Search

The field of Weakly Supervised Learning (WSL) has recently seen a surge of popularity, with numerous papers addressing different types of "supervision deficiencies". In WSL use cases, a variety of situations exists where the collected "information" is imperfect. The paradigm of WSL attempts to list and cover these problems with associated solutions. In this paper, we review the research progress on WSL with the aim to make it as a brief introduction to this field. We present the three axis of WSL cube and an overview of most of all the elements of their facets. We propose three measurable quantities that acts as coordinates in the previously defined cube namely: Quality, Adaptability and Quantity of information. Thus we suggest that Biquality Learning framework can be defined as a plan of the WSL cube and propose to re-discover previously unrelated patches in WSL literature as a unified Biquality Learning literature.

The field of Weakly Supervised Learning (WSL) has recently seen a surge of popularity, with numerous papers addressing different types of "supervision deficiencies", namely: poor quality, non adaptability, and insufficient quantity of labels. Regarding quality, label noise can be of different types, including completely-at-random, at-random or even not-at-random. All these kinds of label noise are addressed separately in the literature, leading to highly specialized approaches. This paper proposes an original, encompassing, view of Weakly Supervised Learning, which results in the design of generic approaches capable of dealing with any kind of label noise. For this purpose, an alternative setting called "Biquality data" is used. It assumes that a small trusted dataset of correctly labeled examples is available, in addition to an untrusted dataset of noisy examples. In this paper, we propose a new reweigthing scheme capable of identifying noncorrupted examples in the untrusted dataset. This allows one to learn classifiers using both datasets. Extensive experiments that simulate several types of label noise and that vary the quality and quantity of untrusted examples, demonstrate that the proposed approach outperforms baselines and state-of-the-art approaches., 8 pages, 7 figures

International audience; Worldwide telecommunications groups are both infrastructure operator and service provider. Hence, when planning the network expansion, these groups must also consider the subscribers dynamics, which they can influence through subsidies. Addressing both aspects together enables them to better optimize the network dimensioning, therefore avoiding unnecessary costs. In this work, the network expansion represents the deployment and/or reinforcement of several technologies (e.g. 2G,3G,4G), assuming that subscribers to a given technology can be served by this technology or older ones. The objective of the resulting optimization problem is to minimize network investments costs and subsides, while being subject to both capacity and strategical constraints, such as minimum coverage and users averaged throughput. We model the customer behavior in response to subsides with S-shape piecewise linear functions, which are linearized. We assess numerically the resulting Mixed-Integer Linear Programming (MILP) formulation on real-life instances focusing on 3G/4G migrations. Our results show the scalability of the MILP model for 2 network generations and 100 sites. Moreover, they underline the cost-benefit of solving a unique optimization problem over the whole time-horizon (5 years) compared to decomposing the problem year by year.

One of the challenges of cloud computing is to optimally and efficiently assign virtual ma- chines to physical machines. The aim of telecommunication operators is to minimize the map- ping cost while respecting constraints regarding location, assignment and capacity. In this paper we first propose an exact formulation leading to a 0-1 bilinear constrained problem. Then we introduce a variety of linear cuts by exploiting the problem structure and present a Lagrange decomposition based B&B algorithm to obtain optimal solutions efficiently. Numerically, we show that our valid inequalities close over 80% of the optimality gap incurred by the well-known McCormick relaxation, and demonstrate the computational advantage of the proposed B&B algorithm with extensive numerical experiments., submitted for journal publication; copyright preserved

International audience; We discuss two approaches to approximate the convex and concave envelopes of bilinear functions over hypercubes. The first approach is based on a semidefinite program. The second approach considers some predefined cover sets of a hypercube and leads to a linear program. Then we establish a connection between the convex envelope of a bilinear function and the concave envelope of a piecewise linear function. Numerical experiments are conducted to compare the two approaches. As an extension, a novel approach is discussed

