- Berkeley, California, United States
The regret-minimization paradigm has emerged as a powerful technique for designing algorithms for online decision-making in adversarial environments. But so far, designing exact minmax-optimal algorithms for minimizing the worst-case... more
The regret-minimization paradigm has emerged as a powerful technique for designing algorithms for online decision-making in adversarial environments. But so far, designing exact minmax-optimal algorithms for minimizing the worst-case regret has proven to be a difficult task in general, with only a few known results in specific settings. In this paper, we present a novel set-valued dynamic programming approach for designing such exact regret-optimal policies for playing repeated games with discounted losses. Our approach first draws the connection between regret minimization, and determining minimal achievable guarantees in repeated games with vector-valued losses. We then characterize the set of these minimal guarantees as the fixed point of a dynamic programming operator defined on the space of Pareto frontiers of convex and compact sets. This approach simultaneously results in the characterization of the optimal strategies that achieve these minimal guarantees, and hence of regret...
Research Interests:
Research Interests:
Abstract : The first part of the work yielded new results on simulated annealing and neural networks while the second part focused on large deviations in high- speed communication networks. Neural Networks, Communication Networks,... more
Abstract : The first part of the work yielded new results on simulated annealing and neural networks while the second part focused on large deviations in high- speed communication networks. Neural Networks, Communication Networks, Simulated Annealing, High-Speed Networks.
Research Interests:
To provide statistical guarantees of QoS, the Internet requires a measurement infrastructure for estimating available resources from actual tra c. In this paper, we outline an algorithm that collects a histogram of the occupancy of a... more
To provide statistical guarantees of QoS, the Internet requires a measurement infrastructure for estimating available resources from actual tra c. In this paper, we outline an algorithm that collects a histogram of the occupancy of a single-server FCFS queue at packet arrival times, and infers the loss rate and delay distribution from such measurements. Direct estimation of such QoS parameters typically leads to estimators with a large variance. To reduce this variance, we t a bu er occupancy model, a sum of exponentials, to the histogram using a weighted least-squares algorithm. Furthermore, we compute batch means to minimize the bias due to the positive correlation between measurements. In this manner, we provide an e cient and robust approach to QoS estimation.
This paper is a brief tutorial on IEEE Time-Sensitive Networks. These networks are designed for applications where latency is critical, such as process control, automotive and aerospace, and augmented or virtual reality.
Research Interests:
We explore a dynamic approach to the problems of call admission and resource allocation for communication networks with connections that are differentiated by their quality of service requirements. In a dynamic approach, the amount of... more
We explore a dynamic approach to the problems of call admission and resource allocation for communication networks with connections that are differentiated by their quality of service requirements. In a dynamic approach, the amount of spare resources is estimated on-line based on feedbacks from the network's quality of service monitoring mechanism. The schemes we propose remove the dependence on accurate traffic models and thus simplify the tasks of supplying traffic statistics required of network users. In this paper we present two dynamic algorithms. The objective of these algorithms is to find the minimum bandwidth necessary to satisfy a cell loss probability constraint at an asynchronous transfer mode (ATM) switch. We show that in both schemes the bandwidth chosen by the algorithm approaches the optimal value almost surely. Furthermore, in the second scheme, which determines the point closest to the optimal bandwidth from a finite number of choices, the expected learning tim...
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
... By. Chen, Chih Liang and Mai, Tony K. Table of Contents. Page. ... sabotaging their own network is an example of insider attack. Denial of Service – An analogy of this problem is a bad guy preventing a good guy from. getting useful... more
... By. Chen, Chih Liang and Mai, Tony K. Table of Contents. Page. ... sabotaging their own network is an example of insider attack. Denial of Service – An analogy of this problem is a bad guy preventing a good guy from. getting useful work done. A good example of this problem is the ...
We describe an approximate dynamic programming (ADP) approach to compute approximations of the optimal strategies and of the minimal losses that can be guaranteed in discounted repeated games with vector-valued losses. Such games... more
We describe an approximate dynamic programming (ADP) approach to compute approximations of the optimal strategies and of the minimal losses that can be guaranteed in discounted repeated games with vector-valued losses. Such games prominently arise in the analysis of regret in repeated decision-making in adversarial environments, also known as adversarial online learning. At the core of our approach is a characterization of the lower Pareto frontier of the set of expected losses that a player can guarantee in these games as the unique fixed point of a set-valued dynamic programming operator. When applied to the problem of regret minimization with discounted losses, our approach yields algorithms that achieve markedly improved performance bounds compared to off-the-shelf online learning algorithms like Hedge. These results thus suggest the significant potential of ADP-based approaches in adversarial online learning.
