Abstract: Embodiments of the disclosed technology include a representation learning model for classification of natural language text. In embodiments, a classification model comprises a feature model and a classifier. The feature model may be hierarchical in nature: data may pass through a series of representations, decreasing in specificity and increasing in generality. Intermediate levels of representation may then be used as automatically learned features to train a statistical classifier. Specifically, the feature model may be based on a hierarchical Pitman-Yor process. In embodiments, once the feature model has been expressed as a Bayesian Belief Network and some aspect of the feature model has been selected for prediction, the feature model may be attached to the classifier. In embodiments, after training, potentially using a mix of labeled and unlabeled data, the classification model can be used to classify documents such as call transcripts based on topics of conversation represented in the transcripts.

Abstract: Systems and methods for an accelerated tuning of hyperparameters of a model supported with prior learnings data include assessing subject models associated with a plurality of distinct sources of transfer tuning data, wherein the assessing includes implementing of: [1] a model relatedness assessment for each of a plurality of distinct pairwise subject models, and [2] a model coherence assessment for each of the plurality of distinct pairwise subject models; constructing a plurality of distinct prior mixture models based on the relatedness metric value and the coherence metric value for each of the plurality of distinct pairwise subject models, identifying sources of transfer tuning data based on identifying a distinct prior mixture model having a satisfactory model evidence fraction; and accelerating a tuning of hyperparameters of the target model based on transfer tuning data associated with the distinct prior mixture model having the satisfactory model evidence fraction.

Abstract: In one embodiment, the disclosed technology involves: digitally generating and storing a machine learning statistical topic model in computer memory, the topic model being programmed to model call transcript data representing words spoken on a call as a function of one or more topics of a set of topics that includes pre-seeded topics and non-pre-seeded topics; programmatically pre-seeding the topic model with a set of keyword groups; programmatically training the topic model using unlabeled training data; conjoining a classifier to the topic model to create a classifier model; programmatically training the classifier model using labeled training data; receiving target call transcript data; programmatically determining at least one of one or more topics of the target call or one or more classifications of the target call; and digitally storing the target call transcript data with additional data indicating the determined topics and/or classifications of the target call.

Abstract: A disclosed example includes implementing a first worker instance and a second worker instance to operate in parallel; running a first tuning operation via the first worker instance to tune first hyperparameters; running a second tuning operation via the second worker instance using a Bayesian-based optimization to determine a hyperparameter configuration to evaluate next; evaluating the hyperparameter configuration for an external model using a surrogate model; and selecting the hyperparameter configuration for the external model.

Abstract: Systems and methods for tuning hyperparameters of a model include receiving a tuning request for tuning hyperparameters, the tuning request includes a first and a second objective function for the machine learning model. The first and second objective functions may output metric values that do not improve uniformly. Systems and methods additionally include defining a joint tuning function that is based on a combination of the first and second objective functions; executing a tuning operation; identifying a Pareto efficient frontier curve defined by a plurality of distinct hyperparameter values; applying metric thresholds to the Pareto efficient frontier curve; demarcating the Pareto efficient frontier curve into at least a first infeasible section and a second feasible section; searching the second feasible section of the Pareto efficient frontier curve for one or more proposed hyperparameter values; and identifying at least a first set of proposed hyperparameter values based on the search.

Abstract: Disclosed examples include after a first tuning of hyperparameters in a hyperparameter space, selecting first hyperparameter values for respective ones of the hyperparameters; generating a polygonal shaped failure region in the hyperparameter space based on the first hyperparameter values; setting the first hyperparameter values to failure before a second tuning of the hyperparameters; and selecting second hyperparameter values for the respective ones of the hyperparameters in a second tuning region after the second tuning of the hyperparameters in the second tuning region, the second tuning region separate from the polygonal shaped failure region.

Abstract: Disclosed examples including generating a joint model based on first and second subject models, the first and second subject models selected based on a relationship between the first and second subject models; selecting the joint model from a plurality of joint models after a determination that entropy data points of the joint model satisfy a threshold, the entropy data points based on multiple tuning trials of the joint model; and providing tuning data associated with the joint model to a tuning session of a target model.

