Multiscale dimensional reduction of data

A method comprising:

• inputting a respective update subset, associated with a respective session, of a sequence of updates associated with a record into a first set of neural network cells;
• for each respective vector of a plurality of session vectors derived from output of the first set of neural network cells in connection with inputting the respective update subset into the first set of neural network cells, inputting (1) the respective vector and a time difference between sessions of a set of sessions into a second set of neural network cells and (2) a respective periodic function output derived from a respective periodic interval segment of a periodic time interval into a third set of neural network cells;
• inputting a first vector set and a second vector set into a fourth set of neural network cells, the first vector set being derived from output of the second set of neural network cells in connection with inputting the respective vectors and the time difference into the second set of neural network cells, the second vector set being derived from output of the third set of neural network cells in connection with inputting the respective vectors and the respective periodic function outputs into the third set of neural network cells;
• and
• pre-pushing data or computing resources based on one or more predicted user actions that correspond to a record-characterizing vector derived from output of the fourth set of neural network cells in connection with inputting the first vector set and the second vector set into the fourth set of neural network cells, wherein the record-characterizing vector represents a user record.

One or more tangible, non-transitory machine-readable media storing instructions that, when executed by one or more processors, effectuate operations comprising:

• inputting a set of update subsets into a first machine learning model;
• for each respective vector of a set of vectors derived from output of the first machine learning model in connection with inputting the set of update subsets into the first machine learning model, inputting (1) the respective vector and a time difference between sessions of a set of sessions into a second machine learning model and (2) a respective periodic function output derived from a respective periodic interval segment of a periodic time interval into a third machine learning model;
• inputting a first vector set and a second vector set into a fourth machine learning model, the first vector set being derived from output of the second machine learning model in connection with inputting the respective vectors and the time difference into the second machine learning model, the second vector set being derived from output of the third machine learning model in connection with inputting the respective vectors and the respective periodic function outputs into the third machine learning model;
• and
• pre-pushing data or computing resources based on one or more predicted user actions that correspond to a set of values derived from output of the fourth machine learning model in connection with inputting the first vector set and the second vector set into the fourth machine learning model.