We consider linear programs involving uncertain parameters and propose a new tractable robust counterpart which contains and generalizes several other models including the existing Affinely Adjustable Robust Counterpart and the Fully Adjustable Robust Counterpart. It consists in selecting a set of poles whose convex hull contains some projection of the uncertainty set, and computing a recourse strategy for each data scenario as a convex combination of some optimized recourses (one for each pole). We show that the proposed multipolar robust counterpart is tractable and its complexity is controllable. Further, we show that under some mild assumptions, two sequences of upper and lower bounds converge to the optimal value of the fully adjustable robust counterpart. To illustrate the approach, a robust problem related to lobbying under some uncertain opinions of authorities is studied. Several numerical experiments are carried out showing the advantages of the proposed robustness framework and evaluating the benefit of adaptability., 35 pages, 1 figure, 9 tables, submitted journal paper

Internet is a worldwide network composed of interconnected but independent networks, called Autonomous Systems. Each network owner has to decide which other networks to interconnect with and how to allocate its traffic among its providers. The financial flows between Autonomous Systems depend on these decisions and raise the key issue of revenue management. In this paper, we propose some models and exact methods for the joint optimization problem of interconnection policy and traffic allocation for a customer AS. This problem is analyzed in the top-percentile pricing framework for the interconnection agreements, and we assess the solution methods using real-life instances.

We analyze some issues of network design and bandwidth allocation in telecommunication systems with congestible resources. The work is closely related to network monitoring and traffic measurement functions that must be carried out on line, in order to overcome congestion caused by an unfavorable traffic pattern or by a failure. In addition to the traditional use of routing controls, our approach achieves network efficiency with capacity assignment and also imposing variable prices for the consumers. A generalized Benders decomposition method is applied to a mixed integer nonlinear programming formulation of the integrated problem of network design and operation. The method exploits the nature of the continuous subproblem, that is a large-scale convex network flow problem with demands sensitive to commodity prices. Some numerical experience suggests that the method is usefull to address both questions of global optimality and competitive pricing in such systems.

It is estimated that the ICT industry contains not less than one billion personal computers (PC), with power consumption ranging from 100 to 200 W per workstation. Virtualization and Thin Client technologies, mediated by broadband network links will have a very significant impact on efforts to reduce energy consumption and to enhance data security throughout the ICT industry. Besides, virtualization is quickly getting into the small and medium-sized business (SMB) market, promising a better (centralized) energy management with means of consolidation and resource reuse. Indeed, multiple virtual machines (VM) or platforms, running Virtual Desktops (VD), can be hosted on a single physical server, requiring only hypervisors and adequate desktop delivery protocols to deliver the VD service to distant users on any suitable device. In this paper, we propose to deliver a VD service over a federation of data centers (DC), sparsely connected to some nodes of the core network. The proposed model aims to provide an adequate resource dimensioning at the DC and the network while minimizing the inherent energy consumption. We prove by means of simulations the importance of user profiling and activity prediction in avoiding over-dimensioning, especially at the DCs. We also show the interest of energy-aware modes at the servers.

International audience; To optimize the quality of service through a telecommunication network, we propose an algorithm based on Lagrangian relaxation. The bundle-type dual algorithm is adapted to the present situation, where the dual function is the sum of a polyhedral function (coming from shortest path problems) and of a smooth function (coming from the congestion function).

The Grid is a promising concept to solve the dilemma of increasingly complex and demanding applications being confronted with the need for a more efficient and flexible use of existing computer resources. Even though Grid technologies have made progress within the context of large enterprises and academic projects, there has not yet been a widespread adoption by public institutions and small enterprises. One barrier to this adoption is the lack of economic paradigms which support the dynamic and efficient sharing of Grid resources by balancing resource scarity and idle capacities. Economic algorithms promise to provide a good fit to the Grid’s inherent strategic dimension by enabling users to express their valuation for computer resources. At the same time they provide incentives to contribute idle resources to the Grid in return for the market price.

International audience