Xerox Corporation

Conduent Business Services, LLC

Palo Alto Research Center Inc.

Guillaume M Bouchard

Victor Ciriza

Jean-Marc Andreoli

Pascal Valobra

Onno R Zoeter

Luis Rafael Ulloa Paredes

Kirk V Pothos

Cedric Archambeau

Florent C Perronnin

Marc Dymetman

Christopher R Dance

Ed H Chi

Julien Perez

Matthew Owen Scrafford

Maria Antonietta Grasso

Bruno Marquié

Gregorio Convertino

Benjamin Vincent Hanrahan

Shawn Kammerdiener

Nicholas Chi-yuen Kong

Shengbo Guo

Laurent Donini

Cedric Philippe C J G Archambeau

Jean-Luc Meunier

Stefania Castellani

Frederic Roulland

Gabriela Csurka

Yves Hoppenot

Francois Ragnet

Jutta Willamowski

Stéphane Clinchant

Marco Bressan

Ray Uri Merriam

Cédric Philippe Charles Jean Ghislain Archambeau

Balaji Lakshminarayanan

Loic Lecerf

Lionel Cazenave

Loïc M Lecerf

James Brinton Henderson

Arturo Mondragon Cardenas

William Michael Darling

Mihajlo Grbovic

Dawei Yin

Cedric P C J G Archambeau

Anil Kumar Nelakanti

Francis Bach

Julien Mairal

Behrouz Behmardi

Each computer of a peer-to-peer (P2P) network performs an iterative computer-based modeling task defined by a set of training data including at least some training data that are not accessible to the other computers of the P2P network, and by a set of parameters including a shared parameter. The modeling task optimizes an objective function comparing a model parameterized by the set of parameters with the training data. Each iteration includes: performing an iterative gradient step update of parameter values stored at the computer based on the objective function; receiving parameter values of the shared parameter from other computers of the P2P network; adjusting the parameter value of the shared parameter stored at the computer by averaging the received parameter values; and sending the parameter value of the shared parameter stored at the computer to other computers of the P2P network.

Distributed and privacy-preserving prediction method

Documents of a set of documents are represented by bag-of-words (BOW) vectors. L labeled topics are provided, each labeled with a word list comprising words of a vocabulary that are representative of the labeled topic and possibly a list of relevant documents. Probabilistic classification of the documents generates for each labeled topic a document vector whose elements store scores of the documents for the labeled topic and a word vector whose elements store scores of the words of the vocabulary for the labeled topic. Non-negative matrix factorization (NMF) is performed to generate a document-topic model that clusters the documents into k topics where k&#x3e;L. NMF factors representing L topics of the k topics are initialized to the document and word vectors for the L labeled topics. In some embodiments the NMF factors representing the L topics initialized to the document and word vectors are frozen, that is, are not updated by the NMF after the initialization.

Joint approach to feature and document labeling

A method and system are disclosed for learning a demand model and simulation parameters from validation information. Validation information is received from automatic fare collection systems and trips are reconstructed from the validation information. Origins, destinations, and arrival/departure times are estimated from the reconstructed trips. A demand model is then generated from the origins, destinations, and times. Assignment model parameters are then learned from the received validation information and demand model via iterative simulations. Infrastructure changes are made to a simulated transportation network based on the assignment and demand model using the learned parameters. A simulated response of the transportation network to the infrastructure change is then output.

Learning mobility user choice and demand models from public transport fare collection data

A collective matrix is constructed, having a diagnostic sessions dimension and a diagnostic state descriptors dimension. The diagnostic state descriptors dimension includes a plurality of symptom fields, a plurality of root cause fields, and a plurality of solution fields. Collective matrix factorization of the collective matrix is performed to generate a factored collective matrix comprising a sessions factor matrix embedding diagnostic sessions and a descriptors factor matrix embedding diagnostic state descriptors. An in-progress diagnostic session is embedded in the factored collective matrix. A symptom or solution is recommended for evaluation in the in-progress diagnostic session based on the embedding. The diagnostic state descriptors dimension may further include at least one information field storing a representation (for example, a bag-of-words representation) of a semantic description of a problem being diagnosed by the in-progress diagnostic session.

Spectral diagnostic engine for customer support call center

In rejection sampling of a function or distribution p over a space X, a proposal distribution qis refined responsive to rejection of a sample x*&#x3b5;X to generate a refined proposal distribution qselected to satisfy the criteria p(x)&#x2266;q(x)&#x2266;q(x) and q(x*)&#x3c;q(x*). In a sampling mode, the sample x* is obtained by random sampling of the space X, the rejection sampling accepts or rejects x* based on comparison of a ratio p(x*)/q(x*) with a random draw, and the refined proposal distribution qis selected to minimize a norm &#x2225;q&#x2225;where &#x3b1;&#x3c;&#x221e;. In an optimization mode, the sample x* is obtained such that q*=q(x*) maximizes qover the space X, the rejection sampling accepts or rejects x* based on a difference between or ratio of q* and p(x*), and the refined proposal distribution qis selected to minimize a norm &#x2225;q&#x2225;=max{q(x)}.

Joint algorithm for sampling and optimization and natural language processing applications of same

Object clustering techniques are disclosed. A nonnegative sparse similarity matrix is constructed for a set of objects. Nonnegative factorization of the nonnegative sparse similarity matrix is performed. Objects of the set of objects are allocated to clusters based on factor matrices generated by the nonnegative factorization of the nonnegative sparse similarity matrix.

