Automated end-to-end machine learning model optimization
Using a training portion of a dataset, a set of component parameters comprising parameters of a component of an object detection model are trained. Using the trained set of component parameters, a set of backbone component weights comprising weights of component types in a backbone portion of the object detection model are trained. Using the trained set of component parameters, a set of backbone link weights comprising weights of links within the backbone portion are trained. Using the trained set of component parameters, a set of head component weights comprising weights of component types in a head portion of the object detection model are trained. Using the trained sets of component parameters, backbone component weights, backbone link weights, and head component weights, a trained object detection model is configured and trained to perform object detection.
1 . A computer-implemented method comprising:
dividing a dataset into a training portion and a validation portion, wherein the dataset is an initial dataset, and wherein the validation portion of the dataset is further divided into a first portion, a second portion, and a third portion;
training, using the training portion of the dataset, a set of component parameters, the set of component parameters comprising parameters of a component of an object detection model;
training, using the trained set of component parameters and the first portion of the validation portion of the dataset, a set of backbone component weights, wherein each backbone component weight in the set of backbone component weights corresponds to a possible backbone component type in a backbone portion of the object detection model;
training, using the trained set of component parameters and the second portion of the validation portion of the dataset, a set of backbone link weights, wherein each backbone link weight within the set of backbone link weights corresponds to a possible link between two backbone components within the backbone portion of in the object detection model;
training, using the trained set of component parameters and the third portion of the validation portion of the dataset, a set of head component weights, a head component weight in the set of head component weights comprising a weight of a component type in a head portion of the object detection model;
configuring, using the trained set of component parameters, the trained set of backbone component weights, the trained set of backbone link weights, and the trained set of head component weights, a trained object detection model;
causing the trained object detection model to perform object detection; and
re-dividing the initial dataset into a second training portion and a second validation portion.
2 . The computer-implemented method of claim 1 , further comprising:
initializing, to a set of pseudorandom values having a normal distribution, the set of component parameters, the set of backbone component weights, the set of backbone link weights, and the set of head component weights.
3 . The computer-implemented method of claim 1 , wherein training the set of component parameters is performed using a stochastic training descent method.
4 . The computer-implemented method of claim 1 , wherein the trained set of component parameters, the set of backbone link weights, and the set of head component weights are held constant during training of the set of backbone component weights.
5 . The computer-implemented method of claim 1 , wherein the trained set of component parameters, the set of backbone component weights, and the set of head component weights are held constant during training of the set of backbone link weights.
6 . The computer-implemented method of claim 1 , wherein the trained set of component parameters, the set of backbone component weights, and the set of backbone link weights are held constant during training of the set of head component weights.
7 . The computer-implemented method of claim 1 , further comprising:
dividing the second validation portion into a fourth portion, a fifth portion, and a sixth portion;
retraining, using the second training portion, the trained set of component parameters;
retraining, using the fourth portion, the trained set of backbone component weights;
retraining, using the fifth portion, the trained set of backbone link weights; and
retraining, using the sixth portion, the trained set of head component weights.
8 . A computer program product for automated end-to-end machine learning model optimization, the computer program product comprising:
one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions comprising:
program instructions to divide a dataset into a training portion and a validation portion, wherein the dataset is an initial dataset, and wherein the validation portion of the dataset is further divided into a first portion, a second portion, and a third portion;
program instructions to train, using the training portion of the dataset, a set of component parameters, the set of component parameters comprising parameters of a component of an object detection model;
program instructions to train, using the trained set of component parameters and the first portion of the validation portion of the dataset, a set of backbone component weights, wherein each backbone component weight in the set of backbone component weights corresponds to a possible backbone component type in a backbone portion of the object detection model;
program instructions to train, using the trained set of component parameters and the second portion of the validation portion of the dataset, a set of backbone link weights, wherein each backbone link weight within the set of backbone link weights corresponds to a possible link between two backbone components within the backbone portion of in the object detection model;
program instructions to train, using the trained set of component parameters and the third portion of the validation portion of the dataset, a set of head component weights, a head component weight in the set of head component weights comprising a weight of a component type in a head portion of the object detection model;
program instructions to configure, using the trained set of component parameters, the trained set of backbone component weights, the trained set of backbone link weights, and the trained set of head component weights, a trained object detection model;
program instructions to causing the trained object detection model to perform object detection; and
program instructions to re-divide the initial dataset into a second training portion and a second validation portion.
