839
Samuele Capobianco, N. Forti, L. Millefiori, P. Braca, P. Willett
In this paper, we propose a deep learning framework for sequence-to-sequence vessel trajectory prediction based on encoder-decoder recurrent neural networks to learn the predictive distribution of maritime patterns from historical Automatic Identification System data and sequentially generate future trajectory estimates given previous observations. Special focus is given on modeling the predictive uncertainty of future estimates arising from the inherent non-deterministic nature of maritime traffic. An attention-based aggregation layer connects the encoder and decoder networks and captures space-time dependencies in sequential data. Experimental results on trajectories from the Danish Maritime Authority dataset demonstrate the effectiveness of the proposed attention-based deep learning model for vessel prediction and show how uncertainty estimates can prove to be extremely informative of the prediction error.
{"title":"Uncertainty-Aware Recurrent Encoder-Decoder Networks for Vessel Trajectory Prediction","authors":"Samuele Capobianco, N. Forti, L. Millefiori, P. Braca, P. Willett","doi":"10.23919/fusion49465.2021.9626839","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626839","url":null,"abstract":"In this paper, we propose a deep learning framework for sequence-to-sequence vessel trajectory prediction based on encoder-decoder recurrent neural networks to learn the predictive distribution of maritime patterns from historical Automatic Identification System data and sequentially generate future trajectory estimates given previous observations. Special focus is given on modeling the predictive uncertainty of future estimates arising from the inherent non-deterministic nature of maritime traffic. An attention-based aggregation layer connects the encoder and decoder networks and captures space-time dependencies in sequential data. Experimental results on trajectories from the Danish Maritime Authority dataset demonstrate the effectiveness of the proposed attention-based deep learning model for vessel prediction and show how uncertainty estimates can prove to be extremely informative of the prediction error.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133840660","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626946
J. H. Ramos, Davis W. Adams, K. Brink, M. Majji
The partial-update filter concept is a recent development that generalizes the Schmidt Kalman filter and extends the range of nonlinearities and uncertainties that a Kalman filter can tolerate. Similar to the Schmidt filter, the intention of the partial-update filter is to ameliorate the negative impact that certain states have within the filter, often due to their poor observability. In contrast with the Schmidt filter, the partial-update filter can update the problematic states at any time step. In practice, the partial-update technique can apply a full (nominal), partial, or no update (Schmidt) to states, depending on user-selected percentages (or weights) that indicate how much of the nominal Kalman update is applied. To date, the update percentages are selected via trial and error, and any change in the system configuration requires re-tuning. Furthermore, because the update percentages are fixed, the partial-update is agnostic to situations where a full update, or even a Schmidt-like filter can be more suitable. To address these drawbacks, this paper proposes two observability informed approaches for online weight selection that do not require manual tuning. The proposed techniques are targeted for systems where the states to be partially updated are only the problematic states. Numerical simulation results demonstrate that the proposed approaches produce estimates comparable to those of a manually fine-tuned fixed partial-update, and that they leverage occasions where local observability increases to produce more accurate estimates.
