Amazon Technologies, Inc.

Unitedhealth Group Incorporated

Ranju Das

Meng Wang

Wei Xia

Weixin Wu

Rahul Bhotika

Barath Balasubramanian

Prakash Krishnan

Joaquin Zepeda Salvatierra

Niels Brouwers

Anant Patel

Shaun Ryan James McDowell

Avinash Aghoram Ravichandran

Gurumurthy Swaminathan

Rakesh Madhavan Nambiar

Jonathan Hedley

Anushri Mainthia

Stefano Stefani

Andrea Olgiati

Gaurav D Ghare

Hao Chen

Xiaofeng Ren

Roman Goldenberg

Joseph Patrick Tighe

Jonathan Wu

Vasant Manohar

Vivek Bhadauria

Stephen Gould

Nitin Singhal

Gowtham Jeyabalan

Praveen Kumar Udayakumar

Subhasis Jethy

Venkatesh Bagaria

Carlos Morato

Sanjay Kumar Singh

Udit Saini

Kuang Han

Renjun Zheng

Shaun Ryan James Mcdowell

An example system for parsing and transforming input data that includes processing circuitry and memory, the memory configured to store the input data. The processing circuitry is configured to determine a first delimiter in the input data. The processing circuitry is configured to determine a plurality of second delimiter hypotheses and parse the input data according to the first delimiter and the plurality of second delimiter hypotheses to generate a plurality of tables that are each associated with a respective one of the plurality of second delimiter hypotheses. The processing circuitry is configured to determine a respective consistency score for each of the plurality of tables and select a table from among the plurality of tables based on the respective consistency score associated with the table. The processing circuitry is configured to format the input data based on the selected table to generate formatted data and output the formatted data.

Hierarchical delimiter identification for parsing of raw data

A visualization tool for machine learning models obtains metadata from a first training node at which a multi-layer machine learning model is being trained. The metadata includes a parameter of an internal layer of the model. The tool determines a plurality of metrics from the metadata, including respective loss function values corresponding to several training iterations of the model. The tool indicates the loss function values and the internal layer parameter values via a graphical interface.

Real-time visualization of machine learning models

Techniques for feedback-based training may include selecting a scoring machine learning model based at least in part on a test metric, and applying the model on an unlabeled dataset to generate, per dataset item of the unlabeled dataset, a prediction and an importance ranking score for the prediction. Techniques for feedback-based training may further include selecting, based on the importance ranking scores, a result of the application of the model on the unlabeled dataset, providing the result and requesting feedback on the result via a graphical user interface, receiving the feedback via the graphical user interface, adding data from the unlabeled dataset into a training dataset when the feedback indicates a verified result, and retraining the model using the training dataset with the data added from the unlabeled dataset to generate a retrained model.

Feedback-based training for anomaly detection

Techniques for anomaly detection are described. An exemplary method includes receiving one or more requests to train an anomaly detection machine learning model using feedback-based training, the request to indicate one or more of a type of analysis to perform, a model selection indication, and a configuration for a training dataset; training the anomaly detection machine learning model according to the one or more requests using the training data; performing feedback-based training on the trained anomaly detection machine learning model; and using the retrained anomaly detection machine learning model.

Anomaly detection using feedback training

Techniques for anomaly detection are described. An exemplary method includes receiving a request to create a training data set from at least one image, the request to include an indication of the at least one image and at least one indication of an operation to perform on the at least one image to generate a plurality of images from the at least one image; creating a training dataset by extracting one or more chunks from a first at least one image according to the request; and receiving one or more requests to train an anomaly detection machine learning model using the created training dataset; and training an anomaly detection machine learning model according to one or more requests using the created training data.

Generation of training data from a single image

People represented in multiple images can be recognized using accurate facial similarity metrics, where the accuracy can be further improved using contextual information. A set of models can be trained to process image data, and facial features can be extracted from a face region of an image and passed to the trained models. Resulting feature vectors can be concatenated and the dimensionality reduced to generate a highly accurate feature vector that is representative of the face in the image. The feature vector can be used to locate similar vectors in a multi-dimensional vector space, where similarity can be determined based at least in part upon the distance between the endpoints of those vectors in the vector space. Context information from the image can be used to adjust the similarity determination. Similar vectors can be clustered together such that the faces represented by those images are associated with the same person.

Context-inclusive face clustering

The present disclosure provides for customizable content moderation using neural networks with fine-grained and dynamic image classification ontology. A content moderation system of the present disclosure may provide a plurality of image categories in which a subset of of image categories may be designated as restricted categories. The restricted categories may be chosen by a content provider or an end-user. The content moderation system may utilize a neural network to classify image data (e.g., still images, video, etc.) into one or more of the plurality of image categories, and determine that an image is a restricted image upon classifying the image into one of the restricted categories. The restricted image may by flagged, rejected, removed, or otherwise filtered upon being classified as a restricted image.

Fine-grain content moderation to restrict images

Techniques for analyzing stored video upon a request are described. For example, a method of receiving a first application programming interface (API) request to analyze a stored video, the API request to include a location of the stored video and at least one analysis action to perform on the stored video; accessing the location of the stored video to retrieve the stored video; segmenting the accessed video into chunks; processing each chunk with a chunk processor to perform the at least one analysis action, each chunk processor to utilize at least one machine learning model in performing the at least one analysis action; joining the results of the processing of each chunk to generate a final result; storing the final result; and providing the final result to a requestor in response to a second API request is described.

Analysis of video content

A knowledge graph (KG) is generated and refined. The generated KG describes direct relationships between different words associated with a particular classification. Initially, a semantic data source, such as a lexical database, is accessed to identify words that are similarly grouped and express a distinct concept. A KG generator creates a sparse KG that provides a direct connection between a seed word and other words. The sparse KG is used by a dense KG generator to create a dense KG. The dense KG generator creates a dense KG that joins each of the different words directly with the seed word for the category. At different points during the creation and refinement of the KG, a user may add or remove one or more connections that affect the creation of the KG.

Generating and refining a knowledge graph

A method and system for analyzing text in an image. Classification and localization information is identified for the image at a word and character level. A detailed profile is generated that includes attributes of the words and characters identified in the image. One or more objects representing a predicted source of the text are identified in the image. In one embodiment, neural networks are employed to determine localization information and classification information associated with the identified object of interest (e.g., a text string, a character, or a text source).

Image text recognition

A knowledge base (KB) is generated and used to classify images. The knowledge base includes a number subcategories of a specified category. Instead of obtaining images just based on a category name, structured and unstructured data sources are used to identify subcategories of the category. Subcategories that are determined to not be relevant to the category may be removed. The remaining data may be used to generate the KB. After identifying the relevant subcategories, representative images are obtained from one or more image sources based on the subcategories identified by the KB. The obtained images and the KB are then used to train an image classifier, such as a neural network or some other machine learning mechanism. After training, the neural network might, for example, classify an object within an image of a car, as a car, but also as a particular brand and model type.

Generating and using a knowledge base for image classification

Growing community of inventors

Seattle, WA, United States of America

Ranju Das