Multi-domain feature enhancement for transfer learning (FTL)
Methods and systems are presented for configuring and training a machine learning model using transfer learning techniques that can transfer knowledge among multiple domains that do not share an identical feature set. Instead of using any feature set associated with a domain, a feature arrangement that combines all of the feature sets associated with the multiple domains in a particular organization for configuring and training the machine learning model. The feature arrangement includes a domain independent section and multiple domain-specific sections corresponding to the multiple domains. The domain independent section includes common features that are common across the multiple domains. Each of the domain-specific sections includes a feature set associated with the corresponding domain. The machine learning model that is configured in this manner can be trained to learn knowledge across the multiple domains and subsequently perform tasks for the multiple domains.
1 . A system, comprising:
a non-transitory memory; and
one or more hardware processors coupled with the non-transitory memory and configured to read instructions from the non-transitory memory to cause the system to perform operations comprising:
obtaining a first plurality of data records comprising first data associated with a first domain and corresponding to a first set of features;
determining a set of common features between the first set of features and a second set of features associated with a second domain, wherein the set of common features is a first subset of the first set of features, and wherein each feature in a second subset of the first set of features is not common with the second set of features;
configuring a machine learning model to perform tasks associated with the first domain and the second domain, wherein the configuring the machine learning model comprises configuring an internal structure of the machine learning model to accept and process input data corresponding to a set of input features arranged in a feature arrangement, wherein the internal structure of the machine learning model comprises (i) a first set of input nodes corresponding to a domain-independent section of the feature arrangement that includes the set of common features and having a first set of connections connecting the first set of input nodes to a first set of internal nodes of the machine learning model and (ii) a second set of input nodes corresponding to a first domain-specific section of the feature arrangement that includes the first set of features comprising both of the first subset of the first set of features and the second subset of the first set of features, wherein the second set of input nodes has a second set of connections connecting the second set of input nodes to a second set of internal nodes of the machine learning model;
generating training data according to the feature arrangement based on the first data;
training the machine learning model using the training data; and
subsequent to the training the machine learning model, performing a task associated with the second domain, using the trained machine learning model, based on second data that is organized according to the feature arrangement.
2 . The system of claim 1 , wherein the operations further comprise:
receiving a transaction request associated with the first domain;
determining, for the transaction request, transaction data corresponding to the first set of features;
generating, using at least the transaction data, inputs for the machine learning model based on the feature arrangement;
providing the inputs to the machine learning model; and
obtaining an outcome from the machine learning model.
3 . The system of claim 1 , wherein the internal structure of the machine learning model further comprises: a third set of input nodes corresponding to a second domain-specific section of the feature arrangement that includes the second set of features.
4 . The system of claim 2 , wherein the generating the inputs comprises:
generating a data structure according to the feature arrangement;
determining a subset of the transaction data corresponding to the set of common features;
inserting the subset of the transaction data in a first set of data fields corresponding to the domain-independent section of the data structure; and
inserting the transaction data in a second set of data fields corresponding to the first domain-specific section of the data structure, wherein the inputs provided to the machine learning model comprise the data structure.
5 . The system of claim 4 , wherein the internal structure of the machine learning model further comprises a third set of input nodes corresponding to a second domain-specific section of the feature arrangement that includes the second set of features, and wherein the generating the inputs further comprises:
inserting default values in a third set of data fields corresponding to the second domain-specific section of the data structure.
6 . The system of claim 1 , wherein the operations further comprise:
obtaining a second plurality of data records comprising second data associated with the second domain and corresponding to the second set of features, wherein the generating the training data is further based on the second data.
7 . The system of claim 1 , wherein the first and second domains represent different services, wherein the first plurality of data records represents first transactions conducted through a first one of the different services.
8 . A method, comprising:
receiving a transaction request associated with a first domain;
obtaining transaction data associated with the transaction request and corresponding to a first set of features to be processed by a machine learning model configured to perform tasks associated with the first domain and a second domain, wherein the machine learning model comprises an internal structure that processes data corresponding to a set of input features arranged in the feature arrangement, wherein the internal structure of the machine learning model comprises (i) a first set of input nodes corresponding to a domain-independent section of the feature arrangement that includes a set of common features that is common between the first set of features and a second set of features associated with the second domain, (ii) a second set of input nodes corresponding to a first domain-specific section of the feature arrangement that includes the first set of features, (iii) a first set of connections that connect the first set of input nodes to a first set of internal nodes of the machine learning model, and (iv) a second set of connections that connect the second set of input nodes to a second set of internal nodes of the machine learning model;
generating a data structure according to the feature arrangement for the machine learning model, wherein the data structure comprises (i) a first portion corresponding to the domain-independent section of the feature arrangement and (ii) a second portion corresponding to the first domain-specific section of the feature arrangement;
extracting, from the transaction data, a subset of the transaction data corresponding to the set of common features;
inserting the subset of the transaction data in the first portion of the data structure;
inserting the transaction data in the second portion of the data structure;
providing the data structure as input data to the machine learning model; and
processing the transaction request based on an output from the machine learning model.
