Method for learning from data with label noise
AssigneeSiemens Healthcare GmbH
InventorsMartin Kraus, Florin-Cristian Ghesu
Life of this patent
Estimating the term…
Abstract
A method and a system are for performing at least one inference task on medical input data. In an embodiment, the system includes: a computing device configured to implement an artificial neural network, ANN. The ANN is configured to receive medical input data as an input and to perform at least one inference task based on the medical input data. The ANN includes an output layer with, for at least one inference task of the at least one inference task, a plurality of output nodes. Each output node corresponds to one quantile out of a plurality of defined quantiles.
Claims24
A system for performing at least one inference task on medical input data, comprising:
- a computing device, configured to implement an artificial neural network (ANN), the ANN being configured to
- receive medical input data as an input, and
- perform the at least one inference task based on the medical input data received;
- and
- wherein the ANN comprises an output layer including, for at least one inference task of the at least one inference task, a plurality of output nodes, and
- wherein each output node of the plurality of output nodes, corresponds to one quantile out of a plurality of defined quantiles.
A computer-implemented method for performing at least one inference task on medical input data, comprising:
- inputting medical input data into an artificial neural network (ANN), the ANN being configured to
- receive medical input data as an input, and
- perform at least one inference task based on the medical input data received;
- processing the medical input data by the ANN;
- and
- outputting by the ANN, for each at least one inference task of the at least one inference task, an output value respectively corresponding to each quantile of a plurality of defined quantiles.
A computer-implemented method for training an artificial neural network (ANN) for at least one inference task on medical input data, comprising:
- providing an ANN, configured to
- receive medical input data as an input and
- perform at least one inference task based on the medical input data;
- wherein the ANN includes an output layer with, for at least one inference task of the at least one inference task, a plurality of output nodes, and wherein each output node of the plurality of output nodes corresponds to one quantile out of a plurality of defined quantiles;
- and
- training the ANN with labelled training data and using a cost function, including an individual loss term, for each of the defined quantiles.
Dependencies are inferred from the claim text.