Pub Date : 2015-09-01DOI: 10.1109/ALLERTON.2015.7447159
Varun Jog, Po-Ling Loh
We derive sharp thresholds for exact recovery of communities in a weighted stochastic block model, where observations are collected in the form of a weighted adjacency matrix, and the weight of each edge is generated independently from a distribution determined by the community membership of its endpoints. Our main result, characterizing the precise boundary between success and failure of maximum likelihood estimation when edge weights are drawn from discrete distributions, involves the Renyi divergence of order 1/2 between the distributions of within-community and between-community edges. When the Renyi divergence is above a certain threshold, meaning the edge distributions are sufficiently separated, maximum likelihood succeeds with probability tending to 1; when the Renyi divergence is below the threshold, maximum likelihood fails with probability bounded away from 0. In the language of graphical channels, the Renyi divergence pinpoints the information-theoretic capacity of discrete graphical channels with binary inputs. Our results generalize previously established thresholds derived specifically for unweighted block models, and support an important natural intuition relating the intrinsic hardness of community estimation to the problem of edge classification. Along the way, we establish a general relationship between the Renyi divergence and the probability of success of the maximum likelihood estimator for arbitrary edge weight distributions. Finally, we discuss consequences of our bounds for the related problems of censored block models and submatrix localization, which may be seen as special cases of the framework developed in our paper.
{"title":"Recovering communities in weighted stochastic block models","authors":"Varun Jog, Po-Ling Loh","doi":"10.1109/ALLERTON.2015.7447159","DOIUrl":"https://doi.org/10.1109/ALLERTON.2015.7447159","url":null,"abstract":"We derive sharp thresholds for exact recovery of communities in a weighted stochastic block model, where observations are collected in the form of a weighted adjacency matrix, and the weight of each edge is generated independently from a distribution determined by the community membership of its endpoints. Our main result, characterizing the precise boundary between success and failure of maximum likelihood estimation when edge weights are drawn from discrete distributions, involves the Renyi divergence of order 1/2 between the distributions of within-community and between-community edges. When the Renyi divergence is above a certain threshold, meaning the edge distributions are sufficiently separated, maximum likelihood succeeds with probability tending to 1; when the Renyi divergence is below the threshold, maximum likelihood fails with probability bounded away from 0. In the language of graphical channels, the Renyi divergence pinpoints the information-theoretic capacity of discrete graphical channels with binary inputs. Our results generalize previously established thresholds derived specifically for unweighted block models, and support an important natural intuition relating the intrinsic hardness of community estimation to the problem of edge classification. Along the way, we establish a general relationship between the Renyi divergence and the probability of success of the maximum likelihood estimator for arbitrary edge weight distributions. Finally, we discuss consequences of our bounds for the related problems of censored block models and submatrix localization, which may be seen as special cases of the framework developed in our paper.","PeriodicalId":112948,"journal":{"name":"2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton)","volume":"46 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116708269","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}
Deep learning based on artificial neural networks is a very popular approach to modeling, classifying, and recognizing complex data such as images, speech, and text. The unprecedented accuracy of deep learning methods has turned them into the foundation of new AI-based services on the Internet. Commercial companies that collect user data on a large scale have been the main beneficiaries of this trend since the success of deep learning techniques is directly proportional to the amount of data available for training. Massive data collection required for deep learning presents obvious privacy issues. Users' personal, highly sensitive data such as photos and voice recordings is kept indefinitely by the companies that collect it. Users can neither delete it, nor restrict the purposes for which it is used. Furthermore, centrally kept data is subject to legal subpoenas and extrajudicial surveillance. Many data owners-for example, medical institutions that may want to apply deep learning methods to clinical records-are prevented by privacy and confidentiality concerns from sharing the data and thus benefitting from large-scale deep learning. In this paper, we present a practical system that enables multiple parties to jointly learn an accurate neural-network model for a given objective without sharing their input datasets. We exploit the fact that the optimization algorithms used in modern deep learning, namely, those based on stochastic gradient descent, can be parallelized and executed asynchronously. Our system lets participants train independently on their own datasets and selectively share small subsets of their models' key parameters during training. This offers an attractive point in the utility/privacy tradeoff space: participants preserve the privacy of their respective data while still benefitting from other participants' models and thus boosting their learning accuracy beyond what is achievable solely on their own inputs. We demonstrate the accuracy of our privacy-preserving deep learning on benchmark datasets.
