Method for synthetic data generation
View Patent ↗A method for generating a synthetic dataset from an original dataset for creating and training machine learning models. The method includes a graphical user interface with inputs for setting values to be used in structuring Maclaurin series at datapoints and modifying results to generate new datapoints for the synthetic dataset.
1 . A computer-implemented method executed by software on a processor of a computerized device for increasing output efficiency of machine learning models by generating specific synthetic datasets enabled to at least train a machine learning model, comprising steps of:
selecting an original graph of data points as an original dataset;
constructing a first Maclaurin series at a first data point in the original dataset;
multiplying an input, X, of the first Maclaurin series by a constant, thereby modifying the first Maclaurin series;
applying a recursive derivative process, deriving general expressions for an nth derivative of the original graph dataset at X=0, that is f n (0) thereby obtaining derivative-based coefficient values corresponding to the original dataset at the first data point;
incorporating the obtained coefficient values into the first modified Maclaurin series, generating a second Maclaurin series;
evaluating the second Maclaurin series at a second datapoint in the original dataset thereby generating a first datapoint in the synthetic dataset;
shifting to each successive data point in the original dataset and repeating the constructing, modifying, incorporating, and evaluating steps to generate additional synthetic data points;
repeating the steps in order until each datapoint in the original dataset has been used to generate a datapoint in the synthetic dataset thereby increasing an amount of synthetic data and increasing a variation level of the synthetic data; and
implementing the synthetic dataset to train the machine learning model thus improving efficiency and accuracy of outputs of the machine learning model while requiring less computational bandwidth commonly used in generative neural network techniques for synthetic data generation and use.
2 . The method of claim 1 wherein the constant used for multiplying is selected as either a positive decimal or a positive integer value.
3 . The method of claim 1 wherein an interactive graphic user interface (GUI) is provided on a display screen of a computerized appliance, the GUI having input mechanisms enabling a user to upload the original dataset, to enter values for variables in the first Maclaurin series and parameters required to implement the steps to generate the synthetic data sets from the original dataset, comprising;
(a) uploading the original dataset,
(b) entering the variable values and parameters, and
(c) selecting a data input, upon which the computerized appliance performs the steps of the method and generates a number of synthetic datasets according to a number of cycles input, and a new dataset consisting of both the original dataset and the synthetic datasets are exported to a spreadsheet file and saved prior to implementation.
4 . The method of claim 3 wherein the GUI further comprises an input to open a graph window on the display screen of the computerized appliance and to display the original and generated synthetic datasets demonstrating a diversity level between the synthetic datasets and the original dataset.
5 . The method of claim 3 wherein the GUI has a “reset” input, that, when selected, clears all parameters.
6 . A system for generating controlled synthetic datasets enabling at least creating and training of machine learning models, comprising:
a computerized appliance having a processor and a display screen;
software executing on a processor of the computerized appliance; and
a graphical user interface (GUI) displayed on the display screen of the computerized appliance by the software executing on the computerized appliance, the GUI having input mechanisms enabling a user to upload an original dataset, to enter values for variables in a Maclaurin series and parameters required to implement steps to generate synthetic data sets from the original dataset, the steps comprising (a) uploading the original dataset, (b) entering the variable values and parameters, and (c) selecting a generate data input, upon which the computerized appliance performs the steps of the method and generates a number of synthetic datasets according to a number of cycles input, and a new dataset consisting of both the original dataset and the new dataset is displayed in a display of the GUI and is exported to spreadsheet file and saved.
7 . The system of claim 6 wherein, upon a user selecting the generate data input the system constructs a first Maclaurin series at a first data point in the original dataset of data points, multiplies the first Maclaurin series by a constant, applies a recursive derivative process deriving general expressions for an nth derivative of the original dataset at X=0, incorporates derived values of the nth derivative f n (0) into the first Maclaurin series, generating a new Maclaurin series, evaluates the new Maclaurin series at a second datapoint in the original graph as a first datapoint in a synthetic dataset, shifts to a second data point in the first Macaurin series, repeats the previous steps to generate a second datapoint in the synthetic dataset; and repeats the process until a last datapoint in the original dataset has been used to generate a final datapoint in the synthetic dataset.
8 . The system of claim 7 wherein the GUI has an input for setting a value for the constant used for multiplying, an input for setting a value for a distance along an x axis of the original graph for which the Maclaurin series will be evaluated, an input for setting a value for a number of iterations for the recursive derivative process, and an input for setting a value for a learning rate as a change in each iteration.
9 . The system of claim 8 wherein the GUI further has an input for creating synthetic data that is diverse a value that may be used to lower the learning rate exponentially, an input for a value that will be a number of nth derivative terms which will be calculated per data point, an input for a value that will be a number of times the overall process will be run, and an input for a value to for a variable in the dataset that will be selected for graphing after the synthetic dataset has been generated.
10 . The system of claim 8 wherein the input for the distance along the x axis for which the first Maclaurin series will be evaluated, and the input for the constant used for multiplying are each a slider on a track that a user may move to increase and decrease values including at least the constant used for multiplying.
11 . The system of claim 10 wherein a value is displayed in the GUI for the distance along the x axis and for the constant used for multiplying proximate the sliders.
12 . The system of claim 9 wherein the input for the value used to lower the learning rate exponentially and the input for the value that will be the number of nth derivative terms which will be calculated per data point are each a slider on a track that a user may move to increase and decrease the values for variables, and a value selected by the slider is displayed proximate the slider.