There is, I think, something ethereal about i —the square root of minus one. I remember first hearing about it at school. It seemed an odd beast at that time—an intruder hovering on the edge of reality.

Usually familiarity dulls this sense of the bizarre, but in the case of i it was the reverse: over the years the sense of its surreal nature intensified. It seemed that it was impossible to write mathematics that described the real world in …

I and i

Machine Learning is the study of methods for programming computers to learn. Computers are applied to a wide range of tasks, and for most of these it is relatively easy for programmers to design and implement the necessary software. However, there are many tasks for which this is difficult or impossible. These can be divided into four general categories. First, there are problems for which there exist no human experts. For example, in modern automated manufacturing facilities, there is a need to predict machine failures before they occur by analyzing sensor readings. Because the machines are new, there are no human experts who can be interviewed by a programmer to provide the knowledge necessary to build a computer system. A machine learning system can study recorded data and subsequent machine failures and learn prediction rules. Second, there are problems where human experts exist, but where they are unable to explain their expertise. This is the case in many perceptual tasks, such as speech recognition, hand-writing recognition, and natural language understanding. Virtually all humans exhibit expert-level abilities on these tasks, but none of them can describe the detailed steps that they follow as they perform them. Fortunately, humans can provide machines with examples of the inputs and correct outputs for these tasks, so machine learning algorithms can learn to map the inputs to the outputs. Third, there are problems where phenomena are changing rapidly. In finance, for example, people would like to predict the future behavior of the stock market, of consumer purchases, or of exchange rates. These behaviors change frequently, so that even if a programmer could construct a good predictive computer program, it would need to be rewritten frequently. A learning program can relieve the programmer of this burden by constantly modifying and tuning a set of learned prediction rules. Fourth, there are applications that need to be customized for each computer user separately. Consider, for example, a program to filter unwanted electronic mail messages. Different users will need different filters. It is unreasonable to expect each user to program his or her own rules, and it is infeasible to provide every user with a software engineer to keep the rules up-to-date. A machine learning system can learn which mail messages the user rejects and maintain the filtering rules automatically. Machine learning addresses many of the same research questions as the fields of statistics, data mining, and psychology, but with differences of emphasis. Statistics focuses on understanding the phenomena that have generated the data, often with the goal of testing different hypotheses about those phenomena. Data mining seeks to find patterns in the data that are understandable by people. Psychological studies of human learning aspire to understand the mechanisms underlying the various learning behaviors exhibited by people (concept learning, skill acquisition, strategy change, etc.).

Machine learning

We will review some of the major results in random graphs and some of the more challenging open problems. We will cover algorithmic and structural questions. We will touch on newer models, including those related to the WWW.

Random graphs

Convergence of Probability Measures. By P. Billingsley. Chichester, Sussex, Wiley, 1968. xii, 253 p. 9 1/4“. 117s.

Convergence of Probability Measures

“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

Network virtualization is a powerful way to run multiple architectures or experiments simultaneously on a shared infrastructure. However, making efficient use of the underlying resources requires effective techniques for virtual network embedding--mapping each virtual network to specific nodes and links in the substrate network. Since the general embedding problem is computationally intractable, past research restricted the problem space to allow efficient solutions, or focused on designing heuristic algorithms. In this paper, we advocate a different approach: rethinking the design of the substrate network to enable simpler embedding algorithms and more efficient use of resources, without restricting the problem space. In particular, we simplify virtual link embedding by: i) allowing the substrate network to split a virtual link over multiple substrate paths and ii) employing path migration to periodically re-optimize the utilization of the substrate network. We also explore node-mapping algorithms that are customized to common classes of virtual-network topologies. Our simulation experiments show that path splitting, path migration,and customized embedding algorithms enable a substrate network to satisfy a much larger mix of virtual networks

/pdf/rethinking-virtual-network-embedding-substrate-support-for-2zvfvo1lxr.pdf

Rethinking virtual network embedding: substrate support for path splitting and migration

This paper presents a quantitative study on the performance of 3G mobile data offloading through WiFi networks. We recruited 97 iPhone users from metropolitan areas and collected statistics on their WiFi connectivity during a two-and-a-half-week period in February 2010. Our trace-driven simulation using the acquired whole-day traces indicates that WiFi already offloads about 65% of the total mobile data traffic and saves 55% of battery power without using any delayed transmission. If data transfers can be delayed with some deadline until users enter a WiFi zone, substantial gains can be achieved only when the deadline is fairly larger than tens of minutes. With 100-s delays, the achievable gain is less than only 2%-3%, whereas with 1 h or longer deadlines, traffic and energy saving gains increase beyond 29% and 20%, respectively. These results are in contrast to the substantial gain (20%-33%) reported by the existing work even for 100-s delayed transmission using traces taken from transit buses or war-driving. In addition, a distribution model-based simulator and a theoretical framework that enable analytical studies of the average performance of offloading are proposed. These tools are useful for network providers to obtain a rough estimate on the average performance of offloading for a given WiFi deployment condition.

