Canary Capital LLC

Zenrobotics Oy

Harri Valpola

Tuomas Lukka

Klaus Greff

Matti Herranen

Antti Yli-Krekola

Eva Koppali

Disclosed is a neural network structure enabling efficient training of the network and a method thereto. The structure is a ladder-type structure wherein one or more lateral input(s) is/are taken to decoding functions. By minimizing one or more cost function(s) belonging to the structure the neural network structure may be trained in an efficient way.

Neural network structure and a method thereto

The invention relates to a method for generating a control system for a target system, wherein: operational data is received; a first neural model component is trained with the received operational data for generating a prediction on a state of the target system based on the received operational data; a second neural model component is trained with the operational data for generating a regularizer for use in inverting the first neural model component; and the control system is generated by inverting the first neural model component by optimization and arranging to apply the regularizer generated with the second neural model component in the optimization. The invention relates also to a system and a computer program product.

Generation of a control system for a target system

Disclosed is a computer-implemented method for training a neural network system including an original neural network and a label generator. The method is based on an idea that the neural network system is trained by a sequence of training steps where at each training step at least one of a plurality of operations is performed and each of the operations gets performed at least once during training of the neural network system. Also disclosed are a neural network system and a computer program product.

Solution for training a neural network system

Disclosed is a computer-implemented method for estimating an uncertainty of a prediction generated by a machine learning system, the method including: receiving first data; training a first machine learning model component of a machine learning system with the received first data, the first machine learning model component is trained to generate a prediction; generating an uncertainty estimate of the prediction; training a second machine learning model component of the machine learning system with second data, the second machine learning model component is trained to generate a calibrated uncertainty estimate of the prediction. Also disclosed is a corresponding system.

Solution for machine learning system

Disclosed is a computer-implemented method for segmenting input data. In the method a plurality of tags is generated; the input data is masked with the plurality of tags; a plurality of output reconstructions is generated by inputting the plurality of masked input data to one of the following: a denoising neural network, a variational autoencoder; a plurality of values representing distances of each plurality of output reconstructions to the input data are determined; a plurality of updated versions of input data is generated by applying at least one of the determined values representing distances of each plurality of output reconstructions to the input data; and updated output reconstructions are generated by inputting the plurality of updated versions of input data to one of the networks. Also disclosed is a method for training the network and a processing unit.

Segmentation of data

A method for generating an output signal of a system based on input data received by the system includes receiving training data and training a neural network for generating the output signal by optimizing a primary cost function and an auxiliary cost function and modulating the auxiliary cost function with a gradient-based attention mask during the training.

Neural network based solution

A method in which an apparatus receives first sensor data from first sensors and determines a target position from the data, the target position may be a position in space or an orientation of a gripper in a robot arm First instructions are issued to the robot arm or the gripper in order to move a gripper to the target position. Force feedback sensor data is received from force feedback sensors associated with either the robot arm or the gripper or from the first sensors. A failure in carrying out the first instructions is determined. Second sensor data is received from the at least one first sensor. Successful gripping of an object is determined from the second sensor data.

Method for the selection of physical objects in a robot system

In the method of the invention, information is selected in a processing unit. The end result to be achieved is final feature information to be used for performing a task. In the method, input information to the processing unit is treated with the intention to find useful information from the input information in view of the task. Feature information representing characteristics of the input information is formed from the input data, the values of the feature information being further processed on the basis of their utility. The method is mainly characterized in that a preliminary prediction is formed from a first context of a sot of input information by means of first parameters and in that the first primary input of the set of input data is preprocessed by forming calculation results by means of second parameters.

Growing community of inventors

Helsinki, Finland

Harri Valpola