HAL - Lille 3
Not a member yet
5967 research outputs found
Sort by
Hybrid Collaborative Filtering with Autoencoders
No full textCollaborative Filtering aims at exploiting the feedback of users to provide personalised recommendations. Such algorithms look for latent variables in a large sparse matrix of ratings. They can be enhanced by adding side information to tackle the well-known cold start problem. While Neu-ral Networks have tremendous success in image and speech recognition, they have received less attention in Collaborative Filtering. This is all the more surprising that Neural Networks are able to discover latent variables in large and heterogeneous datasets. In this paper, we introduce a Collaborative Filtering Neural network architecture aka CFN which computes a non-linear Matrix Factorization from sparse rating inputs and side information. We show experimentally on the MovieLens and Douban dataset that CFN outper-forms the state of the art and benefits from side information. We provide an implementation of the algorithm as a reusable plugin for Torch, a popular Neural Network framework
Gossip Dual Averaging for Decentralized Optimization of Pairwise Functions
No full textInternational audienceIn decentralized networks (of sensors, connected objects, etc.), there is an important need for efficient algorithms to optimize a global cost function , for instance to learn a global model from the local data collected by each computing unit. In this paper, we address the problem of decentralized minimization of pairwise functions of the data points, where these points are distributed over the nodes of a graph defining the communication topology of the network. This general problem finds applications in ranking, distance metric learning and graph inference, among others. We propose new gossip algorithms based on dual averaging which aims at solving such problems both in synchronous and asynchronous settings. The proposed framework is flexible enough to deal with constrained and regularized variants of the optimization problem. Our theoretical analysis reveals that the proposed algorithms preserve the convergence rate of centralized dual averaging up to an additive bias term. We present numerical simulations on Area Under the ROC Curve (AUC) maximization and metric learning problems which illustrate the practical interest of our approach
Combinatorial optimization in networks with Shared Risk Link Groups
No full textInternational audienceThe notion of Shared Risk Link Groups (SRLG) captures survivability issues when a set of links of a network may fail simultaneously. The theory of survivable network design relies on basic combinatorial objects that are rather easy to compute in the classical graph models: shortest paths, minimum cuts, or pairs of disjoint paths. In the SRLG context, the optimization criterion for these objects is no longer the number of edges they use, but the number of SRLGs involved. Unfortunately, computing these combinatorial objects is NP-hard and hard to approximate with this objective in general. Nevertheless some objects can be computed in polynomial time when the SRLGs satisfy certain structural properties of locality which correspond to practical ones, namely the star property (all links affected by a given SRLG are incident to a unique node) and the span 1 property (the links affected by a given SRLG form a connected component of the network). The star property is defined in a multi-colored model where a link can be affected by several SRLGs while the span property is defined only in a mono-colored model where a link can be affected by at most one SRLG. In this paper, we extend these notions to characterize new cases in which these optimization problems can be solved in polynomial time. We also investigate the computational impact of the transformation from the multi-colored model to the mono-colored one. Experimental results are presented to validate the proposed algorithms and principles
A Decentralized and Robust Protocol for Private Averaging over Highly Distributed Data
No full textInternational audienceWe propose a decentralized protocol for a large set of users to privately compute averages over their joint data, which can later be used to learn more complex models. Our protocol can find a solution of arbitrary accuracy, does not rely on a trusted third party and preserves the privacy of users throughout the execution in both the honest-but-curious and malicious adversary models. Furthermore, we design a verification procedure which offers protection against malicious users joining the service with the goal of manipulating the outcome of the algorithm
EmoGame: Towards a Self-Rewarding Methodology for Capturing Children Faces in an Engaging Context
No full textInternational audienceFacial expression datasets are currently limited as most of them only capture the emotional expressions of adults. Researchers have begun to assert the importance of having child exemplars of the various emotional expressions in order to study the interpretation of these expressions developmentally. Capturing children expression is more complicated as the protocols used for eliciting and recording expressions for adults are not necessarily adequate for children. This paper describes the creation of a flexible Emotional Game for capturing children faces in an engaging context. The game is inspired by the well-known Guitar HeroTM gameplay, but instead of playing notes, the player should produce series of expressions. In the current work, we measure the capacity of the game to engage the children and we discuss the requirements in terms of expression recognition needed to ensure a viable gameplay. The preliminary experiments conducted with a group of 12 children with ages between 7 and 11 in various settings and social contexts show high levels of engagement and positive feedback
BOEUF: A Unified Framework for Modeling and Designing Digital Orchestras
No full textInternational audienceOrchestras of Digital Musical Instruments (DMIs) enable new musical collaboration possibilities, extending those of acoustic and electric orchestras. However the creation and development of these orchestras remain constrained. In fact, each new musical collaboration system or orchestra piece relies on a fixed number of musicians, a fixed set of instruments (often only one), and a fixed subset of possible modes of collaboration. In this paper, we describe a unified framework that enables the design of Digital Orchestras with potentially different DMIs and an expand-able set of collaboration modes. It relies on research done on analysis and classification of traditional and digital orchestras, on research in Collaborative Virtual Environments, and on interviews of musicians and composers. The BOEUF framework consists of a classification of modes of collaboration and a set of components for modelling digital orchestras. Integrating this framework into DMIs will enable advanced musical collaboration modes to be used in any digital orchestra, including spontaneous jam sessions
A Novel Algorithm for Adapting the Number of Particles in Particle Filtering
No full textInternational audienc
