International Business Machines Corporation

Stefan A G Van Der Stockt

Craig M Trim

Joel Russell Huffman

Aaron K Baughman

Mary E Rudden

Bernard Freund

Augustina Monica Ragwitz

Erik Rueger

Igor S Ramos

Rahul P Akolkar

Marc Dickenson

Frederik Frank Flöther

Angelo Danducci, Ii

Joseph Lindsey Sharpe, Iii

Kristi A Farinelli

Mark Marrara

Manali Jairam Chanchlani

Alex Swain

Sihang B Fang

Edward J Biddle

Xinyun Zhao

Ellen R Kolsto

Matthias Biniok

Sai Karthik Reddy Ginni

Maximilian Ott

Daniel Kuehn

Zhiyang Yu

In an approach to creating explanatory confusion matrices, responsive to receiving a machine learning model for analysis, a confusion matrix is calculated for the machine learning model, where each cell in the confusion matrix has a corresponding set of data. A link is created from each cell in the confusion matrix to the corresponding set of data. Responsive to a user selecting in a user interface a specific cell of the confusion matrix, the corresponding set of data to the specific cell is displayed on the user interface.

Explanatory confusion matrices for machine learning

A transitive closure data structure is constructed for a pair of features represented in a vector space corresponding to an input dataset. The data structure includes a set of entries corresponding to a set of all possible paths between a first feature in the pair and a second feature in the pair in a graph of the vector space. The data structure is reduced by removing a subset of the set of entries such that only a single entry corresponding to a single path remains in the transitive closure data structure. A feature cross is formed from a cluster of features remaining in a reduced ontology graph resulting from the reducing the transitive closure data structure. A layer is configured in a neural network to represent the feature cross, which causes the neural network to produce a prediction that is within a defined accuracy relative to the dataset.

Feature engineering in neural networks optimization

A computer-implemented method for managing updates to a knowledge graph with incoming change requests is disclosed. The method comprises creating a snapshot of the knowledge graph, building sub-regions of the knowledge graph, classifying each incoming change to the knowledge graph so that it relates to one of the sub-regions, creating at least one change layer, wherein each of the at least one change layer relates to one of the sub-regions, wherein each of the at least one change layer comprises a portion of the incoming change requests, closing the at least one change layer upon reaching a predetermined threshold value for the at least one change layer, and integrating the at least one change layer into the knowledge graph, thereby generating a updated knowledge graph.

Knowledge graph update using change layers

A method, system, and computer program product for visualizing a machine learning model are provided. A confusion matrix and model performance metric data are received from a classification model. For each data point in the confusion matrix, a corresponding pixel is generated. The pixels are grouped into clusters. Each cluster represents a label in the confusion matrix. A centroid is generated for each cluster. Using the model performance metric data, a misclassification indicator arrow is generated for each misclassified data point. The misclassification indicator arrow indicates both the predicted class and the actual class. The clusters, the centroids, and the misclassification indicator arrow are displayed as a graphical visualization of the machine learning model.

Visualizing machine learning model performance for non-technical users

Methods, computer program products, and systems are presented for service cost prediction using machine learning. A claims risk model is formed for predicting a per member per month cost based on variables trained in early prediction models and variables of claim record data from data sources. A training dataset for the claims risk model is modified based on a preconfigured cap value and two distinctive datasets are generated, which trains the claims risk model into a capped submodel and an outlier submodel, respectively. Each submodel makes prediction of the per member per month cost and a slice model interpolates a cost predicted by the capped submodel and another predicted by the outlier submodel. The splice model redistributes the outcomes of the capped submodel and the outlier submodel.

Skew-mitigated evolving prediction model

Gestural zoom operation of an imaging device is facilitated by receiving, by one or more processors, zoom configuration data of a user for configuring gestural zoom operation of the imaging device, where the imaging device includes a gestural zoom mode with a zoom function that is user-configurable, and by configuring for the user, by the one or more processors, the gestural zoom operation of the imaging device for the gestural zoom mode. The configuring uses the received zoom configuration data, and includes configuring the zoom function for the user and mapping a selected portion of the configured zoom function to a gestural range of the user of the imaging device.

User-configurable, gestural zoom facility for an imaging device

Technology for determining an amount of time to wait to retry requests to a representational state transfer (REST) server system for a REST resource, where the time to wait is always chosen to be a prime number of time units (for example, slots, milliseconds). While currently conventional systems will sometimes use a prime number of time units to wait for a retry request, various embodiments of the present invention will always, and invariably, use a prime number of time units. The REST resource may be, for example, a REST application programming interface (API) that is requested by and delivered to a client system using hypertext transfer protocol (HTTP).

Service retry collision prevention using prime numbers

An approach is provided in which the approach initiates a video conference between a first user utilizing a first device and a second user utilizing a second device. The first device uses a camera positioned at a first set of coordinates to capture a first live video feed of the first user from a first alignment perspective, and also displays a second live video feed of the second user on the screen at a second set of coordinates. The approach manipulates the first live video feed such that the manipulated first live video feed captures the first user from a second alignment perspective corresponding to the second set of coordinates, and transmits the manipulated first live video feed to the second device.

Eye contact across digital mediums

A method and system for enhancing a facial image of a user in real time by digital generation of a portion of a facial image using artificial intelligence (AI) during a video conference with a plurality of participants. The method and system including receiving a digital image of a first portion of a user's face in real time from a camera viewing the first portion of the user's face. The camera is unable to view the second portion of the user's face. The method and system includes improving resolution and/or digitally completing the second portion of the user's facial image, using an AI system. The improving resolution including, receiving the digital image at the AI system which includes a Generative Adversarial Network (GAN). The method and system includes generating, in real time, a complete enhanced digital facial image of the user's face using the GAN.

Enhancing digital facial image using artificial intelligence enabled digital facial image generation

Growing community of inventors

Austin, TX, United States of America

Stefan A G Van Der Stockt