Microsoft Technology Licensing, LLC

Cheng Zhang

Sebastian Tschiatschek

Yingzhen Li

Jose Miguel Hernandez Lobato

Reinhard Sebastian Bernhard Nowozin

Anja Thieme

Iain Edward Buchan

Richard Eric Turner

Ameera Patel

Danielle Charlotte Mary Belgrave

Konstantina Palla

Chao Ma

Pashmina Cameron

Yordan Kirilov Zaykov

Katja Hofmann

Sam Michael Devlin

Camilla Alice Longden

Kamil Andrzej Ciosek

Maximilian Igl

Wenbo Gong

José Miguel Hernández Lobato

Angus Lamb

Evgeny Sergeevich Saveliev

Jose Miguel Hernández Lobato

Anna-Lena Popkes

Hiske Catharina Overweg

A computer-implemented method of training an auxiliary machine learning model to predict a set of new parameters of a primary machine learning model, wherein the primary model is configured to transform from an observed subset of a set of real-world features to a predicted version of the set of real-world features.

Auxiliary model for predicting new model parameters

A method of operating a neural network, comprising: at each input node of an input layer, weighting a respective input element received by that node by applying a first class of probability distribution, thereby generating a respective set of output parameters describing an output probability distribution; and from each input node, outputting the respective set of output parameters to one or more nodes in a next, hidden layer of the network, thereby propagating the respective set of output parameters through the hidden layers to an output layer; the propagating comprising, at one or more nodes of at least one hidden layer, combining the sets of input parameters and weighting the combination by applying a second class of probability distribution, thereby generating a respective set of output parameters describing an output probability distribution, wherein the first class of probability distribution is more sparsity inducing than the second class of probability distribution.

Interpretable neural network

A method of training a model comprising a generative network mapping a latent vector to a feature vector, wherein weights in the generative network are modelled as probabilistic distributions. The method comprises: a) obtaining one or more observed data points, each comprising an incomplete observation of the features in the feature vector; b) training the model based on the observed data points to learn values of the weights of the generative network which map the latent vector to the feature vector; c) from amongst a plurality of potential next features to observe, searching for a target feature of the feature vector which maximizes a measure of expected reduction in uncertainty in a distribution of said weights of the generative network given the observed data points so far; and d) outputting a request to collect a target data point comprising at least the target feature.

Collecting observations for machine learning

A computer-implemented method comprising: outputting questions to a user via one or more user devices, and receiving back responses to some of the questions from the user via one or more user devices; over time, controlling the outputting of the questions so as to output the questions under circumstances of different values for each of one or more items of metadata, wherein the one or more items of metadata comprise one or more physical conditions of the user; monitoring whether or not the user responds when the question is output with the different metadata values; training the machine learning algorithm to learn a value of each of the items of metadata which optimizes a reward function, and based thereon selecting a circumstance when the user is exhibiting a particular physical condition to output subsequent questions.

Gathering data in a communication system

A method comprising: receiving observed data points each comprising a vector of feature values, wherein for each data point, the respective feature values are values of different features of a feature vector. Each observed data point represents a respective observation of a ground truth as observed in the form of the respective values of the feature vector. The method further comprises learning parameters of a machine-learning model based on the observed data points. The machine-learning model comprises one or more statistical models arranged to model a causal relationship between the feature vector and a latent vector, a classification, and a manipulation vector. The manipulation vector represents an effect of potential manipulations occurring between the ground truth and the observation thereof as observed via the feature vector. The learning comprises learning parameters of the one or more statistical models to map between the feature vector, latent vector, classification and manipulation vector.

Robustness against manipulations in machine learning

A computer-implemented method comprising: outputting questions to a user via one or more user devices, and receiving back responses to some of the questions from the user via one or more user devices; over time, controlling the outputting of the questions so as to output the questions under circumstances of different values for each of one or more items of metadata, wherein the one or more items of metadata comprise at least a time and/or a location at which a question was output to the user via the one or more user devices; monitoring whether or not the user responds when the question is output with the different metadata values; training the machine learning algorithm to learn a value of each of the items of metadata which optimizes a reward function, and based thereon selecting a time and/or location at which to output subsequent questions.

An apparatus has a memory storing a reinforcement learning policy with an optimization component and a data collection component. The apparatus has a regularization component which applies regularization selectively between the optimization component of the reinforcement learning policy and the data collection component of the reinforcement learning policy. A processor carries out a reinforcement learning process by: triggering execution of an agent according to the policy and with respect to a first task; observing values of variables comprising: an observation space of the agent, an action of the agent; and updating the policy using reinforcement learning according to the observed values and taking into account the regularization.

Generalized reinforcement learning agent

Growing community of inventors

Cambridge, United Kingdom

Cheng Zhang