{"title":"Observability Informed Partial-Update Schmidt Kalman Filter","authors":"J. H. Ramos, Davis W. Adams, K. Brink, M. Majji","doi":"10.23919/fusion49465.2021.9626946","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626946","url":null,"abstract":"The partial-update filter concept is a recent development that generalizes the Schmidt Kalman filter and extends the range of nonlinearities and uncertainties that a Kalman filter can tolerate. Similar to the Schmidt filter, the intention of the partial-update filter is to ameliorate the negative impact that certain states have within the filter, often due to their poor observability. In contrast with the Schmidt filter, the partial-update filter can update the problematic states at any time step. In practice, the partial-update technique can apply a full (nominal), partial, or no update (Schmidt) to states, depending on user-selected percentages (or weights) that indicate how much of the nominal Kalman update is applied. To date, the update percentages are selected via trial and error, and any change in the system configuration requires re-tuning. Furthermore, because the update percentages are fixed, the partial-update is agnostic to situations where a full update, or even a Schmidt-like filter can be more suitable. To address these drawbacks, this paper proposes two observability informed approaches for online weight selection that do not require manual tuning. The proposed techniques are targeted for systems where the states to be partially updated are only the problematic states. Numerical simulation results demonstrate that the proposed approaches produce estimates comparable to those of a manually fine-tuned fixed partial-update, and that they leverage occasions where local observability increases to produce more accurate estimates.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122998492","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9627003
S. Coraluppi, C. Rago, C. Carthel, Brandon Bale
This paper focuses on two challenges in multi-target tracking with passive sensors. The first is the well-known observability problem whereby individual sensor measurements are insufficient to localize targets. The second is the need to relax the usual small-target assumption of at most one measurement per target per scan. Indeed, in some applications such as passive sonar, there are repeated measurements, i.e. multiple detections per target per scan of one sensor. We examine these challenges in a multi-sensor setting and describe the advantages of a distributed MHT solution architecture, with measurement-space tracking following by multi-sensor Cartesian tracking using a robust Cartesian initialization scheme. In the presence of repeated measurements, there are (at least) two viable processing architectures. In both cases we leverage a recently developed generalization to the MHT recursion. We study the relative merits of the two alternative solutions.
{"title":"Distributed MHT with Passive Sensors","authors":"S. Coraluppi, C. Rago, C. Carthel, Brandon Bale","doi":"10.23919/fusion49465.2021.9627003","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9627003","url":null,"abstract":"This paper focuses on two challenges in multi-target tracking with passive sensors. The first is the well-known observability problem whereby individual sensor measurements are insufficient to localize targets. The second is the need to relax the usual small-target assumption of at most one measurement per target per scan. Indeed, in some applications such as passive sonar, there are repeated measurements, i.e. multiple detections per target per scan of one sensor. We examine these challenges in a multi-sensor setting and describe the advantages of a distributed MHT solution architecture, with measurement-space tracking following by multi-sensor Cartesian tracking using a robust Cartesian initialization scheme. In the presence of repeated measurements, there are (at least) two viable processing architectures. In both cases we leverage a recently developed generalization to the MHT recursion. We study the relative merits of the two alternative solutions.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128595073","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626905
S. Koch, Luisa Still, M. Oispuu, W. Koch
This paper deals with shooter localization based on measurements of a sensor network with spatially distributed, non-synchronized single microphones. The acoustic events generated during gunfire – shock wave and muzzle blast – provide information about shooter position and firing direction. A new approach is presented that takes into account the length of the N-shape of the shock wave in addition to the typically used measurement of the time difference of arrival (TDOA) between shock wave and muzzle blast. The accuracy of the new approach is evaluated using Cramér-Rao bounds, Monte Carlo simulations, and measurement experiments. The results are particularly promising in cases where no other approach achieves high accuracy.
{"title":"Shooter Localization Based on TDOA and N-Shape Length Measurements of Distributed Microphones","authors":"S. Koch, Luisa Still, M. Oispuu, W. Koch","doi":"10.23919/fusion49465.2021.9626905","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626905","url":null,"abstract":"This paper deals with shooter localization based on measurements of a sensor network with spatially distributed, non-synchronized single microphones. The acoustic events generated during gunfire – shock wave and muzzle blast – provide information about shooter position and firing direction. A new approach is presented that takes into account the length of the N-shape of the shock wave in addition to the typically used measurement of the time difference of arrival (TDOA) between shock wave and muzzle blast. The accuracy of the new approach is evaluated using Cramér-Rao bounds, Monte Carlo simulations, and measurement experiments. The results are particularly promising in cases where no other approach achieves high accuracy.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117257446","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9627010
V. Nguyen
Situational awareness requires continual learning from observations and adaptive reasoning from domain and contextual knowledge. The integration of reasoning and learning has been a standing goal of machine learning and AI in general, and a pressing need for real-world situational awareness in particular. Representative techniques among the numerous methods proposed include integrating logics with learning formalisms, whether probabilistic graphical models or neural methods. These techniques are motivated by the need to model and exploit the symmetry, regularities and complex relations between entities exhibited in real world scenarios (in the form of relational or graph data) for effective reasoning and learning. In this work, we investigate the benefits of integrating two prominent methods for reasoning and learning with relational/graph data, Markov Logic Networks (or simply Markov Logic) and Graph Neural Networks. The former is well-recognised for its powerful representation and uncertainty handling, while the latter have gained much attention due to their efficiency in handling large-scale graph datasets. This paper reports on the potential benefits of combining their respective strengths and applying them to a use case illustration in the maritime domain, together with empirical results.