Clustering using non-negative matrix factorization on sparse graphs

A penalized loss is optimized using a corpus of language samples respective to a set of parameters of a language model. The penalized loss includes a function measuring predictive accuracy of the language model respective to the corpus of language samples and a penalty comprising a tree-structured norm. The trained language model with optimized values for the parameters generated by the optimizing is applied to predict a symbol following sequence of symbols of the language modeled by the language model. In some embodiments the penalty comprises a tree-structured l-norm, such as a tree-structured l-norm or a tree-structured l-norm. In some embodiments a tree-structured l-norm operates on a collapsed suffix trie in which any series of suffixes of increasing lengths which are always observed in the same context are collapsed into a single node. The optimizing may be performed using a proximal step algorithm.

Language model with structured penalty

In multi-view learning, optimized prediction matrices are determined for V&#x2267;2 views of n objects, and a prediction of a view of an object is generated based on the optimized prediction matrix for that view. An objectiveis optimized, whereinis a set of parameters including at least the V prediction matrices and a concatenated matrix comprising a concatenation of the prediction matrices, andcomprises a sum including at least a loss function for each view, a trace norm of the prediction matrix for each view, and a trace norm of the concatenated matrix.may further include a sparse matrix for each view, withfurther including an element-wisenorm of the sparse matrix for each view.may further include regularization parameters scaling the trace norms of the prediction matrices and the trace norm of the concatenated matrix.

Overlapping trace norms for multi-view learning

In a parallel computing method performed by a parallel computing system comprising a plurality of central processing units (CPUs), a main process executes. Tasks are executed in parallel with the main process on CPUs not used in executing the main process. Results of completed tasks are stored in a cache, from which the main process retrieves completed task results when needed. The initiation of task execution is controlled by a priority ranking of tasks based on at least probabilities that task results will be needed by the main process and time limits for executing the tasks. The priority ranking of tasks is from the vantage point of a current execution point in the main process and is updated as the main process executes. An executing task may be pre-empted by a task having higher priority if no idle CPU is available.

Transportation network micro-simulation with pre-emptive decomposition

Methods and systems are presented for performing one or more printer device management functions in a network, in which affinities between printers are determined from job tracking data to indicate associations between printer devices and user devices, and the affinity data is used to perform one or more printer management functions such as determining printer connections for new or roaming user devices, print job redirection, and identification of underutilized printer device assets.

Methods and apparatus for improved operation of networked printing system

A system and method for solving an optimization problem are disclosed which are based on adaptive weighted stochastic gradient descent. The method includes receiving data to be sampled in optimizing an integral of a second function and, for a number of iterations, sampling the data according to a modified distribution function which depends on at least one parameter, updating at least one parameter of the second function, and updating at least one parameter of the modified distribution function. In a subsequent one of the iterations, the modified distribution function used in sampling the data is based on the updated at least one parameter of the distribution function. Information is output, which can be based on the integral function with the at least one updated parameter.

Adaptive weighted stochastic gradient descent

A method for performing data processing through a pipeline of components includes receiving a set of training observations, each including partial user feedback relating to error in data output by the pipeline for respective input data. Some pipeline components commit errors for at least some of the input data, contributing to an error in the respective output data. A prediction model models a probability of a pipeline component committing an error, given input data. Model parameters are learned using the training observations. For a new observation which includes input data and, optionally, partial user feedback indicating that an error has occurred in processing the new input data, without specifying which pipeline component(s) contributed to the observed error in the output data, a prediction is made as to which of the pipeline components contributed to the error in the output (if any).

Error prediction with partial feedback

A system and method for generating an occupancy model are disclosed. The model is learned using occupancy data for zones, each zone including cells, which are occupied or not at a given time, each with a sensor, which may be reporting or not. The data provides an observed occupancy corresponding to a number of cells in the respective zone which have reporting sensors, and the number of those sensors which are reporting that the respective cell is occupied. The occupancy model is based on a demand model and a sensor noise model which accounts for behavior of the non-reporting sensors. The noise model assumes that the probability of a sensor being in the reporting state is dependent on whether the respective cell is occupied or not. The model can fit the occupancy data better than one which assumes that non-reporting cells are occupied with the same frequency as reporting ones.

System and method for generating an occupancy model

A method operates on observed relationship data between pairs of entities of a set of entities including entities of at least two (and optionally at least three) different entity types. An observed collective symmetric matrix is constructed in which element (n,m)=element (m,n) stores the observed relationship between entities indexed n and m when the observed relationship data includes this observed relationship. A prediction collective symmetric matrix is optimized in order to minimize a loss function comparing the observed collective symmetric matrix and the prediction collective symmetric matrix. A relationship between two entities of the set of entities is predicted using the optimized prediction collective symmetric matrix. Entities of the same entity type may be indexed using a contiguous set of indices such that the entity type maps to a contiguous set of rows and corresponding contiguous set of columns in the observed collective symmetric matrix.

Convex collective matrix factorization

A control system reduces energy consumption in a multi-device system comprising a plurality of devices. The control system includes at least one processor. The processor is programmed to receive a job to be executed, as well as a selection of one of the plurality of devices for executing the job and a transfer cost for transferring the job from the selected device to each of the plurality of devices. A device to execute the job is determined through optimization of a first cost function. The first cost function is based on the device selection and the transfer costs. The job is assigned to the determined device and a time-out for each device in the multi-device system is determined through optimization of a second cost function. The second cost function is based on an expected energy consumption by the multi-device system. The devices are provided with the determined time-outs.

Multi-device powersaving

Growing community of inventors

Saint-Martin-le-Vinoux, France

Guillaume M Bouchard