Abstract: Systems and methods for generating training data for a supervised topic modeling system from outputs of a topic discovery model are described herein. In an embodiment, a system receives a plurality of digitally stored call transcripts and, using a topic model, generates an output which identifies a plurality of topics represented in the plurality of digitally stored call transcripts. Using the output of the topic model, the system generates an input dataset for a supervised learning model by identify a first subset of the plurality of digitally stored call transcripts that include the particular topic, storing a positive value for the first subset, identifying a second subset that do not include the particular topic, and storing a negative value for the second subset. The input training dataset is then used to train a supervised learning model.

Abstract: Systems and methods for tuning hyperparameters of a model includes: receiving a multi-criteria tuning work request for tuning hyperparameters of the model of the subscriber to the remote tuning service, wherein the multi-criteria tuning work request includes: a first objective function of the model to be optimized by the remote tuning service; a second objective function to be optimized by the remote tuning service, the second objective function being distinct from the first objective function; computing a joint tuning func-tion based on a combination of the first objective function and the second objective function; executing a tuning opera-tion of the hyperparameters for the model based on a tuning of the joint function; and identifying one or more proposed hyperparameter values based on one or more hyperparam-eter-based points along a convex Pareto optimal curve.

Abstract: A system and method for accelerated tuning of hyperparameters includes receiving a multi-task tuning work request for tuning hyperparameters of a model, wherein the multi-task tuning work request includes: a full tuning task for tuning hyperparameters, wherein the full tuning task includes a first set of tuning parameters governing a first tuning operation; a partial tuning task for tuning the hyperparameters of the model, wherein the partial tuning task includes a second distinct set of tuning parameters governing a second tuning operation; executing the first tuning operation and the second tuning operation; generating a first suggestion set and a second suggestion set of one or more proposed values for the hyperparameters based on the execution of the full tuning task and the partial tuning task; and setting the partial tuning task as a proxy for the full tuning task thereby accelerating a tuning of the hyperparameters of the model.

Abstract: A system and method for accelerated tuning of hyperparameters includes receiving a multi-task tuning work request for tuning hyperparameters of a model, wherein the multi-task tuning work request includes: a full tuning task for tuning hyperparameters, wherein the full tuning task includes a first set of tuning parameters governing a first tuning operation; a partial tuning task for tuning the hyperparameters of the model, wherein the partial tuning task includes a second distinct set of tuning parameters governing a second tuning operation; executing the first tuning operation and the second tuning operation; generating a first suggestion set and a second suggestion set of one or more proposed values for the hyperparameters based on the execution of the full tuning task and the partial tuning task; and setting the partial tuning task as a proxy for the full tuning task thereby accelerating a tuning of the hyperparameters of the model.

Abstract: Systems and methods for tuning hyperparameters of a model includes: receiving a multi-criteria tuning work request for tuning hyperparameters of the model of the subscriber to the remote tuning service, wherein the multi-criteria tuning work request includes: a first objective function of the model to be optimized by the remote tuning service; a second objective function to be optimized by the remote tuning service, the second objective function being distinct from the first objective function; computing a joint tuning function based on a combination of the first objective function and the second objective function; executing a tuning operation of the hyperparameters for the model based on a tuning of the joint function; and identifying one or more proposed hyperparameter values based on one or more hyperparameter-based points along a convex Pareto optimal curve.

Abstract: In one embodiment, the disclosed technology involves: digitally generating and storing a machine learning statistical topic model in computer memory, the topic model being programmed to model call transcript data representing words spoken on a call as a function of one or more topics of a set of topics that includes pre-seeded topics and non-pre-seeded topics; programmatically pre-seeding the topic model with a set of keyword groups; programmatically training the topic model using unlabeled training data; conjoining a classifier to the topic model to create a classifier model; programmatically training the classifier model using labeled training data; receiving target call transcript data; programmatically determining at least one of one or more topics of the target call or one or more classifications of the target call; and digitally storing the target call transcript data with additional data indicating the determined topics and/or classifications of the target call.

