Microsoft Technology Licensing, LLC

Yi Mao

Qiang Wu

John Carlton Platt

Rahul Kapoor

Royi Ronen

Satish Ramchandra Thatte

Sumit Basu

Rolando Jimenez Salgado

Dengyong Zhou

Weizhu Chen

Erez Altus

Ricard Roma I Dalfó

Jiayuan Huang

Abedelkader Asi

Anuj Bansal

Yarin Kuper

Saji Varkey

Song Wang

Shimrit Rada Bemis

Olga Goldenberg

Syed Muhammad Mujahid Hassan

The disclosure herein describes reducing training bias in outputs generated by a generative language model. A communication segment associated with a communication is obtained by at least one processor of a generative language model. An output value associated with the communication segment is generated by the generative language model. The output value is mapped to a set of training bias values associated with the generative language model and based on the mapping of the output value to a training bias value of the set of training bias values, an alternative output value is generated. The alternative output value is used in a generated segment output for the communication segment. The accuracy of segment outputs generated by the generative language model is improved through reducing or eliminating its training biases.

Reducing biases of generative language models

A system and method infer true labels for multiple items. The inferred labels are generated from judgments. Multiple judges select the judgments from a specified choice of labels for each item. The method includes determining a characterization of judge expertise and item difficulties based on the judgments. The method also includes determining, using maximum entropy, a probability distribution over the specified choice of labels for each judge and item, based on the judgments. The method further includes selecting improved labels for the items from the specified choice such that the entropy over the probability distribution is reduced. The improved labels represent an improvement from the judgments toward the true labels. Additionally, the method includes performing iterative procedure to determine the true labels, the characterizations of judge expertise and the labeling difficulties.

Learning with noisy labels from multiple judges

An annotation system that provides annotations for a corresponding set of entities with regard to a user is presented. The annotation system annotates the entities independently of the selection of the entities to be annotated. Upon receiving a set of entities to annotate, the annotation system identifies a set of annotation relationships between the user and the entity. Each annotation relationship identifies a reason or basis for which the entity may be relevant to the user. The annotation relationships are scored to identify the strength of the relationship between the user and the entity. A configurable threshold number of annotation relationships with the greatest affinity between the user and the entity are selected as the annotation for the entity.

Annotating personalized recommendations

A single virtual image of client information centrally located at an always-on network location for maintaining equivalency among multiple user devices. The image can be accessed by the user devices when coming online to upload and receive changes in the client information. A mid-tier system can be employed as the always-on central location with which the user client machines can communicate to maintain the same set of client information. Services in support thereof include an ownership service for dynamic selection of a designated client machine to take ownership for performing the actions on one client machine and arbitration of duplicate requests, a notification service for allowing data sources to publish cache update instructions to a central place, a roaming service for allowing clients machines to share state with each other, and an encryption service for secure storage and communications of client information.

Single virtual client for multiple client access and equivalency

Growing community of inventors

Redmond, WA, United States of America

Yi Mao