Amazon Technologies, Inc.

University of Southern California

Anil Ramakrishna

Shrikanth Sambasivan Narayanan

Anoop Kumar

Krishna Somandepalli

Nikolaos Malandrakis

Karan Singla

Victor Martinez Palacios

Rahul Gupta

Premkumar Natarajan

Sankaranarayanan Ananthakrishnan

Sriram Venkatapathy

Aram Galstyan

Kshitij Mishra

Tamer Soliman

Yuval Merhav

Zefei Li

Heather Brooke Spetalnick

Techniques for updating a large language model (LLM) to correct generation of undesired responses, such as incorrect outputs, toxic outputs, etc. are described. Typical methods of retraining and fine-tuning are inefficient and computationally expensive for LLMs. Some embodiments of the present disclosure involve identifying a salient layer of the LLM that is responsible for the undesired response and editing only the salient layer. This layer is identified by computing a saliency value for the layer using a mean of gradient values for the layer, and the layer with the greatest saliency value is selected for editing. For editing, a small network is used to update the weights of the selected layer. The LLM is updated to include the edited layer, and the updated LLM is used for future processing.

Systems and methods for updating large language models

Techniques for updating a machine learning (ML) model are described. A device or system may receive input data corresponding to a natural or non-natural language (e.g., gesture) input. Using a first ML model, the device or system may determine the input data corresponds to a data category of a plurality of data categories. Based on the data category, the device or system may select a ML training type from among a plurality of ML training types. Using the input data, the device or system may perform the selected ML training type with respect to a runtime ML model to generate an updated ML model.

Machine learning model updating

A computer implemented method for analyzing media content includes a step of providing a plurality of narrative files formatted in human readable format. Each narrative file includes a script and/or dialogues tagged with character names along with auxiliary information. Each script includes a plurality of portrayals performed by an associated actor or character. Linguistic representations of content of the narrative files in both abstract and semantic forms is determined. The linguistic representations are connected to higher order representations and mental states. The linguistic representations are connected to behavior and action. Interplay between language constructs and demographics of content creators is analyzed. Content representations towards individuals/groups are adapted to reflect heterogeneity in preferences.

Linguistic analysis of differences in portrayal of movie characters

Techniques for using a federated learning framework to update machine learning models for spoken language understanding (SLU) system are described. The system determines which labeled data is needed to update the models based on the models generating an undesired response to an input. The system identifies users to solicit labeled data from, and sends a request to a user device to speak an input. The device generates labeled data using the spoken input, and updates the on-device models using the spoken input and the labeled data. The updated model data is provided to the system to enable the system to update the system-level (global) models.

Spoken language understanding models

Growing community of inventors

Los Angeles, CA, United States of America

Anil Ramakrishna