Mobile data offloading: how much can WiFi deliver?

Energy-efficiency, one of the major design goals in wireless cellular networks, has received much attention lately, due to increased awareness of environmental and economic issues for network operators. In this paper, we develop a theoretical framework for BS energy saving that encompasses dynamic BS operation and the related problem of user association together. Specifically, we formulate a total cost minimization that allows for a flexible tradeoff between flow-level performance and energy consumption. For the user association problem, we propose an optimal energy-efficient user association policy and further present a distributed implementation with provable convergence. For the BS operation problem (i.e., BS switching on/off), which is a challenging combinatorial problem, we propose simple greedy-on and greedy-off algorithms that are inspired by the mathematical background of submodularity maximization problem. Moreover, we propose other heuristic algorithms based on the distances between BSs or the utilizations of BSs that do not impose any additional signaling overhead and thus are easy to implement in practice. Extensive simulations under various practical configurations demonstrate that the proposed user association and BS operation algorithms can significantly reduce energy consumption.

/pdf/base-station-operation-and-user-association-mechanisms-for-ps5xv62lec.pdf

Base Station Operation and User Association Mechanisms for Energy-Delay Tradeoffs in Green Cellular Networks

We explore value-based solutions for multi-agent reinforcement learning (MARL) tasks in the centralized training with decentralized execution (CTDE) regime popularized recently. However, VDN and QMIX are representative examples that use the idea of factorization of the joint action-value function into individual ones for decentralized execution. VDN and QMIX address only a fraction of factorizable MARL tasks due to their structural constraint in factorization such as additivity and monotonicity. In this paper, we propose a new factorization method for MARL, QTRAN, which is free from such structural constraints and takes on a new approach to transforming the original joint action-value function into an easily factorizable one, with the same optimal actions. QTRAN guarantees more general factorization than VDN or QMIX, thus covering a much wider class of MARL tasks than does previous methods. Our experiments for the tasks of multi-domain Gaussian-squeeze and modified predator-prey demonstrate QTRAN's superior performance with especially larger margins in games whose payoffs penalize non-cooperative behavior more aggressively.

/pdf/qtran-learning-to-factorize-with-transformation-for-5gt8vi6gwk.pdf

QTRAN: Learning to Factorize with Transformation for Cooperative Multi-Agent Reinforcement Learning.

This paper focuses on congestion control over multi-hop, wireless networks. In a wireless network, an important constraint that arises is that due to the MAC (Media Access Control) layer. Many wireless MACs use a time-division strategy for channel access, where, at any point in space, the physical channel can be accessed by a single user at each instant of time. In this paper, we develop a fair hop-by-hop congestion control algorithm with the MAC constraint being imposed in the form of a channel access time constraint, using an optimization-based framework. In the absence of delay, we show that this algorithm are globally stable using a Lyapunov-function-based approach. Next, in the presence of delay, we show that the hop-by-hop control algorithm has the property of spatial spreading. In other words, focused loads at a particular spatial location in the network get "smoothed" over space. We derive bounds on the "peak load" at a node, both with hop-by-hop control, as well as with end-to-end control, show that significant gains are to be had with the hop-by-hop scheme, and validate the analytical results with simulation.

/pdf/hop-by-hop-congestion-control-over-a-wireless-multi-hop-4ilxn0zovu.pdf

Yung Yi

Papers

Rethinking virtual network embedding: substrate support for path splitting and migration

Mobile data offloading: how much can WiFi deliver?

Base Station Operation and User Association Mechanisms for Energy-Delay Tradeoffs in Green Cellular Networks

QTRAN: Learning to Factorize with Transformation for Cooperative Multi-Agent Reinforcement Learning.

Hop-by-hop congestion control over a wireless multi-hop network