arXiv subjects

Atilla Eryilmaz

Publications and source records attributed to Atilla Eryilmaz.

At least 37 records · Page 2Linked to original sources

arXiv2020

Group-Fair Online Allocation in Continuous Time

The theory of discrete-time online learning has been successfully applied in many problems that involve sequential decision-making under uncertainty. However, in many applications including contractual hiring in online freelancing platforms and server allocation in cloud computing systems, the outcome of each action is observed only after a random and action-dependent time. Furthermore, as a consequence of certain ethical and economic concerns, the controller may impose deadlines on the completion of each task, and require fairness across different groups in the allocation of total time budget $B$. In order to address these applications, we consider continuous-time online learning problem with fairness considerations, and present a novel framework based on continuous-time utility maximization. We show that this formulation recovers reward-maximizing, max-min fair and proportionally fair allocation rules across different groups as special cases. We characterize the optimal offline policy, which allocates the total time between different actions in an optimally fair way (as defined by the utility function), and impose deadlines to maximize time-efficiency. In the absence of any statistical knowledge, we propose a novel online learning algorithm based on dual ascent optimization for time averages, and prove that it achieves $\tilde{O}(B^{-1/2})$ regret bound.

cs.LG↗

arXiv2020

Continuous-Time Multi-Armed Bandits with Controlled Restarts

Time-constrained decision processes have been ubiquitous in many fundamental applications in physics, biology and computer science. Recently, restart strategies have gained significant attention for boosting the efficiency of time-constrained processes by expediting the completion times. In this work, we investigate the bandit problem with controlled restarts for time-constrained decision processes, and develop provably good learning algorithms. In particular, we consider a bandit setting where each decision takes a random completion time, and yields a random and correlated reward at the end, with unknown values at the time of decision. The goal of the decision-maker is to maximize the expected total reward subject to a time constraint $τ$. As an additional control, we allow the decision-maker to interrupt an ongoing task and forgo its reward for a potentially more rewarding alternative. For this problem, we develop efficient online learning algorithms with $O(\log(τ))$ and $O(\sqrt{τ\log(τ)})$ regret in a finite and continuous action space of restart strategies, respectively. We demonstrate an applicability of our algorithm by using it to boost the performance of SAT solvers.

cs.LG↗

arXiv2020

Budget-Constrained Bandits over General Cost and Reward Distributions

We consider a budget-constrained bandit problem where each arm pull incurs a random cost, and yields a random reward in return. The objective is to maximize the total expected reward under a budget constraint on the total cost. The model is general in the sense that it allows correlated and potentially heavy-tailed cost-reward pairs that can take on negative values as required by many applications. We show that if moments of order $(2+γ)$ for some $γ> 0$ exist for all cost-reward pairs, $O(\log B)$ regret is achievable for a budget $B>0$. In order to achieve tight regret bounds, we propose algorithms that exploit the correlation between the cost and reward of each arm by extracting the common information via linear minimum mean-square error estimation. We prove a regret lower bound for this problem, and show that the proposed algorithms achieve tight problem-dependent regret bounds, which are optimal up to a universal constant factor in the case of jointly Gaussian cost and reward pairs.

cs.LG↗

arXiv2019

Predictive Scheduling for Virtual Reality

A significant challenge for future virtual reality (VR) applications is to deliver high quality-of-experience, both in terms of video quality and responsiveness, over wireless networks with limited bandwidth. This paper proposes to address this challenge by leveraging the predictability of user movements in the virtual world. We consider a wireless system where an access point (AP) serves multiple VR users. We show that the VR application process consists of two distinctive phases, whereby during the first (proactive scheduling) phase the controller has uncertain predictions of the demand that will arrive at the second (deadline scheduling) phase. We then develop a predictive scheduling policy for the AP that jointly optimizes the scheduling decisions in both phases. In addition to our theoretical study, we demonstrate the usefulness of our policy by building a prototype system. We show that our policy can be implemented under Furion, a Unity-based VR gaming software, with minor modifications. Experimental results clearly show visible difference between our policy and the default one. We also conduct extensive simulation studies, which show that our policy not only outperforms others, but also maintains excellent performance even when the prediction of future user movements is not accurate.

