Abstract: A machine learning (ML) system and methods are provided that are configured to detect concept drift in ML models. The system includes a processor and a computer readable medium operably coupled thereto, the computer readable medium comprising a plurality of instructions stored in association therewith that are accessible to, and executable by, the processor, to perform drift detection operations which include receiving a first data set for use during online training of a first ML model, determining a change to an uncertainty bound metric associated with classifiers for features utilized by the first ML model, identifying that the first data set causes the concept drift with the online training of the first ML model, determining characterization information about a type of the concept drift, generating an ML update paradigm based on the concept drift and the characterization information, alerting an ML model updater of the ML update paradigm.

Abstract: The present disclosure provides a machine learning system and method configured to induce neuron activity in a neural network of the machine learning system. Each of the system and method selects a neural network with a multilayer perceptron and performs incremental learning cycle on the multilayer perceptron. An input neuron is modified by strengthening connections between the input neuron and additional neurons. A second input neuron may be modified by weakening connections between the second input neuron and additional neurons. Activation functions associated with the neurons in the multilayer perceptron may be adjusted. Batches of data are run through the multilayer perceptron until a set constraint is met, at which point a prediction is generated for each of the batches from input data from the neural network.

Abstract: A computerized-method for analyzing financial data to improve performance of a concept-drift-detector that is providing alerts of drift to an update component of a machine learning model for fraud prediction and detection, is provided herein. The computerized-method includes retrieving a time-series data of financial transactions having one or more features, during a time unit. For each feature, detecting a process of values of the feature to determine a type of the process. When the type of the process of a feature is determined as nonstationary, determining a subtype thereof and if the process is feasible for rectification to a stationary process, rectifying it. When the type of the process is determined as stationary, determining its subtype and when the type of the process is determined as nonstationary and the process is not feasible for rectification, forwarding the time-series data, the type of the process and the subtype to the concept-drift-detector.

Abstract: A computerized-method for maintaining ethical Artificial-Intelligence by generating a representative-training-sample-dataset for a fraud-detection Machine-Learning (ML) model, by: (i) operating a representative-dataset-preparation module to generate a representative-training-sample-dataset by operating balanced-sampling on randomly-selected preconfigured-number of financial-transactions. The balanced-sampling may be operated by applying a configurable-rule on at least two values of a parameter of non-sensitive PII parameters of each financial-transaction by a low-frequency value; (ii) training the fraud-detection ML model on the representative-training-sample-dataset; and (iii) deploying the trained fraud-detection ML model in a finance-system in test-environment, and operating the trained fraud-detection ML model on a stream-of-financial-transactions to predict a risk-score for each financial-transaction.

Abstract: System and method for generating case-based data including receiving, input data describing an event of interest; if the input data is not in a format of a property graph then transforming the input data into a first property graph describing the event of interest and representing a first network, wherein the first property graph includes a plurality of network elements and properties of at least some of the plurality of network elements, wherein the network elements include entities and links describing relationships between the entities; changing a network element in the first property graph to create a second property graph of a new network; and using the second property graph as the case-based data. New properties may be generated for the entities and links.

Abstract: A computerized-method for real-time detection of real concept drift in predictive machine learning models, by processing high-speed streaming data. The computerized-method includes: receiving a real-time data stream having labeled and unlabeled instances. Obtaining a window of ‘n’ instances having a portion of the ‘n’ instances as reliable labels. Computing posterior distribution of the reliable labels; and operating a Drift-Detection (DD) module. The DD module is configured to: operate a kernel density estimation on the computed posterior distribution for sensitivity control of the DD module; operate an error rate function on the estimated kernel density to yield an error value; and train an incremental estimator module, according to the kernel density estimation. When the error value is not above a preconfigured drift threshold repeating operations (i) through (iii), else when the error value is above the preconfigured drift threshold, at least one concept drift related action takes place.

Abstract: A risk-prediction-preparation module to generate a risk-prediction-model, is provided herein. The risk-prediction-preparation module includes accessing a data-storage of transactions to operate a group-by operation on transactions related to data-points, according to a logical-entity into entities. Then, clustering entities of a clean-financial dataset into clusters. Selecting data-points of: (a) entities from the clusters to a first dataset and (b) a preconfigured amount of entities randomly to a second dataset. Selecting all entities that have at least one ‘fraudulent’ data-points in at least one related data-point to add all the entities to the first dataset and the second dataset. Using vectorized and scaled extracted features for training a first machine-learning-model of fraud detection on the first dataset and training a second machine-learning-model of fraud detection on the second dataset to collect results.

Abstract: A device, system, and method for training a machine learning model using incremental learning without forgetting. A sequence of training tasks may be respectively associated with training samples and corresponding labels. A subset of shared model parameters common to the training tasks and a subset of task-specific model parameters not common to the training tasks may be generated. The machine learning model may be trained in each of a plurality of sequential task training iteration by generating the task-specific parameters for the current training iteration by applying a propagator to the training samples associated with the current training task and constraining the training of the model for the current training task by the training samples associated with a previous training task in a previous training iteration, and classifying the samples for the current training task based on the current and previous training task samples.

