Abstract: Systems and methods for efficient speech representation are disclosed. In one embodiment, a method for efficient speech representation may include training a teacher model using training data to get embeddings from intermediate and/or final layers of the teacher model; training a student model using training data processed with audio distortion to have outputs matching the embeddings from the teacher model; and injecting known hand-crafted audio features into intermediate or final layers of the student model.

Abstract: Systems and methods for unit test generation using reinforcement learning augmented transformer architectures are disclosed. A method may include: receiving raw data for source code from a database; identifying a function for which a unit test will be generated and an existing unit test for that function; receiving the function and the existing unit test; generating a generated unit test for the function using the function for the unit test and the existing unit test using a deep learning model; applying a loss function to the generated unit test, wherein the loss function is based on a comparison between the generated unit test and the existing unit test and results of the application of the loss function are fed back to the transformer computer program; simulating the generated unit test using a simulator; generating scalar feedback; and refining the generated unit test using the scalar feedback.

Abstract: A global partitioning-based method for anonymizing a dataset of biometric data may include an anonymization computer program: (1) receiving a value k representing a number of records to hide a biometric datum among, a value t that represents a t-closeness parameter for a t-close distribution, a weight parameter, and a first number of features to retain for determining an attribute of interest; (2) receiving the attribute of interest; (3) calculating a distribution of the attribute of interest in a biometric dataset; (4) splitting the biometric dataset into a plurality of k-sized clusters that satisfy the t-close distribution; (5) anonymizing each biometric datum in the plurality of k-sized clusters using a weighted average of landmarks for the biometric datums in k-sized clusters using the weight parameter; (6) adding each anonymized biometric datum into an anonymized biometric dataset; and (7) persisting the anonymized biometric dataset.

Abstract: Systems and methods for scalable and flexible federated learning frameworks are disclosed. A method may include: (1) receiving, by a computer program executed by an electronic device and from a client, a project for federated learning using a training federation, the training federation comprising a plurality of clients; (2) generating, by the computer program, a configuration file that reflects a set-up for the training federation; (3) receiving, by the computer program, files necessary to build containers, wherein at least some of the files are customized by the client; (4) generating, by the computer program, containers comprising the configuration file and files necessary to build the containers; and (5) deploying, by the computer program, the containers to a client compute environment for the client as a client node, wherein the client node is configured to join the training federation as a server and/or a participant.

Abstract: Systems and methods for simultaneous network pruning and parameter optimization are disclosed. A method may include: (1) receiving a network to optimize, the network comprising a plurality of layers; (2) selecting layer pruning and/or channel pruning for the layers within the network; (3) providing a gating module at layers within the network, wherein each gating module opens or closes a gate in the gating module based on an output of a binary head; (4) training parameters for the network; (5) extracting gate open/close features from the gating modules; (6) optimizing a loss function for the network using a polarization regularizer to reach a consensus static sub-network; and (7) updating parameters for each gating module in the network consistent with the consensus static sub-network.

Abstract: Systems and methods for auto-thresholding using pairwise feature cross-correlation for hyperparameter value selection are disclosed. A method may include a hyperparameter value optimization computer program: receiving data to be used by a clustering algorithm; receiving a selection of a hyperparameter value to tune; for each possible hyperparameter value, executing the clustering algorithm resulting in a set of clusters for each hyperparameter value; extracting a series of cluster features from the set of clusters; performing pairwise cross-correlation on the series of cluster features resulting in potential candidates for an optimal hyperparameter value; aggregating maximum or minimum values for the hyperparameter value at their respective indices; selecting an optimum value for the hyperparameter value; and outputting the optimum value for the hyperparameter value to the clustering algorithm.

Abstract: Systems and methods for generating runtime predictions in distributed computer architectures are disclosed. According to one embodiment, a method for generating runtime predictions in a distributed computer architectures may include: (1) receiving, by a runtime prediction computer program executed by an electronic device, training data regarding completion of a plurality of risk compute jobs; (2) extracting, by the runtime prediction computer program, bucketing statistics and instrument level features from the data; (3) training, by the runtime prediction computer program, a machine learning model with the training data, the extracted bucketing statistics, and the instrument level features; (4) receiving, by the runtime prediction computer program, a plurality of jobs for a period of time; and (5) calculating, by the runtime prediction computer program, an instrument cost for each of the jobs using the machine learning model.