cs.NI↗

arXiv2019

On the Performance of Mobility-Aware D2D Caching Networks

The increase in demand for spectrum-based services forms a bottleneck in wireless networks. Device-to-Device (D2D) caching networks tackle this problem by exploiting user's behavior predictability and the possibility of sharing data between them to alleviate the network congestion. However, capturing mobility statistics allows Service Providers (SPs) to enhance their caching strategies. In this work, we introduce a mobility-aware D2D caching network where SP harnesses user demand and mobility statistics to minimize the incurred service cost through an optimal caching policy. We investigate two caching schemes: centralized and decentralized caching schemes. In the centralized caching scheme, SP makes the caching decision towards its cost minimization to increase its profit. However, the complexity of optimal caching policy grows exponentially with the number of users. Therefore, we discuss a greedy caching algorithm which has a polynomial order complexity. We also use this greedy algorithm to establish upper and lower bounds on the proactive service gain achieved by the optimal caching policy. In the decentralized caching scheme, users take over and make their caching decisions, in a distributed fashion affected by the SP pricing policy, towards their payment minimization. We formulated the tension between the SP and users as a Stackelberg game. The Best response analysis was used to identify a sub-game perfect Nash equilibrium (SPNE) between users. The optimal solution of proposed model was found to depend on the SP reward preference, which affects the assigned memory in users' devices. We found some regimes for the reward value where the SPNE was non-unique. A fair allocation caching policy was adopted to choose one of these SPNEs. To understand the impact of user behavior, we investigated some special cases to explore how user's mobility statistics affect their caching decision.

cs.NI↗

arXiv2018

Optimal Learning for Dynamic Coding in Deadline-Constrained Multi-Channel Networks

We study the problem of serving randomly arriving and delay-sensitive traffic over a multi-channel communication system with time-varying channel states and unknown statistics. This problem deviates from the classical exploration-exploitation setting in that the design and analysis must accommodate the dynamics of packet availability and urgency as well as the cost of each channel use at the time of decision. To that end, we have developed and investigated an index-based policy UCB-Deadline, which performs dynamic channel allocation decisions that incorporate these traffic requirements and costs. Under symmetric channel conditions, we have proved that the UCB-Deadline policy can achieve bounded regret in the likely case where the cost of using a channel is not too high to prevent all transmissions, and logarithmic regret otherwise. In this case, we show that UCB-Deadline is order-optimal. We also perform numerical investigations to validate the theoretical findings, and also compare the performance of the UCB-Deadline to another learning algorithm that we propose based on Thompson Sampling.

cs.NI↗

arXiv2018

Proactive Resource Allocation with Predictable Channel Statistics

The behavior of users in relatively predictable, both in terms of the data they request and the wireless channels they observe. In this paper, we consider the statistics of such predictable patterns of the demand and channel jointly across multiple users, and develop a novel predictive resource allocation method. This method is shown to provide performance benefits over a reactive approach, which ignores these patterns and instead aims to satisfy the instantaneous demands, irrespective of cost to the system. In particular, we show that our proposed method is able to attain a novel fundamental bound on the achievable cost, as the service window grows. Through numerical evaluation, we gain insights into how different uncertainty sources affect the decisions and the cost.

cs.IT↗

arXiv2017

Reward Maximization Under Uncertainty: Leveraging Side-Observations on Networks

We study the stochastic multi-armed bandit (MAB) problem in the presence of side-observations across actions that occur as a result of an underlying network structure. In our model, a bipartite graph captures the relationship between actions and a common set of unknowns such that choosing an action reveals observations for the unknowns that it is connected to. This models a common scenario in online social networks where users respond to their friends' activity, thus providing side information about each other's preferences. Our contributions are as follows: 1) We derive an asymptotic lower bound (with respect to time) as a function of the bi-partite network structure on the regret of any uniformly good policy that achieves the maximum long-term average reward. 2) We propose two policies - a randomized policy; and a policy based on the well-known upper confidence bound (UCB) policies - both of which explore each action at a rate that is a function of its network position. We show, under mild assumptions, that these policies achieve the asymptotic lower bound on the regret up to a multiplicative factor, independent of the network structure. Finally, we use numerical examples on a real-world social network and a routing example network to demonstrate the benefits obtained by our policies over other existing policies.

