International Business Machines Corporation

Claire Abu-Assal

Ramani Ranjan Routray

Abhinandan Kelgere Ramesh

Sheng Hua Bao

Mu Qiao

Cartic Ramakrishnan

Bruce Light Hillsberg

Pathirage Dinindu Sujan Udayanga Perera

A mechanism is provided in a data processing system for training a computer implemented model. The mechanism determines an operation for which the computer implemented model is to be trained. The mechanism performs a statistical analysis of an enterprise dataset for an enterprise to generate one or more statistical distributions of cases and features correlated with the operation for which the computer implemented model is to be trained. The mechanism selects a subset of cases in the enterprise dataset for annotation based on the one or more statistical distributions of cases and features. The mechanism annotates the selected subset of cases to generate an annotated training dataset. The mechanism trains the computer implemented model, using the annotated training dataset, to perform the operation.

Cognitive data preparation for deep learning model training

In training a new neural network, batches of the new training dataset are generated. An epoch of batches is passed through the new neural network using an initial weight (&#x3b8;). An area minimized (A) under an error function curve and an accuracy for the epoch are calculated. It is then determined whether a set of conditions are met, where the set of conditions includes whether Ais less than an average area (A_avg) from a training of an existing neural network and whether the accuracy is within a predetermined threshold. When the set of conditions are not met, a new &#x3b8; is calculated by modifying a dynamic learning rate (&#x3b2;) by an amount proportional to a ratio of A/A_avg and by calculating the new &#x3b8; using the modified &#x3b2; according to &#x3b8;:&#xb1;&#x3b8;&#x2212; The process is repeated a next epoch until the set of conditions are met.

Optimization of model generation in deep learning neural networks using smarter gradient descent calibration

Growing community of inventors

Pismo Beach, CA, United States of America

Claire Abu-Assal