Research Interests:
WiFi networks suffer from severe network utility degradation due to the usage of diverse modulation and coding schemes. The proportional-fair allocation, that has been shown to be a good remedy, can be enforced through the proper... more
WiFi networks suffer from severe network utility degradation due to the usage of diverse modulation and coding schemes. The proportional-fair allocation, that has been shown to be a good remedy, can be enforced through the proper selection of contention window values. This has been achieved so far for centralized systems by an explicit solution of an optimization problem or, as proposed recently, by following a learning-based approach. In this paper, we present the first fully distributed solution in which each of the WiFi nodes independently tunes its contention window to achieve proportional fairness. Our solution is therefore applicable also for a set of collocated, unconnected WiFi networks. We compare the throughput and air-time allocation that this algorithm achieves to the values achieved by standard WiFi binary exponential back-off and values achieved by known centralized algorithms.
Research Interests:
We describe an approximate dynamic programming (ADP) approach to compute approximately optimal strategies and approximations of the minimal losses that can be guaranteed in discounted repeated games with vector losses. At the core of our... more
We describe an approximate dynamic programming (ADP) approach to compute approximately optimal strategies and approximations of the minimal losses that can be guaranteed in discounted repeated games with vector losses. At the core of our approach is a characterization of the lower Pareto frontier of the set of expected losses that a player can guarantee in these games as the unique fixed point of a set-valued dynamic programming (DP) operator. This fixed point can be approximated by an iterative application of this DP operator compounded by a polytopic set approximation, beginning with a single point. Each iteration can be computed by solving a set of linear programs corresponding to the vertices of the polytope. We derive rigorous bounds on the error of the resulting approximation and the performance of the corresponding approximately optimal strategies. We discuss an application to regret minimization in repeated decision-making in adversarial environments, where we show that this...
Research Interests:
As vehicles get equipped with increasingly complex sensors and processors, the communication requirements become more demanding. Traditionally, vehicles have used specialized networking technologies designed to guarantee bounded... more
As vehicles get equipped with increasingly complex sensors and processors, the communication requirements become more demanding. Traditionally, vehicles have used specialized networking technologies designed to guarantee bounded latencies, such a the Controller Area Network (CAN) bus. Recently, some have used dedicated technologies to transport signals from cameras, lidars, radars, and ultrasonic sensors. In parallel, IEEE working groups are defining Ethernet standards for time-sensitive networks (TSN). This paper describes an Ethernet-based architecture with provable guaranteed performance and simple configuration that is suitable for supporting the communication requirements of many vehicles.
Research Interests:
The book is designed for a junior/senior level course. Applications drive the material: PageRank, Multiplexing, Digital Link, Tracking, Speech Recognition, Route Planning and more. Topics include Markov chains, detection, coding,... more
The book is designed for a junior/senior level course. Applications drive the material: PageRank, Multiplexing, Digital Link, Tracking, Speech Recognition, Route Planning and more. Topics include Markov chains, detection, coding, estimation, Viterbi algorithm, expectation maximization, clustering, compressed sensing, recommender systems, Kalman Filter, Markov decision problems, LQG, and channel capacity. Matlab examples are used to simulate models and to implement the algorithms. Appendices provide the necessary background in basic probability and linear algebra. See https://sites.google.com/site/walrandpeecs/home.
Research Interests:
This paper proposes a proof of the convergence of a distributed and asynchronous version of the Kiefer-Wolfowitz algorithm where the agents do not exchange information with one another.
Research Interests:
Research Interests:
The objective of this note is to explore the existence of an effective bandwidth for ATM (asynchronous transfer mode) traffic. We show that such an effective bandwidth exists for a class of stationary Gaussian sources. We also show that... more
The objective of this note is to explore the existence of an effective bandwidth for ATM (asynchronous transfer mode) traffic. We show that such an effective bandwidth exists for a class of stationary Gaussian sources. We also show that the effective bandwidth cannot be defined for general non-stationary sources by providing an elemen- tary counter-example.
Research Interests:
We consider a fork-join network with arrival and service times forming a stationary and ergodic process. The usual stability condition, namely that the input rate is strictly less than all the service rates, is proved to be valid in this... more
We consider a fork-join network with arrival and service times forming a stationary and ergodic process. The usual stability condition, namely that the input rate is strictly less than all the service rates, is proved to be valid in this general case. Finally we extend the result to the case where there is random routing.
Research Interests:
Modern processing networks often consist of heterogeneous servers with widely varying capabilities, and process job flows with complex structure and requirements. A major challenge in designing efficient scheduling policies in these... more
Modern processing networks often consist of heterogeneous servers with widely varying capabilities, and process job flows with complex structure and requirements. A major challenge in designing efficient scheduling policies in these networks is the lack of reliable estimates of system parameters, and an attractive approach for addressing this challenge is to design robust policies, i.e. policies that do not use system parameters such as arrival and/or service rates for making scheduling decisions. In this paper we propose a general framework for the design of robust policies. The main technical novelty is the use of a stochastic gradient projection method that reacts to queue-length changes in order to find a balanced allocation of service resources to incoming tasks. We illustrate our approach on two broad classes of processing systems, namely the flexible fork-join networks and the flexible queueing networks, and prove the rate stability of our proposed policies for these networks...