cs.LG↗

arXiv2015

Low-complexity Optimal Scheduling over Correlated Fading Channels with ARQ Feedback

We investigate the downlink scheduling problem under Markovian ON/OFF fading channels, where the instantaneous channel state information is not directly accessible, but is revealed via ARQ-type feedback. The scheduler can exploit the temporal correlation/channel memory inherent in the Markovian channels to improve network performance. However, designing low-complexity and throughput-optimal algorithms under temporal correlation is a challenging problem. In this paper, we find that under an average number of transmissions constraint, a low-complexity index policy is throughput-optimal. The policy uses Whittle's index value, which was previously used to capture opportunistic scheduling under temporally correlated channels. Our results build on the interesting finding that, under the intricate queue length and channel memory evolutions, the importance of scheduling a user is captured by a simple multiplication of its queue length and Whittle's index value. The proposed queue-based index policy has provably low complexity. Numerical results show that significant throughput gains can be realized by exploiting the channel memory using the proposed low-complexity policy.

cs.NI↗

arXiv2015

Downlink Scheduling over Markovian Fading Channels

We consider the scheduling problem in downlink wireless networks with heterogeneous, Markov-modulated, ON/OFF channels. It is well-known that the performance of scheduling over fading channels relies heavily on the accuracy of the available Channel State Information (CSI), which is costly to acquire. Thus, we consider the CSI acquisition via a practical ARQ-based feedback mechanism whereby channel states are revealed at the end of only scheduled users' transmissions. In the assumed presence of temporally-correlated channel evolutions, the desired scheduler must optimally balance the exploitation-exploration trade-off, whereby it schedules transmissions both to exploit those channels with up-to-date CSI and to explore the current state of those with outdated CSI. In earlier works, Whittle's Index Policy had been suggested as a low-complexity and high-performance solution to this problem. However, analyzing its performance in the typical scenario of statistically heterogeneous channel state processes has remained elusive and challenging, mainly because of the highly-coupled and complex dynamics it possesses. In this work, we overcome these difficulties to rigorously establish the asymptotic optimality properties of Whittle's Index Policy in the limiting regime of many users. More specifically: (1) we prove the local optimality of Whittle's Index Policy, provided that the initial state of the system is within a certain neighborhood of a carefully selected state; (2) we then establish the global optimality of Whittle's Index Policy under a recurrence assumption that is verified numerically for our problem. These results establish that Whittle's Index Policy possesses analytically provable optimality characteristics for scheduling over heterogeneous and temporally-correlated channels.

cs.NI↗

arXiv2014

Proactive Data Download and User Demand Shaping for Data Networks

In this work, we propose and study optimal proactive resource allocation and demand shaping for data networks. Motivated by the recent findings on the predictability of human behavior patterns in data networks, and the emergence of highly capable handheld devices, our design aims to smooth out the network traffic over time and minimize the data delivery costs. Our framework utilizes proactive data services as well as smart content recommendation schemes for shaping the demand. Proactive data services take place during the off-peak hours based on a statistical prediction of a demand profile for each user, whereas smart content recommendation assigns modified valuations to data items so as to render the users' demand less uncertain. Hence, our recommendation scheme aims to boost the performance of proactive services within the allowed flexibility of user requirements. We conduct theoretical performance analysis that quantifies the leveraged cost reduction through the proposed framework. We show that the cost reduction scales at the same rate as the cost function scales with the number of users. Further, we prove that \emph{demand shaping} through smart recommendation strictly reduces the incurred cost even below that of proactive downloads without recommendation.

cs.IT↗

arXiv2014

Throughput-Optimal Scheduling Design with Regular Service Guarantees in Wireless Networks

