MACHINE LEARNING INFERENCE SYSTEM

1 . A machine learning inference system comprising:

a machine learning model configured to receive sample data, process the sample data, and produce an output; and

a confidence module communicatively coupled to the machine learning model and configured to perform the following steps:

receiving data pertaining to the sample data;

analyzing the data pertaining to the sample data using a mathematical operation and/or a machine learning algorithm, wherein the machine learning algorithm is not a deep machine learning algorithm;

determining a confidence score for the machine learning model based on the analysis; and

only if the confidence score is below a predetermined confidence threshold, triggering retraining of the machine learning model.

2 . The machine learning inference system of claim 1 , wherein the data pertaining to the sample data comprises one or more of: the sample data, metadata of the sample data, the output of the machine learning model from processing the sample data, adapted sample data, a Softmax score, the Softmax score calculated by applying a Softmax operator to the output of the machine learning model.

3 . The machine learning inference system of claim 2 , wherein analyzing the data pertaining to the sample data comprises creating a distribution using the Softmax score and calculating the divergence between the distribution and a uniform distribution, wherein the divergence is an unscaled version of the confidence score.

4 . The machine learning inference system of claim 1 , further comprising:

a data minder module communicatively coupled to the machine learning model and configured to perform the following steps:

receiving training data, wherein the training data is the data used to train the machine learning model;

receiving the sample data prior to the machine learning model;

comparing, using a data representation model, the sample data to the training data to determine a similarity score;

only if the similarity score is above or equal to a first predetermined similarity threshold, sending the sample data to the machine learning model for processing.

5 . The machine learning inference system of claim 4 , further comprising a data remapping module communicatively coupled to the data minder module, wherein the data minder module is further configured to:

if the similarity score is below the first predetermined similarity threshold but above or equal to a second predetermined similarity threshold, sending the sample data to the data remapping module.

6 . The machine learning inference system of claim 4 , wherein the data minder module is further configured to:

if the similarity score is below a second predetermined similarity threshold, not sending the sample data to the machine learning model for processing.

7 . The machine learning inference system of claim 1 , further comprising:

a data remapping module communicatively coupled to the machine learning model and configured to perform the following steps:

receiving training data, wherein the training data is the data used to train the machine learning model;

receiving the sample data prior to the machine learning model;

performing domain adaptation on the sample data to increase the similarity between the sample data and the training data;

sending the adapted sample data to the machine learning model for processing.

8 . The machine learning inference system of claim 7 , further comprising a data minder module, the data remapping module communicatively coupled to the data minder module, wherein the

sample data is received from the data minder module.

9 . The machine learning inference system of claim 1 , further comprising:

an adversarial defense module communicatively coupled to the machine learning model and configured to perform the following steps:

creating a reference activation signature for the machine learning model at a first point in time;

creating a sample activation signature for the machine learning model at a second point in time, subsequent to the first point in time;

comparing the sample activation signature to the reference activation signature to detect an anomaly in the sample activation signature; and

if an anomaly is detected, sending the sample data to an exception path.

10 . The machine learning inference system of claim 1 , further comprising:

an update module communicatively coupled to the machine learning model and configured to perform the following steps:

receiving first performance data relating to the performance of the machine learning model at a first point in time and storing the first performance data;

receiving second performance data relating to the performance of the machine learning model at a second point in time, subsequent to the first point in time, and storing the second performance data;

calculating a data drift for the machine learning model using the second performance data and the first performance data;

if the data drift is above a first predetermined drift threshold, triggering retraining of the machine learning model.

11 . The machine learning inference system of claim 10 , wherein the update module is further configured to predict performance data of the retained machine learning model by processing a validation dataset, and compare the predicted performance data against validation performance data of the machine learning model when processing the validation dataset.

12 . The machine learning inference system of claim 11 , wherein the update module is further configured to revert the retained machine learning model back to the machine learning module if the predicted performance data of the retained machine learning module is lower than the validation performance data of machine learning module.

13 . The machine learning inference system of claim 1 , further comprising:

a sampling module communicatively coupled to the machine learning model and configured to perform the following steps:

sampling, based on a data representation model, the machine learning model output at a first point in time; and

determining first performance data for the machine learning model.

14 . The machine learning inference system of claim 1 , further comprising:

a data interpretability module communicatively coupled to the machine learning model and configured to perform the following steps:

generating a remapped image of the sample data that identifies salient features of the sample data used by the machine learning model to arrive at the output.

15 . The machine learning inference system of claim 1 , further comprising a graphical processing unit, wherein the graphical processing unit is configured to perform processing tasks for one or more of: the machine learning model, a data minder module, a data remapping module, an adversarial defense module, and an update module.