{"title":"Privacy-preserving deep learning","authors":"R. Shokri, Vitaly Shmatikov","doi":"10.1145/2810103.2813687","DOIUrl":"https://doi.org/10.1145/2810103.2813687","url":null,"abstract":"Deep learning based on artificial neural networks is a very popular approach to modeling, classifying, and recognizing complex data such as images, speech, and text. The unprecedented accuracy of deep learning methods has turned them into the foundation of new AI-based services on the Internet. Commercial companies that collect user data on a large scale have been the main beneficiaries of this trend since the success of deep learning techniques is directly proportional to the amount of data available for training. Massive data collection required for deep learning presents obvious privacy issues. Users' personal, highly sensitive data such as photos and voice recordings is kept indefinitely by the companies that collect it. Users can neither delete it, nor restrict the purposes for which it is used. Furthermore, centrally kept data is subject to legal subpoenas and extrajudicial surveillance. Many data owners-for example, medical institutions that may want to apply deep learning methods to clinical records-are prevented by privacy and confidentiality concerns from sharing the data and thus benefitting from large-scale deep learning. In this paper, we present a practical system that enables multiple parties to jointly learn an accurate neural-network model for a given objective without sharing their input datasets. We exploit the fact that the optimization algorithms used in modern deep learning, namely, those based on stochastic gradient descent, can be parallelized and executed asynchronously. Our system lets participants train independently on their own datasets and selectively share small subsets of their models' key parameters during training. This offers an attractive point in the utility/privacy tradeoff space: participants preserve the privacy of their respective data while still benefitting from other participants' models and thus boosting their learning accuracy beyond what is achievable solely on their own inputs. We demonstrate the accuracy of our privacy-preserving deep learning on benchmark datasets.","PeriodicalId":112948,"journal":{"name":"2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton)","volume":"183 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115552707","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 : 2015-09-01DOI: 10.1109/ALLERTON.2015.7447179
M. Sharifzadeh, D. Schonfeld
This paper presents a novel approach to the optimization and performance bounds of video steganography. Hypothesis testing is used to derive the probability of detection and false alarm for a cooperator with a priori knowledge of a carrier signal and an attacker for whom the carrier signal is unknown. The result is then used to optimize the statistical performance of a well-known video steganography method (i.e., secure spread spectrum watermarking) while ensuring limits on the statistical performance of video steganalysis. In addition, the channel capacity for video steganography and steganalsyis are ascertained under the proposed statistical model. It is then used to characterize an optimal information-theoretic criterion for video steganography subject to performance bounds on statistical steganalysis. Theoretical and numerical results demonstrate the consistency of both the statistical and information-theoretic approaches to the optimization of video steganography.
{"title":"Statistical and information-theoretic optimization and performance bounds of video steganography","authors":"M. Sharifzadeh, D. Schonfeld","doi":"10.1109/ALLERTON.2015.7447179","DOIUrl":"https://doi.org/10.1109/ALLERTON.2015.7447179","url":null,"abstract":"This paper presents a novel approach to the optimization and performance bounds of video steganography. Hypothesis testing is used to derive the probability of detection and false alarm for a cooperator with a priori knowledge of a carrier signal and an attacker for whom the carrier signal is unknown. The result is then used to optimize the statistical performance of a well-known video steganography method (i.e., secure spread spectrum watermarking) while ensuring limits on the statistical performance of video steganalysis. In addition, the channel capacity for video steganography and steganalsyis are ascertained under the proposed statistical model. It is then used to characterize an optimal information-theoretic criterion for video steganography subject to performance bounds on statistical steganalysis. Theoretical and numerical results demonstrate the consistency of both the statistical and information-theoretic approaches to the optimization of video steganography.","PeriodicalId":112948,"journal":{"name":"2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115134001","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 : 2015-09-01DOI: 10.1109/ALLERTON.2015.7447089
Javad Heydari, A. Tajer, H. Poor
The problem of quickest data-adaptive and sequential search for clusters in a Gauss-Markov random field is considered. In the existing literature, such search for clusters is often performed using fixed sample size and non-adaptive strategies. In order to accommodate large networks, in which data adaptivity leads to significant gains in detection quality and agility, in this paper sequential and data-adaptive detection strategies are proposed and are shown to enjoy asymptotic optimality. The quickest detection problem is abstracted by adopting an acyclic dependency graph to model the mutual effects of different random variables in the field and decision making rules are derived for general random fields and specialized for Gauss-Markov random fields. Performance evaluations demonstrate the gains of the data-adaptive schemes over existing techniques in terms of sampling complexity and error exponents.