Motivated by the regular service requirements of video applications for improving Quality-of-Experience (QoE) of users, we consider the design of scheduling strategies in multi-hop wireless networks that not only maximize system throughput but also provide regular inter-service times for all links. Since the service regularity of links is related to the higher-order statistics of the arrival process and the policy operation, it is highly challenging to characterize and analyze directly. We overcome this obstacle by introducing a new quantity, namely the time-since-last-service (TSLS), which tracks the time since the last service. By combining it with the queue-length in the weight, we propose a novel maximum-weight type scheduling policy, called Regular Service Guarantee (RSG) Algorithm. The unique evolution of the TSLS counter poses significant challenges for the analysis of the RSG Algorithm. To tackle these challenges, we first propose a novel Lyapunov function to show the throughput optimality of the RSG Algorithm. Then, we prove that the RSG Algorithm can provide service regularity guarantees by using the Lyapunov-drift based analysis of the steady-state behavior of the stochastic processes. In particular, our algorithm can achieve a degree of service regularity within a factor of a fundamental lower bound we derive. This factor is a function of the system statistics and design parameters and can be as low as two in some special networks. Our results, both analytical and numerical, exhibit significant service regularity improvements over the traditional throughput-optimal policies, which reveals the importance of incorporating the metric of time-since-last-service into the scheduling policy for providing regulated service.

cs.NI↗

arXiv2014

Scheduling in Time-correlated Wireless Networks with Imperfect CSI and Stringent Constraint

In a wireless network, the efficiency of scheduling algorithms over time-varying channels depends heavily on the accuracy of the Channel State Information (CSI), which is usually quite ``costly'' in terms of consuming network resources. Scheduling in such systems is also subject to stringent constraints such as power and bandwidth, which limit the maximum number of simultaneous transmissions. In the meanwhile, communication channels in wireless systems typically fluctuate in a time-correlated manner. We hence design schedulers to exploit the temporal-correlation inherent in channels with memory and ARQ-styled feedback from the users for better channel state knowledge, under the assumption of Markovian channels and the stringent constraint on the maximum number of simultaneously active users. We model this problem under the framework of a Partially Observable Markov Decision Processes. In recent work, a low-complexity optimal solution was developed for this problem under a long-term time-average resource constraint. However, in real systems with instantaneous resource constraints, how to optimally exploit the temporal correlation and satisfy realistic stringent constraint on the instantaneous service remains elusive. In this work, we incorporate a stringent constraint on the simultaneously scheduled users and propose a low-complexity scheduling algorithm that dynamically implements user scheduling and dummy packet broadcasting. We show that the throughput region of the optimal policy under the long-term average resource constraint can be asymptotically achieved in the stringent constrained scenario by the proposed algorithm, in the many users limiting regime.

cs.NI↗

arXiv2013

Asymptotically Tight Steady-State Queue Length Bounds Implied By Drift Conditions

The Foster-Lyapunov theorem and its variants serve as the primary tools for studying the stability of queueing systems. In addition, it is well known that setting the drift of the Lyapunov function equal to zero in steady-state provides bounds on the expected queue lengths. However, such bounds are often very loose due to the fact that they fail to capture resource pooling effects. The main contribution of this paper is to show that the approach of "setting the drift of a Lyapunov function equal to zero" can be used to obtain bounds on the steady-state queue lengths which are tight in the heavy-traffic limit. The key is to establish an appropriate notion of state-space collapse in terms of steady-state moments of weighted queue length differences, and use this state-space collapse result when setting the Lyapunov drift equal to zero. As an application of the methodology, we prove the steady-state equivalent of the heavy-traffic optimality result of Stolyar for wireless networks operating under the MaxWeight scheduling policy.

math.PR↗

arXiv2012

A Fast-CSMA Algorithm for Deadline-Constrained Scheduling over Wireless Fading Channels

Recently, low-complexity and distributed Carrier Sense Multiple Access (CSMA)-based scheduling algorithms have attracted extensive interest due to their throughput-optimal characteristics in general network topologies. However, these algorithms are not well-suited for serving real-time traffic under time-varying channel conditions for two reasons: (1) the mixing time of the underlying CSMA Markov Chain grows with the size of the network, which, for large networks, generates unacceptable delay for deadline-constrained traffic; (2) since the dynamic CSMA parameters are influenced by the arrival and channel state processes, the underlying CSMA Markov Chain may not converge to a steady-state under strict deadline constraints and fading channel conditions. In this paper, we attack the problem of distributed scheduling for serving real-time traffic over time-varying channels. Specifically, we consider fully-connected topologies with independently fading channels (which can model cellular networks) in which flows with short-term deadline constraints and long-term drop rate requirements are served. To that end, we first characterize the maximal set of satisfiable arrival processes for this system and, then, propose a Fast-CSMA (FCSMA) policy that is shown to be optimal in supporting any real-time traffic that is within the maximal satisfiable set. These theoretical results are further validated through simulations to demonstrate the relative efficiency of the FCSMA policy compared to some of the existing CSMA-based algorithms.