9 . The computer program product of claim 8 , further comprising:
program instructions to initialize, to a set of pseudorandom values having a normal distribution, the set of component parameters, the set of backbone component weights, the set of backbone link weights, and the set of head component weights.
10 . The computer program product of claim 8 , wherein training the set of component parameters is performed using a stochastic training descent method.
11 . The computer program product of claim 8 , wherein the trained set of component parameters, the set of backbone link weights, and the set of head component weights are held constant during training of the set of backbone component weights.
12 . The computer program product of claim 8 , wherein the trained set of component parameters, the set of backbone component weights, and the set of head component weights are held constant during training of the set of backbone link weights.
13 . The computer program product of claim 8 , wherein the trained set of component parameters, the set of backbone component weights, and the set of backbone link weights are held constant during training of the set of head component weights.
14 . The computer program product of claim 8 , further comprising:
program instructions to divide the dataset into a second training portion and a second validation portion;
program instructions to divide the second validation portion into a fourth portion, a fifth portion, and a sixth portion;
program instructions to retrain, using the second training portion, the trained set of component parameters;
program instructions to retrain, using the fourth portion, the trained set of backbone component weights;
program instructions to retrain, using the fifth portion, the trained set of backbone link weights; and
program instructions to retrain, using the sixth portion, the trained set of head component weights.
15 . The computer program product of claim 8 , wherein the stored program instructions are stored in the at least one of the one or more storage media of a local data processing system, and wherein the stored program instructions are transferred over a network from a remote data processing system.
16 . The computer program product of claim 8 , wherein the stored program instructions are stored in the at least one of the one or more storage media of a server data processing system, and wherein the stored program instructions are downloaded over a network to a remote data processing system for use in a computer readable storage device associated with the remote data processing system.
17 . The computer program product of claim 8 , wherein the computer program product is provided as a service in a cloud environment.
18 . A computer system comprising one or more processors, one or more computer-readable memories, and one or more computer-readable storage devices, and program instructions stored on at least one of the one or more storage devices for execution by at least one of the one or more processors via at least one of the one or more memories, the stored program instructions comprising:
program instructions to divide a dataset into a training portion and a validation portion, wherein the dataset is an initial dataset, and wherein the validation portion of the dataset is further divided into a first portion, a second portion, and a third portion;
program instructions to train, using the training portion of the dataset, a set of component parameters, the set of component parameters comprising parameters of a component of an object detection model;
program instructions to train, using the trained set of component parameters and the first portion of the validation portion of the dataset, a set of backbone component weights, wherein each backbone component weight in the set of backbone component weights corresponds to a possible backbone component type in a backbone portion of the object detection model;
program instructions to train, using the trained set of component parameters and the second portion of the validation portion of the dataset, wherein each backbone link weight within the set of backbone link weights corresponds to a possible link between two backbone components within the backbone portion of in the object detection model;
program instructions to train, using the trained set of component parameters and the third portion of the validation portion of the dataset, a set of head component weights, a head component weight in the set of head component weights comprising a weight of a component type in a head portion of the object detection model;
program instructions to configure, using the trained set of component parameters, the trained set of backbone component weights, the trained set of backbone link weights, and the trained set of head component weights, a trained object detection model;
program instructions to causing the trained object detection model to perform object detection; and
program instructions to re-divide the initial dataset into a second training portion and a second validation portion.
19 . The computer system of claim 18 , further comprising:
program instructions to initialize, to a set of pseudorandom values having a normal distribution, the set of component parameters, the set of backbone component weights, the set of backbone link weights, and the set of head component weights.
20 . The computer system of claim 19 , wherein training the set of component parameters is performed using a stochastic training descent method.