Pre-training of an encoder with unlabelled data by masked property prediction

A method for training an encoder that is configured for encoding a record of measurement data into a representation in a working space, the measurement data including measurement values that are associated with positions in space, the method comprising the following steps:

• providing at least one training record of measurement data;
• designating cells in space, wherein at least some of the cells contain positions in space to which measurement values in the at least one training record of measurement data are associated;
• determining, for at least one group of the cells, an indication of a predetermined property;
• creating a masked record of measurement data by removing, from the at least one training record of measurement data, all measurement values that are associated with positions in space that lie in at least one to-be-masked cell;
• providing the masked record of measurement data to the encoder, thereby obtaining a representation;
• decoding, by a decoder, the representation into a reconstructed indication of the predetermined property for at least one group of cells;
• rating, by means of a predetermined loss function, how well the reconstructed indication of the predetermined property corresponds to the indication of the predetermined property determined from the at least one training record of measurement data;
• and
• optimizing parameters that characterize a behavior of the encoder towards a goal of improving the rating by the loss function upon further processing of training records.

A non-transitory computer-readable data carrier on which is stored a computer program including machine-readable instructions for training an encoder that is configured for encoding a record of measurement data into a representation in a working space, the measurement data including measurement values that are associated with positions in space, the instructions, when executed by one or more computers and/or compute instances, causing the one or more computers and/or compute instances to perform the following steps comprising:

• providing at least one training record of measurement data;
• designating cells in space, wherein at least some of the cells contain positions in space with which measurement values in the at least one training record of measurement data are associated;
• determining, for each of the cells, based on the at least one training record of measurement data, an indication of a predetermined property;
• creating a masked record of measurement data by removing, from the at least one training record of measurement data, all measurement values that are associated with positions in space that lie in at least one to-be-masked cell;
• providing the masked record of measurement data to the encoder, thereby obtaining a representation;
• decoding, by a decoder, the representation into a reconstructed indication of the predetermined property for at most neighboring cells of not-masked cells;
• rating, using a predetermined loss function, how well the reconstructed indication of the predetermined property corresponds to the indication of the predetermined property determined from the at least one training record of measurement data;
• and
• optimizing parameters that characterize a behavior of the encoder towards the goal of improving the rating by the loss function upon further processing of training records.

One or more computers and/or compute instances including a non-transitory computer-readable data carrier on which is stored a computer program including machine-readable instructions for training an encoder that is configured for encoding a record of measurement data into a representation in a working space, the measurement data including measurement values that are associated with positions in space, the instructions, when executed by the one or more computers and/or compute instances, causing the one or more computers and/or compute instances to perform the following steps comprising:

• providing at least one training record of measurement data;
• designating cells in space, wherein at least some of the cells contain positions in space with which measurement values in the at least one training record of measurement data are associated;
• determining, for each of the cells, based on the at least one training record of measurement data, an indication of a predetermined property;
• creating a masked record of measurement data by removing, from the at least one training record of measurement data, all measurement values that are associated with positions in space that lie in at least one to-be-masked cell;
• providing the masked record of measurement data to the encoder, thereby obtaining a representation;
• decoding, by a decoder, the representation into a reconstructed indication of the predetermined property for at most neighboring cells of not-masked cells;
• rating, using a predetermined loss function, how well the reconstructed indication of the predetermined property corresponds to the indication of the predetermined property determined from the at least one training record of measurement data;
• and
• optimizing parameters that characterize a behavior of the encoder towards the goal of improving the rating by the loss function upon further processing of training records.