cs.NI↗

arXiv2011

Exploiting Channel Memory for Joint Estimation and Scheduling in Downlink Networks

We address the problem of opportunistic multiuser scheduling in downlink networks with Markov-modeled outage channels. We consider the scenario in which the scheduler does not have full knowledge of the channel state information, but instead estimates the channel state information by exploiting the memory inherent in the Markov channels along with ARQ-styled feedback from the scheduled users. Opportunistic scheduling is optimized in two stages: (1) Channel estimation and rate adaptation to maximize the expected immediate rate of the scheduled user; (2) User scheduling, based on the optimized immediate rate, to maximize the overall long term sum-throughput of the downlink. The scheduling problem is a partially observable Markov decision process with the classic 'exploitation vs exploration' trade-off that is difficult to quantify. We therefore study the problem in the framework of Restless Multi-armed Bandit Processes (RMBP) and perform a Whittle's indexability analysis. Whittle's indexability is traditionally known to be hard to establish and the index policy derived based on Whittle's indexability is known to have optimality properties in various settings. We show that the problem of downlink scheduling under imperfect channel state information is Whittle indexable and derive the Whittle's index policy in closed form. Via extensive numerical experiments, we show that the index policy has near-optimal performance. Our work reveals that, under incomplete channel state information, exploiting channel memory for opportunistic scheduling can result in significant performance gains and that almost all of these gains can be realized using an easy-to-implement index policy.

cs.NI↗

arXiv2011

Proactive Resource Allocation: Harnessing the Diversity and Multicast Gains

This paper introduces the novel concept of proactive resource allocation through which the predictability of user behavior is exploited to balance the wireless traffic over time, and hence, significantly reduce the bandwidth required to achieve a given blocking/outage probability. We start with a simple model in which the smart wireless devices are assumed to predict the arrival of new requests and submit them to the network T time slots in advance. Using tools from large deviation theory, we quantify the resulting prediction diversity gain} to establish that the decay rate of the outage event probabilities increases with the prediction duration T. This model is then generalized to incorporate the effect of the randomness in the prediction look-ahead time T. Remarkably, we also show that, in the cognitive networking scenario, the appropriate use of proactive resource allocation by the primary users improves the diversity gain of the secondary network at no cost in the primary network diversity. We also shed lights on multicasting with predictable demands and show that the proactive multicast networks can achieve a significantly higher diversity gain that scales super-linearly with T. Finally, we conclude by a discussion of the new research questions posed under the umbrella of the proposed proactive (non-causal) wireless networking framework.

cs.IT↗

arXiv2011

Throughput-Delay Analysis of Random Linear Network Coding for Wireless Broadcasting

In an unreliable single-hop broadcast network setting, we investigate the throughput and decoding-delay performance of random linear network coding as a function of the coding window size and the network size. Our model consists of a source transmitting packets of a single flow to a set of $n$ users over independent erasure channels. The source performs random linear network coding (RLNC) over $k$ (coding window size) packets and broadcasts them to the users. We note that the broadcast throughput of RLNC must vanish with increasing $n$, for any fixed $k.$ Hence, in contrast to other works in the literature, we investigate how the coding window size $k$ must scale for increasing $n$. Our analysis reveals that the coding window size of $Θ(\ln(n))$ represents a phase transition rate, below which the throughput converges to zero, and above which it converges to the broadcast capacity. Further, we characterize the asymptotic distribution of decoding delay and provide approximate expressions for the mean and variance of decoding delay for the scaling regime of $k=ω(\ln(n)).$ These asymptotic expressions reveal the impact of channel correlations on the throughput and delay performance of RLNC. We also show how our analysis can be extended to other rateless block coding schemes such as the LT codes. Finally, we comment on the extension of our results to the cases of dependent channels across users and asymmetric channel model.

cs.IT↗