{"title":"Markov Logic meets Graph Neural Networks: A Study for Situational Awareness","authors":"V. Nguyen","doi":"10.23919/fusion49465.2021.9627010","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9627010","url":null,"abstract":"Situational awareness requires continual learning from observations and adaptive reasoning from domain and contextual knowledge. The integration of reasoning and learning has been a standing goal of machine learning and AI in general, and a pressing need for real-world situational awareness in particular. Representative techniques among the numerous methods proposed include integrating logics with learning formalisms, whether probabilistic graphical models or neural methods. These techniques are motivated by the need to model and exploit the symmetry, regularities and complex relations between entities exhibited in real world scenarios (in the form of relational or graph data) for effective reasoning and learning. In this work, we investigate the benefits of integrating two prominent methods for reasoning and learning with relational/graph data, Markov Logic Networks (or simply Markov Logic) and Graph Neural Networks. The former is well-recognised for its powerful representation and uncertainty handling, while the latter have gained much attention due to their efficiency in handling large-scale graph datasets. This paper reports on the potential benefits of combining their respective strengths and applying them to a use case illustration in the maritime domain, together with empirical results.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126799038","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626952
Fei He, N. Rao, Chris Y. T. Ma
Job scheduling at supercomputing facilities is important for achieving high utilization of these valuable resources while ensuring effective execution of jobs submitted by users. The jobs are scheduled according to their specified resource demands such as expected job completion times, and the available resources based on allocations. Jobs that overrun their allocated times are terminated, for example, after a grace-period. It is non-trivial and often very complex for users to accurately estimate the completion times of their jobs, and consequently they face a dilemma: underestimate the job time to have a higher priority and risk job termination due to overrun, or overestimate it to ensure its completion and risk its delayed execution. In this paper, we investigate whether providing grace-period can benefit facility performance by developing a game- theoretic model between a facility provider and multiple users for a simplified scheduling scenario based on job execution times. We present closed-form expressions for the provider’s and user’s best-response strategies to maximize their respective utility functions. We describe conditions under which offering a grace-period is advantageous to both facility provider and users by deriving the Nash equilibrium of the game.
{"title":"Game-Theoretic Approach for Grace-Period Policy in Supercomputers","authors":"Fei He, N. Rao, Chris Y. T. Ma","doi":"10.23919/fusion49465.2021.9626952","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626952","url":null,"abstract":"Job scheduling at supercomputing facilities is important for achieving high utilization of these valuable resources while ensuring effective execution of jobs submitted by users. The jobs are scheduled according to their specified resource demands such as expected job completion times, and the available resources based on allocations. Jobs that overrun their allocated times are terminated, for example, after a grace-period. It is non-trivial and often very complex for users to accurately estimate the completion times of their jobs, and consequently they face a dilemma: underestimate the job time to have a higher priority and risk job termination due to overrun, or overestimate it to ensure its completion and risk its delayed execution. In this paper, we investigate whether providing grace-period can benefit facility performance by developing a game- theoretic model between a facility provider and multiple users for a simplified scheduling scenario based on job execution times. We present closed-form expressions for the provider’s and user’s best-response strategies to maximize their respective utility functions. We describe conditions under which offering a grace-period is advantageous to both facility provider and users by deriving the Nash equilibrium of the game.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124884288","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：