Abstract: A computerized-system and method for generating a reduced-size superior labeled training-dataset for a high-accuracy machine-learning-classification model for extreme class imbalance by: (a) retrieving minority and majority class instances to mark them as related to an initial dataset; (b) retrieving a sample of majority instances; (c) selecting an instance to operate a clustering classification model on it and the instances marked as related to the initial dataset to yield clusters; (d) operating a learner model to: (i) measure each instance in the yielded clusters according to a differentiability and an indicativeness estimators; (ii) mark measured instances as related to an intermediate training dataset according to the differentiability and the indicativeness estimators; (e) repeating until a preconfigured condition is met; (f) applying a variation estimator on all marked instances as related to an intermediate training dataset to select most distant instances; and (g) marking the instances as related to

Abstract: A computerized-method for real-time detection of real concept drift in predictive machine learning models, by processing high-speed streaming data. The computerized-method includes: receiving a real-time data stream having labeled and unlabeled instances. Obtaining a window of ‘n’ instances having a portion of the ‘n’ instances as reliable labels. Computing posterior distribution of the reliable labels; and operating a Drift-Detection (DD) module. The DD module is configured to: operate a kernel density estimation on the computed posterior distribution for sensitivity control of the DD module; operate an error rate function on the estimated kernel density to yield an error value; and train an incremental estimator module, according to the kernel density estimation. When the error value is not above a preconfigured drift threshold repeating operations (i) through (iii), else when the error value is above the preconfigured drift threshold, at least one concept drift related action takes place.

Abstract: An artificial intelligence system configured to detect anomalies in transaction data sets. The system includes a processor and a computer readable medium operably coupled thereto, the computer readable medium comprising a plurality of instructions stored in association therewith that are accessible to, and executable by, the processor, to perform modeling operations which include receiving a first data set for training a first machine learning model to detect anomalies in the transaction data sets using a machine learning technique, accessing at least one micro-model trained using at least one second data set separate from the first data set, determining risk scores from the first data set using the at least one micro-model, enriching the first data set with the risk scores, and determining the first machine learning model for the enriched first data set using the machine learning technique.

Abstract: A computerized-system and method for generating a reduced-size superior labeled training-dataset for a high-accuracy machine-learning-classification model for extreme class imbalance by: (a) retrieving minority and majority class instances to mark them as related to an initial dataset; (b) retrieving a sample of majority instances; (c) selecting an instance to operate a clustering classification model on it and the instances marked as related to the initial dataset to yield clusters; (d) operating a learner model to: (i) measure each instance in the yielded clusters according to a differentiability and an indicativeness estimators; (ii) mark measured instances as related to an intermediate training dataset according to the differentiability and the indicativeness estimators; (e) repeating until a preconfigured condition is met; (f) applying a variation estimator on all marked instances as related to an intermediate training dataset to select most distant instances; and (g) marking the instances as related to

Abstract: A computer implemented method and system for optimization of model parameters of at least one predictive model for detecting suspicious financial activity. The processor may select a reduced set of key indicators and corresponding scores to optimize from each of the at least one predictive model, each key indicator and corresponding score in the reduced set having an influence ranking above a predetermined influence ranking. The processor may select a best performing model candidate based on an evaluation of each reduced set of key indicators and corresponding scores. The processor may preform gradient-ascent optimization on the best performing model candidate and the at least one random model to generate a set of at least two new models for each of the best performing model candidate and the at least one random model. The processor may select the new model with the highest performance ranking.

Abstract: System and method for generating case-based data including receiving, input data describing an event of interest; if the input data is not in a format of a property graph then transforming the input data into a first property graph describing the event of interest and representing a first network, wherein the first property graph includes a plurality of network elements and properties of at least some of the plurality of network elements, wherein the network elements include entities and links describing relationships between the entities; changing a network element in the first property graph to create a second property graph of a new network; and using the second property graph as the case-based data. New properties may be generated for the entities and links.

Abstract: A computer implemented method and system for optimization of model parameters of at least one predictive model for detecting suspicious financial activity. The processor may select a reduced set of key indicators and corresponding scores to optimize from each of the at least one predictive model, each key indicator and corresponding score in the reduced set having an influence ranking above a predetermined influence ranking. The processor may select a best performing model candidate based on an evaluation of each reduced set of key indicators and corresponding scores. The processor may preform gradient-ascent optimization on the best performing model candidate and the at least one random model to generate a set of at least two new models for each of the best performing model candidate and the at least one random model. The processor may select the new model with the highest performance ranking.

Abstract: A system or method may include receiving, by a processor, data describing a network, wherein the network includes a plurality of entities and links describing relationships between the plurality of entities. The method may further include identifying a set of seed entities from the plurality of entities based on predefined rules. The method may further include generating a set of sub-networks based on the set of identified seed entities, wherein each of the sub-networks may include one or more other entities of the plurality of entities having at least one link to the at least one seed entity. The method may further include calculating a risk score for each of the generated sub-networks.