{"title":"Quickest detection of Gauss-Markov random fields","authors":"Javad Heydari, A. Tajer, H. Poor","doi":"10.1109/ALLERTON.2015.7447089","DOIUrl":"https://doi.org/10.1109/ALLERTON.2015.7447089","url":null,"abstract":"The problem of quickest data-adaptive and sequential search for clusters in a Gauss-Markov random field is considered. In the existing literature, such search for clusters is often performed using fixed sample size and non-adaptive strategies. In order to accommodate large networks, in which data adaptivity leads to significant gains in detection quality and agility, in this paper sequential and data-adaptive detection strategies are proposed and are shown to enjoy asymptotic optimality. The quickest detection problem is abstracted by adopting an acyclic dependency graph to model the mutual effects of different random variables in the field and decision making rules are derived for general random fields and specialized for Gauss-Markov random fields. Performance evaluations demonstrate the gains of the data-adaptive schemes over existing techniques in terms of sampling complexity and error exponents.","PeriodicalId":112948,"journal":{"name":"2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126697341","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 : 2015-09-01DOI: 10.1109/ALLERTON.2015.7447153
Iñaki Estella Aguerri, A. Zaidi
We study the transmission over a cloud radio access network, in which multiple base stations (BS) are connected to a central processor (CP) via finite-capacity backhaul links. Focusing on maximizing the allowed sum-rate, we develop a lattice based coding scheme that generalizes both compute-and-forward and successive Wyner-Ziv coding for this model. The scheme builds on Cover and El Gamal partial-decode-compress-and-forward and is shown to strictly outperform the best of the aforementioned two popular schemes. The results are illustrated through some numerical examples.
{"title":"Partial compute-compress-and-forward for limited backhaul uplink multicell processing","authors":"Iñaki Estella Aguerri, A. Zaidi","doi":"10.1109/ALLERTON.2015.7447153","DOIUrl":"https://doi.org/10.1109/ALLERTON.2015.7447153","url":null,"abstract":"We study the transmission over a cloud radio access network, in which multiple base stations (BS) are connected to a central processor (CP) via finite-capacity backhaul links. Focusing on maximizing the allowed sum-rate, we develop a lattice based coding scheme that generalizes both compute-and-forward and successive Wyner-Ziv coding for this model. The scheme builds on Cover and El Gamal partial-decode-compress-and-forward and is shown to strictly outperform the best of the aforementioned two popular schemes. The results are illustrated through some numerical examples.","PeriodicalId":112948,"journal":{"name":"2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton)","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124386567","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 : 2015-09-01DOI: 10.1109/ALLERTON.2015.7447143
Osman Yağan
We introduce a new random key predistribution scheme for securing heterogeneous wireless sensor networks. Each of the n sensors in the network is classified into r classes according to some probability distribution μ = {μ1, ..., μr}. Before deployment, a class i sensor is assigned Ki cryptographic keys that are selected uniformly at random from a common pool of P keys, for each i = 1, ..., r. Once deployed, a pair of sensors can establish a secure communication channel if and only if they have a key in common. We model the topology of this network by an inhomogeneous random key graph. We establish scaling conditions on the parameters P and {K1, ..., Kr} so that the this graph has no isolated nodes with high probability. The result is given in the form of a zero-one law with the number of sensors n growing unboundedly large. An analogous result is also conjectured for the property of graph connectivity.
