Caching of machine learning model training parameters

1. A method, comprising:

training, by a first training process, a machine learning model using a given training dataset comprising a plurality of input training data elements and respective ones of a plurality of output labels, wherein the training learns one or more parameters of the machine learning model, using the given training dataset to learn to recognize a given output label of the plurality of output labels for a given input training data element, wherein the one or more learned parameters of the machine learning model comprise one or more weights for respective connections between a plurality of layers of the machine learning model, and wherein the trained machine learning model generates one or more of: (i) at least one prediction and (ii) at least one classification, wherein the one or more learned parameters of the machine learning model are distinct from the plurality of input training data elements and from the plurality of output labels; and

caching, in at least one cache memory, at least one of the one or more learned parameters of the machine learning model from the training with the given training dataset, wherein the cached at least one learned parameter of the machine learning model is reused for a subsequent training, by a subsequent training process, wherein the subsequent training process is distinct from the first training process, wherein the cached at least one learned parameter of the machine learning model is provided, in response to a request by the subsequent training process, to the subsequent training process by a processor-based cache manager that manages the at least one cache memory for a plurality of training processes;

wherein the method is performed by at least one processing device comprising a processor coupled to a memory.

2. The method of claim 1 , wherein the caching is performed after each of a plurality of iterations of the training of the machine learning model.

3. The method of claim 2 , wherein a given cached iteration of the training of the machine learning model is identified using a key based at least in part on one or more of: (i) a hash of the given training dataset, (ii) a hash of the at least one learned parameter of the machine learning model following the given cached iteration, and (iii) one or more hyperparameters of the machine learning model following the given cached iteration.

4. The method of claim 3 , wherein the key for the given cached iteration of the training of the machine learning model is evaluated to determine if the given cached iteration is in the at least one cache memory.

5. The method of claim 2 , wherein a given cached iteration of the training of the machine learning model comprises the trained machine learning model following the given cached iteration, checkpoints of the given cached iteration, and a response time of the given cached iteration.

6. The method of claim 2 , wherein the caching of a given iteration of the training of the machine learning model occurs when the given cached iteration is not found in the at least one cache memory.

7. The method of claim 6 , wherein the at least one cache memory is accessible by one or more of: (a) one or more physical processing devices, and (b) one or more virtual processing devices that implement the training of the machine learning model.

8. An apparatus comprising:

at least one processing device comprising a processor coupled to a memory;

the at least one processing device being configured to implement the following steps:

training, by a first training process, a machine learning model using a given training dataset comprising a plurality of input training data elements and respective ones of a plurality of output labels, wherein the training learns one or more parameters of the machine learning model, using the given training dataset to learn to recognize a given output label of the plurality of output labels for a given input training data element, wherein the one or more learned parameters of the machine learning model comprise one or more weights for respective connections between a plurality of layers of the machine learning model, and wherein the trained machine learning model generates one or more of: (i) at least one prediction and (ii) at least one classification, wherein the one or more learned parameters of the machine learning model are distinct from the plurality of input training data elements and from the plurality of output labels; and

caching, in at least one cache memory, at least one of the one or more learned parameters of the machine learning model from the training with the given training dataset, wherein the cached at least one learned parameter of the machine learning model is reused for a subsequent training, by a subsequent training process, wherein the subsequent training process is distinct from the first training process, wherein the cached at least one learned parameter of the machine learning model is provided, in response to a request by the subsequent training process, to the subsequent training process by a processor-based cache manager that manages the at least one cache memory for a plurality of training processes.

9. The apparatus of claim 8 , wherein the caching is performed after each of a plurality of iterations of the training of the machine learning model.

10. The apparatus of claim 9 , wherein a given cached iteration of the training of the machine learning model is identified using a key based at least in part on one or more of: (i) a hash of the given training dataset, (ii) a hash of the at least one learned parameter of the machine learning model following the given cached iteration, and (iii) one or more hyperparameters of the machine learning model following the given cached iteration.

11. The apparatus of claim 10 , wherein the key for the given cached iteration of the training of the machine learning model is evaluated to determine if the given cached iteration is in the at least one cache memory.

12. The apparatus of claim 9 , wherein a given cached iteration of the training of the machine learning model comprises the trained machine learning model following the given cached iteration, checkpoints of the given cached iteration, and a response time of the given cached iteration.

13. The apparatus of claim 9 , wherein the caching of a given iteration of the training of the machine learning model occurs when the given cached iteration is not found in the at least one cache memory.

14. The apparatus of claim 13 , wherein the at least one cache memory is accessible by one or more of: (a) one or more physical processing devices, and (b) one or more virtual processing devices that implement the training of the machine learning model.

15. A non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device to perform the following steps:

training, by a first training process, a machine learning model using a given training dataset comprising a plurality of input training data elements and respective ones of a plurality of output labels, wherein the training learns one or more parameters of the machine learning model, using the given training dataset to learn to recognize a given output label of the plurality of output labels for a given input training data element, wherein the one or more learned parameters of the machine learning model comprise one or more weights for respective connections between a plurality of layers of the machine learning model, and wherein the trained machine learning model generates one or more of: (i) at least one prediction and (ii) at least one classification, wherein the one or more learned parameters of the machine learning model are distinct from the plurality of input training data elements and from the plurality of output labels; and

caching, in at least one cache memory, at least one of the one or more learned parameters of the machine learning model from the training with the given training dataset, wherein the cached at least one learned parameter of the machine learning model is reused for a subsequent training, by a subsequent training process, wherein the subsequent training process is distinct from the first training process, wherein the cached at least one learned parameter of the machine learning model is provided, in response to a request by the subsequent training process, to the subsequent training process by a processor-based cache manager that manages the at least one cache memory for a plurality of training processes.

16. The non-transitory processor-readable storage medium of claim 15 , wherein the caching is performed after each of a plurality of iterations of the training of the machine learning model.

17. The non-transitory processor-readable storage medium of claim 16 , wherein a given cached iteration of the training of the machine learning model is identified using a key based at least in part on one or more of: (i) a hash of the given training dataset, (ii) a hash of the at least one learned parameter of the machine learning model following the given cached iteration, and (iii) one or more hyperparameters of the machine learning model following the given cached iteration.

18. The non-transitory processor-readable storage medium of claim 17 , wherein the key for the given cached iteration of the training of the machine learning model is evaluated to determine if the given cached iteration is in the at least one cache memory.

19. The non-transitory processor-readable storage medium of claim 16 ,

wherein a given cached iteration of the training of the machine learning model comprises the trained machine learning model following the given cached iteration, checkpoints of the given cached iteration, and a response time of the given cached iteration.

20. The non-transitory processor-readable storage medium of claim 16 , wherein the caching of a given iteration of the training of the machine learning model occurs when the given cached iteration is not found in the at least one cache memory.