Abstract: Systems and methods for improving efficiency and control compliance across software development life cycles using domain-specific controls are disclosed. In one embodiment, a method may include: (1) receiving, by an analysis and recommendation service computer program, an identification of a domain for code; (2) identifying, by the analysis and recommendation service computer program, rules and/or control patterns for the domain; (3) receiving, by the analysis and recommendation service computer program, the code; (4) checking, by the analysis and recommendation service computer program, the code for compliance with the rules or control patterns; and (5) deploying, by analysis and recommendation service computer program, the code in response to the code being in compliance.

Abstract: A method for facilitating identification of secrets in source code by using machine learning is provided. The method includes retrieving a plurality of files from a repository, each of the plurality of files including a source code file; parsing the source code file to identify a training feature; associating a predetermined label with the training feature, the predetermined label corresponding to a secret label and a non-secret label; training a model by using the training feature and the corresponding predetermined label; receiving, via a graphical user interface, a test file, the test file including a set of source codes; parsing the set of source codes to identify a feature; and determining, by using the model, a first characteristic of the feature.

Abstract: A method for combining an embedding of a graph having a plurality of nodes and edge connections and classifications of distributed ledger-based activities may include: receiving information for historical distributed ledger-based activities comprising identifications of the plurality of nodes in the graph as involved in illicit activities, licit activities, or unknown; applying sampling to sample labels and associated features; training a first classifier based on the plurality of historical distributed ledger-based activities, the node information, and the sample labels; receiving a current distributed ledger-based activity from the distributed ledger network; predicting a classification for the current distributed ledger-based activity using the first classifier; extracting features from the current distributed ledger-based activity; enriching the extracted features based on one or more neighbors of the nodes involved in the current distributed ledger-based activity; reclassifying the current distributed le

Abstract: Systems and methods for weakly supervised unit test quality scoring are disclosed. According to one embodiment, a method may include: parsing, by a generative model computer program, a plurality of code snippets in a repository using an abstract syntax tree; receiving, by the generative model computer program, a plurality of binary labelling functions from a labelling function repository; creating, by the generative model computer program, a labelling matrix by applying the binary labelling functions to the parsed code snippets; training, by the generative model computer program, a probabilistic generative model using the labelling matrix resulting in a vector of pseudo-labels; building, by a unit test scoring computer program, a discriminative model, wherein the discriminative model receives an array of real value inputs; and training, by the unit test scoring computer program, the discriminative model using the parsed code snippets, the vector of pseudo labels, and the abstract syntax tree.

Abstract: Systems and methods for risk diagnosis of cryptocurrency addresses on blockchains using anonymous and public information are disclosed. In one embodiment, a method may include a risk diagnosis computer program executed by a server: receiving data with labels and data without labels from public data databases and/or anonymous data databases; fitting the data without labels into a plurality of clusters using unsupervised learning; assigning a risk level to each of the plurality of clusters; verifying the risk level for each of the plurality of clusters using the data with labels; receiving unseen data; and predicting a risk level associated with the unseen data by constructing features and text embeddings from the unseen data, clustering the unseen data based on a distance measurement to one of the plurality of clusters, and returning a risk level associated with the cluster.

Abstract: A method for facilitating automatic assignment of code topics is provided. The method includes accessing a database, the database including data that is associated with a known code topic and data that is associated with an unknown code topic; parsing the database to identify a data file that includes a code snippet, document strings, and/or dependencies. Dividing the identified data file into a training data set and a testing data set; generating a first set of features for each element of the code snippet, document strings, and dependencies, for the training data set; generating a second set of features for the testing data set; identifying, by using a model, a representative feature for the training data set based on frequency distribution of the training data set, the representative feature including a corresponding probability; and determining, by using clustering, a code topic for the data file.

