Automatic profile extraction in data streams using recurrent neural networks
In various embodiments, a process for automatic profile extraction in data streams using recurrent neural networks includes receiving input sequence data associated with a stream of events and using a plurality of trained recurrent neural network machine learning models at least in part in parallel to determine different embedding output sets that represent at least a portion of the input sequence data in a plurality of different embedding spaces. The process includes providing the different embedding output sets to one or more classifier machine learning models to determine one or more classifier results, and using the one or more classifier results to provide a prediction output.
1 . A method comprising:
receiving input sequence data associated with a stream of events;
using a plurality of trained recurrent neural network machine learning models at least in part in parallel to determine different embedding output sets that represent at least a portion of the input sequence data in a plurality of different embedding spaces;
providing the different embedding output sets to one or more classifier machine learning models to determine one or more classifier results, wherein:
the one or more classifier machine learning models includes a respective classifier for a respective one of the trained recurrent neural network machine learning models;
each of the one or more classifier machine learning models is configured to determine a score; and
using the one or more classifier results to provide a prediction output including by combining the score determined by each of the one or more classifier machine learning models.
2 . The method of claim 1 , wherein at least one of the different embedding output sets represents at least a portion of the input sequence data in an embedding space of lower dimension than an original input space of the input sequence data.
3 . The method of claim 1 , wherein:
a first one of the plurality of trained recurrent neural network machine learning models is configured for a first entity;
a second one of the plurality of trained recurrent neural network machine learning models is configured for a second entity; and
using the plurality of trained recurrent neural network machine learning models at least in part in parallel includes:
determining, by the first one of the plurality of trained recurrent neural network machine learning models, a first embedding output set included in the different embedding output sets based on the received input sequence data; and
determining, by the second one of the plurality of trained recurrent neural network machine learning models, a second embedding output set included in the different embedding output sets based on the received input sequence data.
4 . The method of claim 1 , wherein:
the one or more classifier machine learning models are configured to extract relevant information to determine the one or more classifier results; and
the one or more classifier results includes a score.
5 . The method of claim 1 , wherein using the one or more classifier results to provide the prediction output includes using raw data in combination with the combined score.
6 . The method of claim 1 , wherein using the one or more classifier results to provide the prediction output includes weighting a first entity more heavily than a second entity based at least in part on the first entity being expected to be more indicative of a characteristic than the second entity.
7 . The method of claim 1 , wherein the stream of events includes transactions.
8 . The method of claim 7 , wherein the one or more classifier results includes a score associated with a likelihood of an associated transaction being fraudulent.
9 . The method of claim 7 , wherein the prediction output includes a value indicating an associated transaction as being fraudulent.
10 . The method of claim 1 , wherein each of the trained recurrent neural network machine learning models is configured to learn context for a respective entity associated with an event in the stream of events.
11 . The method of claim 1 , wherein a first subset of the trained recurrent neural network machine learning models is trained in parallel with a second subset of the trained recurrent neural network machine learning models.
12 . The method of claim 1 , wherein at least one of the plurality of recurrent neural network machine learning models is trained using supervised training.
13 . The method of claim 1 , wherein at least one of the plurality of recurrent neural network machine learning models is trained using unsupervised training.
14 . The method of claim 1 , further comprising, prior to using the plurality of trained recurrent neural network machine learning models at least in part in parallel to determine the different embedding output sets that represent at least a portion of the input sequence data in the plurality of different embedding spaces:
pre-processing the input sequence data including by determining at least one feature from the input sequence data.
15 . The method of claim 1 , further comprising performing a database operation in parallel for a plurality of states associated with the plurality of trained recurrent neural network machine learning models.
16 . A method comprising:
receiving input sequence data associated with a stream of events;
using a plurality of trained recurrent neural network machine learning models at least in part in parallel to determine different embedding output sets that represent at least a portion of the input sequence data in a plurality of different embedding spaces including by, for each of the plurality of trained recurrent neural network machine learning models:
retrieving a first state for a first transaction in the stream of events, wherein the first state is:
a default state in the event the first transaction is associated with an entity that has not been previously seen, or
a saved recurrent neural network state for an entity associated with the first transaction in the event the first transaction is associated with an entity that has been previously seen;
providing the retrieved first state and an input data based on the first transaction to a respective recurrent neural network machine learning model to determine a new second state; and
updating the saved recurrent neural network state for the entity associated with the first transaction to be the new second state;
providing the different embedding output sets to one or more classifier machine learning models to determine one or more classifier results; and
using the one or more classifier results to provide a prediction output, wherein the prediction output is determined using at least (i) the first state and (ii) the input data based on the first transaction.
17 . A system, comprising:
a processor configured to:
receive input sequence data associated with a stream of events;
use a plurality of trained recurrent neural network machine learning models at least in part in parallel to determine different embedding output sets that represent at least a portion of the input sequence data in a plurality of different embedding spaces;
provide the different embedding output sets to one or more classifier machine learning models to determine one or more classifier results, wherein:
the one or more classifier machine learning models includes a respective classifier for a respective one of the trained recurrent neural network machine learning models;
each of the one or more classifier machine learning models is configured to determine a score; and
use the one or more classifier results to provide a prediction output including by combining the score determined by each of the one or more classifier machine learning models; and
a memory coupled to the processor and configured to provide the processor with instructions.
18 . A computer program product embodied in a non-transitory computer readable medium and comprising computer instructions for:
receiving input sequence data associated with a stream of events;
using a plurality of trained recurrent neural network machine learning models at least in part in parallel to determine different embedding output sets that represent at least a portion of the input sequence data in a plurality of different embedding spaces;
providing the different embedding output sets to one or more classifier machine learning models to determine one or more classifier results, wherein:
the one or more classifier machine learning models includes a respective classifier for a respective one of the trained recurrent neural network machine learning models;
each of the one or more classifier machine learning models is configured to determine a score; and
using the one or more classifier results to provide a prediction output including by combining the score determined by each of the one or more classifier machine learning models.