Abstract
In machine learning applications, a system takes a data set and runs it through the machine learning
model to create a prediction. It is necessary for some systems to do this on a continual basis for
example, a computer vision system continually evaluates the environment to recognize objects. In other
systems it is not necessary to run data through the model every time new data is available and may not
result in a significantly different prediction if it was. In these systems, it would be more cost‐effective to
space out runs of the model. Since running a machine learning model can be expensive, having a
method to reduce the frequency of runs can provide significant savings to a business. In a continuously
monitored system, every time data arrives a new prediction is done and in many cases the prediction is
not significantly different than the previous prediction. This consumes resources which affects the
responsiveness of the system and increases operation cost. By only doing predictions when it’ likely the
new prediction will be significantly different, the system improves responsiveness and reduces the cost
of operation. In a batch system, predictions are done on a fixed schedule. This method would change
the system such that it would only add a record to the batch when it’ likely the new prediction will be
significantly different. This also would improve system responsiveness and reduces cost of operation.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.
Recommended Citation
INC, HP, "MULTI-FACTOR ADAPTIVE MACHINE LEARNING EVALUATION IN PRODUCTION ENVIRONMENTS", Technical Disclosure Commons, (November 21, 2018)
https://www.tdcommons.org/dpubs_series/1694