9 . The method of claim 8 , wherein the internal structure of the machine learning model further comprises a third set of input nodes corresponding to a second domain-specific section of the feature arrangement that includes the second set of features, and wherein the method further comprises:
inserting default values in a third portion of the data structure corresponding to the second domain-specific section of the feature arrangement.
10 . The method of claim 9 , wherein the default values comprise null values.
11 . The method of claim 8 , further comprising:
obtaining a first plurality of records comprising first data associated with the first domain and corresponding to the first set of features;
generating a training data record according to the feature arrangement based on a first record in the first plurality of records, wherein the generating the training data record comprises:
extracting a portion of the first record corresponding to the set of common features;
inserting the portion of the first record in the domain-independent section of the training data record; and
inserting the first record in the first domain-specific section of the training data record; and
training the machine learning model using the training data record.
12 . The method of claim 11 , wherein the internal structure of the machine learning model further comprises a third set of input nodes corresponding to a second domain-specific section of the feature arrangement that includes the second set of features, and wherein the generating the training data record further comprises inserting default values in the second domain-specific section of the training data record.
13 . The method of claim 8 , further comprising:
receiving a second transaction request associated with the second domain;
obtaining second transaction data associated with the second transaction request and corresponding to the second set of features;
generating a second data structure according to the feature arrangement for the machine learning model based on the second transaction data;
providing the second data structure to the machine learning model; and
processing the second transaction request based on a second outcome from the machine learning model.
14 . The method of claim 13 , wherein the generating the second data structure further comprises:
extracting, from the second transaction data, a portion of the second transaction data corresponding to the set of common features;
inserting the subset of the second transaction data in the domain-independent section of the second data structure; and
inserting the second transaction data in a second domain-specific section of the second data structure that corresponds to the second set of features.
15 . A non-transitory machine-readable medium having stored thereon machine-readable instructions executable to cause a machine to perform operations comprising:
obtaining a first plurality of data records comprising first data associated with a first domain and corresponding to a first set of features;
determining a second set of features associated with a second domain different from the first domain, wherein the second set of features is different from the first set of features;
determining a set of common features between the first set of features and the second set of features, wherein the set of common features is a first subset of the first set of features, and wherein each feature in a second subset of the first set of features is not common with the second set of features;
accessing a machine learning model configured to perform tasks associated with the first domain and the second domain, wherein the machine learning model comprises an internal structure to process data corresponding to a set of input features arranged in a feature arrangement associated with the first domain and the second domain, wherein the internal structure of the machine learning model comprises (i) a first set of input nodes corresponding to a domain-independent section of the feature arrangement that includes the set of common features and having a first set of connections connecting the first set of input nodes to a first set of internal nodes of the machine learning model and (ii) a second set of input nodes corresponding to a first domain-specific section of the feature arrangement that includes the first set of features and having a second set of connections connecting the second set of input nodes to a second set of internal nodes of the machine learning model;
generating a data structure for organizing the first data in a format usable to train the machine learning model, wherein the data structure comprises a first set of data fields corresponding to the set of common features and a second set of data fields corresponding to the first set of features;
populating the data structure based on the first data; and
training the machine learning model using the data structure.
16 . The non-transitory machine-readable medium of claim 15 , wherein the first domain is associated with a first type of transaction conducted via a first online portal, and wherein the second domain is associated with a second type of transaction conducted via a second online portal.
17 . The non-transitory machine-readable medium of claim 16 , wherein the first type of transaction comprises a login transaction for logging a user in to the first online portal.
18 . The non-transitory machine-readable medium of claim 16 , wherein the first type of transaction comprises a payment transaction for transferring funds between two accounts via the first online portal.
19 . The non-transitory machine-readable medium of claim 15 , wherein the internal structure of the machine learning model further comprises a third set of input nodes corresponding to a second domain-specific section of the feature arrangement that includes the second set of features, and wherein the data structure further comprises a third set of data fields corresponding to the second set of features.
20 . The non-transitory machine-readable medium of claim 15 , wherein the operations further comprise:
receiving a transaction request associated with the first domain;
determining, for the transaction request, transaction data corresponding to the first set of features;
generating, using at least the transaction data, input data for the machine learning model according to an arrangement associated with the data structure;
providing the input data to the machine learning model;
obtaining an outcome from the machine learning model; and
processing the transaction request based on the outcome.