BUILDING MANAGEMENT SYSTEM FOR IMPLEMENTING MACHINE-LEARNING ON EDGE DEVICES
Systems and methods described herein are directed to the optimization and implementation of machine-learning models on edge devices. A building management system can receive optimization criteria to optimize a first machine-learning model for a target platform. The first machine-learning model has one or more model parameters. The building management system transforms, based on the target platform and the optimization criteria, at least one datatype of the one or more model parameters of the first machine-learning model to generate a second machine-learning model. The building management system determines, using a verification dataset, an accuracy of the second machine-learning model, and retrains the second machine-learning model using a training dataset responsive to the accuracy being less than a predetermined threshold.
1 . A building management system, comprising:
one or more processors coupled to non-transitory memory, the one or more processors configured to:
receive optimization criteria to optimize a first machine-learning model for a target platform, the first machine-learning model having one or more model parameters;
transform, based on the target platform and the optimization criteria, at least one datatype of the one or more model parameters of the first machine-learning model to generate a second machine-learning model;
determine, using a verification dataset, an accuracy of the second machine-learning model; and
retrain the second machine-learning model using a training dataset responsive to the accuracy being less than a predetermined threshold.
2 . The building management system of claim 1 , wherein the one or more processors are further configured to prune at least one parameter of the second machine-learning model.
3 . The building management system of claim 2 , wherein the one or more processors are further configured to retrain the second machine-learning model responsive to pruning the at least one parameter of the second machine-learning model.
4 . The building management system of claim 2 , wherein the one or more processors are further configured to select the at least one parameter of the second machine-learning model based on a layer type of a machine-learning layer including the at least one parameter.
5 . The building management system of claim 2 , wherein the one or more processors are further configured to prune the at least one parameter of the second machine-learning model according to a transfer learning process.
6 . The building management system of claim 1 , wherein the one or more processors are further configured to modify a runtime of the second machine-learning model based on the target platform.
7 . The building management system of claim 1 , wherein the at least one datatype is a 64-bit or a 32-bit floating point type, and the one or more model parameters of the second machine-learning model are transformed to include a 32-bit, 16-bit, or 8-bit floating point datatype.
8 . The building management system of claim 1 , wherein the one or more processors are further configured to determine, using the verification dataset, a second accuracy of the second machine-learning model responsive to retraining the second machine-learning model.
9 . A method, comprising:
receiving, by one or more processors coupled to non-transitory memory, optimization criteria to optimize a first machine-learning model for a target platform, the first machine-learning model having one or more model parameters;
transforming, by the one or more processors, based on the target platform and the optimization criteria, at least one datatype of the one or more model parameters of the first machine-learning model to generate a second machine-learning model;
determining, by the one or more processors, using a verification dataset, an accuracy of the second machine-learning model; and
retraining, by the one or more processors, the second machine-learning model using a training dataset responsive to the accuracy being less than a predetermined threshold.
10 . The method of claim 9 , further comprising pruning, by the one or more processors, at least one parameter of the second machine-learning model.
11 . The method of claim 10 , further comprising retraining, by the one or more processors, the second machine-learning model responsive to pruning the at least one parameter of the second machine-learning model.
12 . The method of claim 10 , further comprising selecting, by the one or more processors, the at least one parameter of the second machine-learning model based on a layer type of a machine-learning layer including the at least one parameter.
13 . The method of claim 10 , further comprising pruning, by the one or more processors, the at least one parameter of the second machine-learning model according to a transfer learning process.
14 . The method of claim 9 , further comprising modifying, by the one or more processors, a runtime of the second machine-learning model based on the target platform.
15 . The method of claim 9 , wherein the at least one datatype is a 64-bit or a 32-bit floating point type, and the one or more model parameters of the second machine-learning model are transformed to include a 32-bit, 16-bit, or 8-bit floating point datatype.
16 . The method of claim 9 , further comprising determining, by the one or more processors, using the verification dataset, a second accuracy of the second machine-learning model responsive to retraining the second machine-learning model.
17 . A non-transitory computer-readable medium with instructions embodied thereon that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
receiving optimization criteria to optimize a first machine-learning model for a target platform, the first machine-learning model having one or more model parameters;
transforming, based on the target platform and the optimization criteria, at least one datatype of the one or more model parameters of the first machine-learning model to generate a second machine-learning model;
determining, using a verification dataset, an accuracy of the second machine-learning model; and
retraining the second machine-learning model using a training dataset responsive to the accuracy being less than a predetermined threshold.
18 . The non-transitory computer-readable medium of claim 17 , wherein the operations further comprise pruning at least one parameter of the second machine-learning model.
19 . The non-transitory computer-readable medium of claim 18 , wherein the operations further comprise retraining the second machine-learning model responsive to pruning the at least one parameter of the second machine-learning model.
20 . The non-transitory computer-readable medium of claim 18 , wherein the operations further comprise selecting the at least one parameter of the second machine-learning model based on a layer type of a machine-learning layer including the at least one parameter.