{"title":"Absence of isolated nodes in inhomogeneous random key graphs","authors":"Osman Yağan","doi":"10.1109/ALLERTON.2015.7447143","DOIUrl":"https://doi.org/10.1109/ALLERTON.2015.7447143","url":null,"abstract":"We introduce a new random key predistribution scheme for securing heterogeneous wireless sensor networks. Each of the n sensors in the network is classified into r classes according to some probability distribution μ = {μ1, ..., μr}. Before deployment, a class i sensor is assigned Ki cryptographic keys that are selected uniformly at random from a common pool of P keys, for each i = 1, ..., r. Once deployed, a pair of sensors can establish a secure communication channel if and only if they have a key in common. We model the topology of this network by an inhomogeneous random key graph. We establish scaling conditions on the parameters P and {K1, ..., Kr} so that the this graph has no isolated nodes with high probability. The result is given in the form of a zero-one law with the number of sensors n growing unboundedly large. An analogous result is also conjectured for the property of graph connectivity.","PeriodicalId":112948,"journal":{"name":"2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton)","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121625175","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 : 2015-09-01DOI: 10.1109/ALLERTON.2015.7447058
Radoslav Ivanov, Nikolay A. Atanasov, M. Pajic, George J. Pappas, Insup Lee
This paper presents the context-aware filter, an estimation technique that incorporates context measurements, in addition to the regular continuous measurements. Context measurements provide binary information about the system's context which is not directly encoded in the state; examples include a robot detecting a nearby building using image processing or a medical device alarming that a vital sign has exceeded a predefined threshold. These measurements can only be received from certain states and can therefore be modeled as a function of the system's current state. We focus on two classes of functions describing the probability of context detection given the current state; these functions capture a wide variety of detections that may occur in practice. We derive the corresponding context-aware filters, a Gaussian Mixture filter and another closed-form filter with a posterior distribution whose moments are derived in the paper. Finally, we evaluate the performance of both classes of functions through simulation of an unmanned ground vehicle.
{"title":"Robust estimation using context-aware filtering","authors":"Radoslav Ivanov, Nikolay A. Atanasov, M. Pajic, George J. Pappas, Insup Lee","doi":"10.1109/ALLERTON.2015.7447058","DOIUrl":"https://doi.org/10.1109/ALLERTON.2015.7447058","url":null,"abstract":"This paper presents the context-aware filter, an estimation technique that incorporates context measurements, in addition to the regular continuous measurements. Context measurements provide binary information about the system's context which is not directly encoded in the state; examples include a robot detecting a nearby building using image processing or a medical device alarming that a vital sign has exceeded a predefined threshold. These measurements can only be received from certain states and can therefore be modeled as a function of the system's current state. We focus on two classes of functions describing the probability of context detection given the current state; these functions capture a wide variety of detections that may occur in practice. We derive the corresponding context-aware filters, a Gaussian Mixture filter and another closed-form filter with a posterior distribution whose moments are derived in the paper. Finally, we evaluate the performance of both classes of functions through simulation of an unmanned ground vehicle.","PeriodicalId":112948,"journal":{"name":"2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton)","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121887880","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 : 2015-09-01DOI: 10.1109/ALLERTON.2015.7447022
Sunghyun Kim, Changho Suh
We investigate the role of relays in multiple access channels (MACs) with bursty user traffic, where intermittent data traffic restricts the users to bursty transmissions. Specifically, we examine a K-user bursty MIMO Gaussian MAC with a relay, where bursty traffic of each user is governed by a Bernoulli random process. As our main result, we characterize the degrees of freedom (DoF) region. To this end, we extend noisy network coding, in which relays compress-and-forward, to achieve the DoF cut-set bound. From this result, we establish the necessary and sufficient condition for attaining collision-free DoF performances. Also, we show that relays can provide a DoF gain which scales to some extent with additional relay antennas. Our results have practical implications in various scenarios of wireless systems, such as the Internet of Things (IoT) and media access control protocols.