Abstract: A method for automatically providing library package recommendations is disclosed. The method includes aggregating, via a communication interface, sets of source code from a repository, each of the sets of source code including a code snippet that relates to a portion of source code for a software program; initiating, by using a model, code context analysis of the code snippet to identify an alternative library package and a replaceable library package; performing, by using the model, implemented library analysis of the code snippet to determine a complementary library package; generating recommendations for the code snippet, the recommendations including information that relates to the complementary library package, the alternative library package, and the replaceable library package; and associating, in the repository, the generated recommendations with the code snippet.

Abstract: Systems and methods for automated code analysis and tagging are disclosed. In one embodiment, a method for automated code analysis and tagging may include: (1) receiving, by a code annotation computer program executed by a computer processor, a training code snippet from a training codebase; (2) parsing, by the code annotation computer program, the training code snippet into a data structure; (3) quantifying, by the code annotation computer program, the data structure, (4) parsing, by the code annotation computer program, a docstring associated with the training code snippet into a plurality of keywords; (5) quantifying, by the code annotation computer program, the plurality of keywords; and (6) training, by the code annotation computer program, a code annotation model based on a similarity between the quantified data structure and a smoothing parameter for a Dirichlet prior smoothing estimate.

Abstract: A method for automatically providing library package recommendations is disclosed. The method includes aggregating, via a communication interface, sets of source code from a repository, each of the sets of source code including a code snippet that relates to a portion of source code for a software program; initiating, by using a model, code context analysis of the code snippet to identify an alternative library package and a replaceable library package; performing, by using the model, implemented library analysis of the code snippet to determine a complementary library package; generating recommendations for the code snippet, the recommendations including information that relates to the complementary library package, the alternative library package, and the replaceable library package; and associating, in the repository, the generated recommendations with the code snippet.

Abstract: Systems and methods for weak supervision classification with probabilistic generative latent variable models are disclosed. A method for weak supervision classification with probabilistic generative latent variable models may include: (1) receiving, by a generative model computer program, a plurality of records from a database; (2) receiving, by the generative model computer program, a plurality of user-defined label functions; (3) labeling, by the generative model computer program, each of the plurality of records with each of the plurality of user-defined label functions; (4) representing, by the generative model computer program, the plurality of records that are labeled with the user-defined label functions in a matrix; (5) performing, by the generative model computer program, probabilistic latent variable model analysis on the matrix using a probabilistic generative latent variable model; and (6) outputting, by the generative model computer program, a labeled dataset for the plurality of records.

Abstract: Systems and methods for locality preserving federated learning are disclosed. In one embodiment, a method for locality preserving federated learning may include: (1) receiving, at an aggregator computer program and from each of a plurality of clients, weights for each client's local machine learning model; (2) generating, by the aggregator computer program, an averaged machine learning model based on the received weights; (3) sharing, by the aggregator computer program, the averaged machine learning model with the plurality of clients; and (4) applying, by each client, a scaling factor to the averaged machine learning model to update its local machine learning model.

Abstract: Systems and methods for federated model validation and data verification are disclosed. A method may include: (1) receiving, by a local computer program executed by client system, a federated machine learning model from a federated model server; (2) testing, by the local computer program and using a policy service, the federated machine learning model for vulnerabilities to attacks; (3) accepting, by the local computer program, the federated machine learning model in response to the federated machine learning model passing the testing; (4) training, by the local computer program, the federated machine learning model using input data comprising local data and outputting training parameters; (5) identifying, by the local computer program using the policy service, accidental leakage and/or contamination by comparing the training parameters to the input data; and (6) providing, by the local computer program, the training parameters to the federated model server.

Abstract: A method may include: retrieving a plurality of code snippets from code repositories; generating a syntax representation, a property representation for each of the code snippets; receiving a query comprising a query code snippet, natural language keywords, and/or a string pattern; performing string-based matching and parser/syntax tree matching on the query and the tree representations, syntax matching on the query and the syntax representations, and property matching on the query and the property representations, wherein each of the matchings results in a score; combining the scores of the string-based matching, the parser/syntax tree matching, the syntax matching, and/or the property matching; identifying a plurality of code snippets of interest based on the combined scores; classifying the code snippets of interest using a machine learning classifier; outputting a list of the code snippets of interest with their classifications; and training the machine learning classifier based on user feedback.