Abstract: Embodiments of the disclosed technology include a representation learning model for classification of natural language text. In embodiments, a classification model comprises a feature model and a classifier. The feature model may be hierarchical in nature: data may pass through a series of representations, decreasing in specificity and increasing in generality. Intermediate levels of representation may then be used as automatically learned features to train a statistical classifier. Specifically, the feature model may be based on a hierarchical Pitman-Yor process. In embodiments, once the feature model has been expressed as a Bayesian Belief Network and some aspect of the feature model has been selected for prediction, the feature model may be attached to the classifier. In embodiments, after training, potentially using a mix of labeled and unlabeled data, the classification model can be used to classify documents such as call transcripts based on topics of conversation represented in the transcripts.

Abstract: Systems and methods for improving machine learning systems used to model topics on a plurality of calls are described herein. In an embodiment, a server computer receives plurality of digitally stored call transcripts that have been prepared from digitally recorded voice calls. The server computer uses a topic model of an artificial intelligence machine learning system, the topic model modeling words of a call as a function of one or more word distributions for each topic of a plurality of topics, to generate an output of the topic model which identifies the plurality of topics represented in the plurality of call transcripts. The server computer computes, for a particular topic of the plurality of topics a first value representing a vocabulary of the particular topic and a second value representing a consistency of the particular topic in two more call transcripts of the plurality of call transcripts which include the particular topic.

Abstract: Systems and methods for generating training data for a supervised topic modeling system from outputs of a topic discovery model are described herein. In an embodiment, a system receives a plurality of digitally stored call transcripts and, using a topic model, generates an output which identifies a plurality of topics represented in the plurality of digitally stored call transcripts. Using the output of the topic model, the system generates an input dataset for a supervised learning model by identify a first subset of the plurality of digitally stored call transcripts that include the particular topic, storing a positive value for the first subset, identifying a second subset that do not include the particular topic, and storing a negative value for the second subset. The input training dataset is then used to train a supervised learning model.

Abstract: In one embodiment, the disclosed technology involves: digitally generating and storing a machine learning statistical topic model in computer memory, the topic model being programmed to model call transcript data representing words spoken on a call as a function of one or more topics of a set of topics that includes pre-seeded topics and non-pre-seeded topics; programmatically pre-seeding the topic model with a set of keyword groups; programmatically training the topic model using unlabeled training data; conjoining a classifier to the topic model to create a classifier model; programmatically training the classifier model using labeled training data; receiving target call transcript data; programmatically determining at least one of one or more topics of the target call or one or more classifications of the target call; and digitally storing the target call transcript data with additional data indicating the determined topics and/or classifications of the target call.

Abstract: Systems and methods for generating training data for a supervised topic modeling system from outputs of a topic discovery model are described herein. In an embodiment, a system receives a plurality of digitally stored call transcripts and, using a topic model, generates an output which identifies a plurality of topics represented in the plurality of digitally stored call transcripts. Using the output of the topic model, the system generates an input dataset for a supervised learning model by identify a first subset of the plurality of digitally stored call transcripts that include the particular topic, storing a positive value for the first subset, identifying a second subset that do not include the particular topic, and storing a negative value for the second subset. The input training dataset is then used to train a supervised learning model.

Abstract: Disclosed examples include after a first tuning of hyperparameters in a hyperparameter space, selecting first hyperparameter values for respective ones of the hyperparameters; generating a polygonal shaped failure region in the hyperparameter space based on the first hyperparameter values; setting the first hyperparameter values to failure before a second tuning of the hyperparameters; and selecting second hyperparameter values for the respective ones of the hyperparameters in a second tuning region after the second tuning of the hyperparameters in the second tuning region, the second tuning region separate from the polygonal shaped failure region.

Abstract: A disclosed example includes implementing a first worker instance and a second worker instance to operate in parallel running a first tuning operation via the first worker instance to tune first hyperparameters; running a second tuning operation via the second worker instance using a Bayesian-based optimization to determine a hyperparameter configuration to evaluate next; evaluating the hyperparameter configuration for an external model using a surrogate model; and selecting the hyperparameter configuration for the external model.