Research Interests:
Many distributed multiple access (MAC) protocols use an exponential backoff mechanism. In that mechanism, a node picks a random backoff time uniformly in an interval that doubles in size after a collision. When used in an ad-hoc network... more
Many distributed multiple access (MAC) protocols use an exponential backoff mechanism. In that mechanism, a node picks a random backoff time uniformly in an interval that doubles in size after a collision. When used in an ad-hoc network spanning multiple interference domains, this backoff mechanism is unfair towards nodes in the middle of the network. Indeed, such nodes tend to
Research Interests:
We propose some improvements of TCP Vegas and compare its performance characteristics with TCP Reno. We argue through analysis that TCP Vegas, with its better bandwidth estimation scheme, uses the network resources more efficiently and... more
We propose some improvements of TCP Vegas and compare its performance characteristics with TCP Reno. We argue through analysis that TCP Vegas, with its better bandwidth estimation scheme, uses the network resources more efficiently and fairly than TCP Reno. Simulation results are given that support the results of the analysis.
Research Interests:
ABSTRACT
Research Interests:
Research Interests:
ABSTRACT
Research Interests:
ABSTRACT
Research Interests:
The cμ rule is optimal for arbitrary arrival processes provided that the service times are geometric and the service discipline is preemptive.
Research Interests:
In this paper, we present a prototype of an overlay network that provides Quality of Service routing for ows in an ad- hoc wireless network. A distributed scheduler offers MAC layer reservation capabilities and a best effort trafc... more
In this paper, we present a prototype of an overlay network that provides Quality of Service routing for ows in an ad- hoc wireless network. A distributed scheduler offers MAC layer reservation capabilities and a best effort trafc maximization algorithm completes the architecture. We use network measurements and simulation to improve the decisions. The design is tested through a working
Research Interests:
Research Interests:
Research Interests:
We consider joint traffic management of Variable andAvailable Bit Rate (VBR,ABR) services for ATM networks.For VBR service we consider loss sensitive calladmission for virtual trunks of such traffic. For ABRtraffic we propose a flow... more
We consider joint traffic management of Variable andAvailable Bit Rate (VBR,ABR) services for ATM networks.For VBR service we consider loss sensitive calladmission for virtual trunks of such traffic. For ABRtraffic we propose a flow control mechanism based onmeasurement of the current queue length and bandwidthavailability, as well as tracking the current activitylevel of sources, to adjust an explicit bound onthe source transmission rates. Integration is consideredby assessing the...
Research Interests:
Research Interests:
This paper provides proofs of the rate stability, Harris recurrence, and epsilon-optimality of CSMA algorithms where the backoff parameter of each node is based on its backlog. These algorithms require only local information and are easy... more
This paper provides proofs of the rate stability, Harris recurrence, and epsilon-optimality of CSMA algorithms where the backoff parameter of each node is based on its backlog. These algorithms require only local information and are easy to implement. The setup is a network of wireless nodes with a fixed conflict graph that identifies pairs of nodes whose simultaneous transmissions conflict.
Research Interests:
Research Interests:
Research Interests:
ABSTRACT Consider a counting process {N(t)>0}. In the one-dimensional case it is shown that the set of distributions of N(t) for t>0. This result is a "marginal"version of the characterization of a point... more
ABSTRACT Consider a counting process {N(t)>0}. In the one-dimensional case it is shown that the set of distributions of N(t) for t>0. This result is a "marginal"version of the characterization of a point process by its stochastic intensity. It is shown that the correspondng results have to be modifed in the multi-dimensional case. These results are particularized to the case of point processes with Poisson marginals.
Research Interests:
The capacity of an arbitrary ad-hoc network is difficult to estimate due to interference between the links. We use a conflict graph that models this interference relationship to determine if a set of flow rates can be accommodated. Us-... more
The capacity of an arbitrary ad-hoc network is difficult to estimate due to interference between the links. We use a conflict graph that models this interference relationship to determine if a set of flow rates can be accommodated. Us- ing the cliques (complete subgraphs) of the conflict graph, we derive constraints that are sufficient for a set of flow rates
Research Interests:
One of the challenges facing the networking industry today is to increase the profitability of Internet services. This calls for economic mechanisms that can enable providers to charge more for better services and collect a fair share of... more
One of the challenges facing the networking industry today is to increase the profitability of Internet services. This calls for economic mechanisms that can enable providers to charge more for better services and collect a fair share of the increased revenues. In this paper, we present a generic pricing model for Internet services jointly offered by a group of providers. We show that noncooperative pricing strategies may lead to unfair distribution of profit and may even discourage future upgrades to the network. As an alternative, we propose a fair revenue-sharing policy based on the weighted proportional fairness criterion. We show that this fair allocation policy encourages collaboration among providers, and hence can produce higher profits for all providers. Based on the analysis, we suggest a scalable algorithm for providers to implement this policy in a distributed way and study its convergence property.