Neural network systems for abstract reasoning

A computerized system for solving a cognitive task that includes learning abstract properties, the system operable to form:

• an input unit adapted to access datasets characterizing the abstract properties;
• a neural network module, which is connected to the input unit, wherein the neural network module comprises a first neural network, a structural operation unit, a second neural network, and a third neural network, wherein:
• the first neural network is configured to generate first embeddings, upon execution based on datasets as accessed by the input unit;
• the structural operation unit is configured to form pairs of the first embeddings generated by the first neural network, the pairs formed corresponding to pairs of the datasets;
• the second neural network is configured to generate second embeddings that capture relational properties of the pairs of the datasets, upon execution based on data corresponding to pairs as formed by the structural operation unit;
• and
• the third neural network is configured to produce output values, upon execution based on second embeddings as generated by the second neural network, and an output unit connected to the neural network module and configured to learn one or more abstract properties of the datasets based on output values as obtained from the third neural network, to solve the cognitive task.

A computer program product for solving a cognitive task that includes learning abstract properties, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by one or more processors, to cause to:

• access datasets characterizing the abstract properties;
• input the accessed datasets into a first neural network and execute the first neural network to generate first embeddings;
• form pairs of the first embeddings generated, the pairs formed corresponding to pairs of the datasets;
• input data corresponding to the formed pairs into a second neural network, and execute the second neural network to generate second embeddings that capture relational properties of the pairs of the datasets;
• execute a third neural network based on the second embeddings to obtain output values;
• and
• learn one or more abstract properties of the datasets based on the output values obtained, to solve the cognitive task.