{"title":"Degrees of freedom of bursty multiple access channels with a relay","authors":"Sunghyun Kim, Changho Suh","doi":"10.1109/ALLERTON.2015.7447022","DOIUrl":"https://doi.org/10.1109/ALLERTON.2015.7447022","url":null,"abstract":"We investigate the role of relays in multiple access channels (MACs) with bursty user traffic, where intermittent data traffic restricts the users to bursty transmissions. Specifically, we examine a K-user bursty MIMO Gaussian MAC with a relay, where bursty traffic of each user is governed by a Bernoulli random process. As our main result, we characterize the degrees of freedom (DoF) region. To this end, we extend noisy network coding, in which relays compress-and-forward, to achieve the DoF cut-set bound. From this result, we establish the necessary and sufficient condition for attaining collision-free DoF performances. Also, we show that relays can provide a DoF gain which scales to some extent with additional relay antennas. Our results have practical implications in various scenarios of wireless systems, such as the Internet of Things (IoT) and media access control protocols.","PeriodicalId":112948,"journal":{"name":"2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121776101","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 : 2015-09-01DOI: 10.1109/ALLERTON.2015.7447064
Yunshu Liu, J. Walsh
A 2N - 1 dimensional vector is said to be entropic if each of its entries can be regarded as the joint entropy of a particular subset of N discrete random variables. The explicit characterization of the closure of the region of entropic vectors Γ̅*N is unknown for N ≥ 4. A systematic approach is proposed to generate the list of non-isomorphic distribution supports for the purpose of calculating and optimizing entropic vectors. It is shown that a better understanding of the structure of the entropy region can be obtained by constructing inner bounds based on these supports. The constructed inner bounds based on different supports are compared both in full dimension and in a transformed three dimensional space of Csirmaz and Matúš.
{"title":"Non-isomorphic distribution supports for calculating entropic vectors","authors":"Yunshu Liu, J. Walsh","doi":"10.1109/ALLERTON.2015.7447064","DOIUrl":"https://doi.org/10.1109/ALLERTON.2015.7447064","url":null,"abstract":"A 2N - 1 dimensional vector is said to be entropic if each of its entries can be regarded as the joint entropy of a particular subset of N discrete random variables. The explicit characterization of the closure of the region of entropic vectors Γ̅*N is unknown for N ≥ 4. A systematic approach is proposed to generate the list of non-isomorphic distribution supports for the purpose of calculating and optimizing entropic vectors. It is shown that a better understanding of the structure of the entropy region can be obtained by constructing inner bounds based on these supports. The constructed inner bounds based on different supports are compared both in full dimension and in a transformed three dimensional space of Csirmaz and Matúš.","PeriodicalId":112948,"journal":{"name":"2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton)","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122181311","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 : 2015-09-01DOI: 10.1109/ALLERTON.2015.7447007
D. Kalathil, R. Rajagopal
Demand response is a key component of existing and future grid systems facing increased variability and peak demands. Scaling demand response requires efficiently predicting individual responses for large numbers of consumers while selecting the right ones to signal. This paper proposes a new online learning problem that captures consumer diversity, messaging fatigue and response prediction. We use the framework of multi-armed bandits model to address this problem. This yields simple and easy to implement index based learning algorithms with provable performance guarantees.
{"title":"Online learning for demand response","authors":"D. Kalathil, R. Rajagopal","doi":"10.1109/ALLERTON.2015.7447007","DOIUrl":"https://doi.org/10.1109/ALLERTON.2015.7447007","url":null,"abstract":"Demand response is a key component of existing and future grid systems facing increased variability and peak demands. Scaling demand response requires efficiently predicting individual responses for large numbers of consumers while selecting the right ones to signal. This paper proposes a new online learning problem that captures consumer diversity, messaging fatigue and response prediction. We use the framework of multi-armed bandits model to address this problem. This yields simple and easy to implement index based learning algorithms with provable performance guarantees.","PeriodicalId":112948,"journal":{"name":"2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton)","volume":"210 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2015-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131511386","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：