International Business Machines Corporation

Vmware, Inc.

Other

Ji Hui Yang

Jing James Xu

Si Er Han

Xue Ying Zhang

Xiao Ming Ma

Jun Song Wang

Dakuo Wang

Jing Xu

Chuang Gan

Bei Chen

Jun Wang

Wen Pei Yu

Steven G Barbee

Dhavalkumar C Patel

Syed Yousaf Shah

Xiaoyang Yang

Lei Gao

Peter Daniel Kirchner

Horst Cornelius Samulowitz

Long H Vu

Yang Zhang

Gregory Bramble

Saket Sathe

Xuan-Hong Dang

Arunima Chaudhary

Rui Wang

Abel Nicolas Valente

Daniel Karl I Weidele

John Dillon Eversman

Voranouth Supadulya

Jing Min Xu

Dong Hai Yu

Yonghua Lin

Lingfei Wu

Petros Zerfos

Thanh Lam Hoang

Xing Wei

Mo Yu

Wesley M Gifford

Craig Reginald Tomlyn

Sier Han

Qian Pan

Lin Ju

Song Bo

Yi Wang

Xiaoding Bian

Xuye Liu

Long Vu

Jun Qi Zhang

Zhiyuan Wang

Zongyao Zhang

Jishuo Yang

Kai Xa Li

Jing Jia

A method, system, and computer program product for incremental machine learning for a parametric machine learning model are disclosed. The method may include processing samples comprising historical samples and new samples with an existing parametric machine learning model to obtain at least one prediction residual of each of the samples, wherein the existing parametric machine learning model was trained based on the historical samples. The method may further include clustering the samples based on the at least one prediction residual of each of the samples and features of each of the samples. The method may further include sampling samples in each cluster to ensure that each cluster includes substantially similar number of sampled samples. The method may further include updating the existing parametric machine learning model to obtain an updated parametric machine learning model based on sampled samples in each cluster.

Incremental machine learning for a parametric machine learning model

A method, a computer system, and a computer program product for receiving a plurality of unreduced training data records (TDRs), using a first autoencoder on the plurality of unreduced TDRs to obtain a first plurality of reduced TDRs, clustering the first plurality of reduced TDRs into a plurality of clusters, wherein each first reduced TDR cluster is assigned a respective cluster label, transferring the respective cluster label from each first reduced TDR cluster to a corresponding unreduced TDR cluster, and performing stratified sampling to form K data blocks of unreduced TDRs from a plurality of unreduced TDR clusters.

Dimension reduction in the context of unsupervised learning

A computer-implemented method for anomaly detection for a time series data is provided. Aspects include receiving a time series data including a plurality of sequential data points, calculating an expected next value for the time series data based on the plurality of sequential data points, and receiving an actual next value corresponding to the time series data. Aspects also include calculating an anomaly strength estimate based on the expected next value and the actual next value, identifying one of a plurality of anomaly detection pipelines based on the anomaly strength estimate and a portrait associated with each of the plurality of anomaly detection pipelines, and obtaining an anomaly prediction by inputting the time series data and the actual next value into the one of the plurality of anomaly detection pipelines.

Anomaly detection for time series data

A computer-implemented method for treating post-modeling data includes computing, sequentially for each category of a feature, a category importance (CI) value. The CI value is based on a model accuracy change when records of a category being examined are reassigned to a remaining set of categories of the feature according to a cumulative distribution of records among the remaining set of categories of the feature, wherein the remaining set of categories include all categories of the feature, except for the category being examined. A post-modeling category is performed to merge of each category having the CI value less than a CI value threshold.

Key category identification and visualization

An approach to time-series data point anomaly detection may be presented. Data point anomalies in time-series data can cause a cascade of incorrect predictions in a time-series data prediction model. Presented herein may be an approach to decompose a time-series training data set into elementary components, such as seasonal, trend and residual. The approach may determine one or more confidence intervals for elementary components of data points including level shift, variance, and outlier. From these confidence intervals, new data points can be analyzed and identified as anomaly data points. The approach may also prevent anomaly data points from being incorporated into a time series data prediction model, reducing prediction error in the prediction model.

Optimization of time-series anomaly detection

Embodiments of the present disclosure relate to a method, system, and computer program product for efficient execution of a decision tree. According to the method, respective target values of a plurality of attributes of a target entity are obtained. Representations of a plurality of leaf nodes of a decision tree are obtained. Each of the representations indicates respective statistic values of a plurality of attributes of historical entities and a statistic prediction result determined from historical prediction results output at a respective one of the plurality of leaf nodes for the historical entities. Distance measures between the target entity and the plurality of leaf nodes are determined based on the target values and the statistic values indicated by the representations of the plurality of leaf nodes. A target prediction result for the target entity is determined based on the distance measures and the statistic prediction results of the historical entities.

Efficient execution of a decision tree

Performing predictive analysis on running batch jobs is provided. A series of batch end time predictive models is retrieved according to a sequence of milestone jobs in a batch of jobs. Retrieved batch end time predictive models are assembled into an aggregate batch end time predictive model to increase accuracy and stability of an end time prediction of the batch of jobs. The aggregate batch end time predictive model is utilized to predict an end time of the batch of jobs during running of the batch of jobs to form a predicted end time of the batch of jobs.

Predictive analysis on running batch jobs

To automate time series forecasting machine learning pipeline generation, a data allocation size of time series data may be determined based on one or more characteristics of a time series data set. The time series data may be allocated for use by candidate machine learning pipelines based on the data allocation size. Features for the time series data may be determined and cached by the candidate machine learning pipelines. Predictions of each of the candidate machine learning pipelines using at least the one or more features may be evaluated. A ranked list of machine learning pipelines may be automatically generated from the candidate machine learning pipelines for time series forecasting based upon evaluating predictions of each of the one or more candidate machine learning pipelines.

Automated time series forecasting pipeline generation

Automated development and training of deep forest models for analyzing data by growing a random forest of decision trees using data, determining Out-of-bag (OOB) predictions for the forest, appending the OOB predictions to the data set, and growing an additional forest using the data set including the appended OOB predictions, and combining the output of the additional forest, then utilizing the model to classify data outside the training data set.

Deep forest model development and training

An approach is provided in which the approach generates a parallel chart based on multiple records that includes a set of variable values corresponding to a set of variables. To generate the parallel chart, the approach arranges the set of variables on the parallel chart in a variable order based on at least one variable arrangement rule. The approach arranges an initial variable value order for each one of the set of variables, and computes a lucidity score based on the variable order and the initial variable value order of each of the set of variables. The lucidity score is a measurement of the clarity of the parallel chart. The approach adjusts the variable value order of at least one of the set of variables to increase the lucidity score and optimizes the clarity of the parallel chart based on the adjusted variable value order.

Parallel chart generator

Provided are a computer-implemented method, a system, and a computer program product. The method comprises extracting features from a plurality of base models in an ensemble model. The plurality of base models are configured to provide respective prediction results. The ensemble model is configured to provide an overall prediction result from the prediction results of the plurality of base models. The features are associated with time performance of the base models. The method further comprises clustering the plurality of base models into a plurality of clusters based on the extracted features. The method further comprises assigning the plurality of base models to a plurality of parallel computation units based on the plurality of clusters.

Parallelized scoring for ensemble model

A method of clustering data generated by an unknown model, the method including accessing the data, wherein the data includes a prediction target and a confidence, extracting a data group with high prediction confidence from the data, wherein the data group comprises a plurality of data cases, and where each of the data cases is described by a plurality of predictors, identifying high rank predictors of each the data cases in the data group, transforming the data group into a transformed data group including only the high rank predictors for each of the data cases, wherein the high rank predictors are ranked within each of the data cases included in the transformed data group, clustering the transformed data group to generate clusters, and profiling the clusters to extract an insight about the unknown model.

Model-free high confidence data clustering

An embodiment includes parsing conversation data to extract a message dataset and a user dataset. The embodiment classifies the message dataset into a category using machine learning processing and identifies the category as a top category based at least in part on an amount of the conversation data associated with the category. The embodiment generates impact data associated with the user dataset based on actions in the conversation data by the user. The embodiment generates role data associated with the user by applying a rule to the conversation data for the user. The embodiment generates key index data associated with the message dataset by identifying interactions with a message represented by the message dataset. The embodiment generates output data arranged according to a specified data format that is compatible with a user interface.

Summarizing conversations in a messaging application with internal social network discovery

In an approach to efficient model adjustment utilizing a dual model system, one or more computer processors create a subset of a dataset utilizing a trained primary model; create a secondary model based on the created subset of the dataset; calculate a confidence of a case utilizing the trained primary model, wherein the confidence is a robustness indicator of a model indicating a capacity of the model to meet or exceed performance when applied to the dataset; responsive to the calculated confidence not exceeding a confidence threshold, generate a score of the case utilizing the created secondary model; responsive to an incorrect classification, update the created subset of the dataset with the case; retrain the secondary model utilizing the updated subset of the dataset.

Dual model incremental learning

Systems, computer-implemented methods, and computer program products to facilitate visualization of a model selection process are provided. According to an embodiment, a system can comprise a memory that stores computer executable components and a processor that executes the computer executable components stored in the memory. The computer executable components can comprise an interaction backend handler component that obtains one or more assessment metrics of a model pipeline candidate. The computer executable components can further comprise a visualization render component that renders a progress visualization of the model pipeline candidate based on the one or more assessment metrics.

Visualization of a model selection process in an automated model selection system

Growing community of inventors

Beijing, China

Ji Hui Yang