Amazon Technologies, Inc.

Vasant Manohar

Vivek Bhadauria

Andrea Olgiati

Jonathan Hedley

Viswanath Sankaranarayanan

Gowtham Jeyabalan

Nitin Singhal

Stefano Stefani

Joseph Spencer King

Gaurav D Ghare

Roman Goldenberg

Dylan Chandler Thomas

Eric Allen Menninga

Wei Xia

Joseph Patrick Tighe

Ranju Das

Rakesh Madhavan Nambiar

Anand Dhandhania

Shubham Chandra Gupta

Mugunthan Govindaraju

Davide Modolo

Kunwar Yashraj Singh

Stephen Gould

Praveen Kumar Udayakumar

Yuri Natanzon

Ashley Alonzo Ricardo Karl Mitchell

Sridhar Godavarthy

Keith Young Johnson

Praveenkumar Udayakumar

Sean R Flynn

Binglei Du

Palak Mehta

Kuang Han

Renjun Zheng

Janarthanan Lakshmipathy

A network-based service may provide a machine learning model for different clients. The network-based service may implement an interface that allows a client to identify a test data set for validating versions of the machine learning model specifically for the client. When a new version of the machine learning model is created, a validation test using the test data set identified by the client may be used. Results of the validation test may be used to make a decision regard whether to migrate workloads for the client to the new version of the machine learning model.

Client-specific validation for migrating to a version of a machine learning model using client-supplied test data

Objects detected in data may be filtered from an object recognition index. Data for object detection may be received. An object detection technique may be applied to the data to detect an object. If the object does not satisfy indexing criteria for the object recognition index, then the detected object may be excluded from the object recognition index.

Filtering detected objects from an object recognition index according to extracted features

Technologies are disclosed for efficiently identifying objects in videos using deep neural networks and motion information. Using the disclosed technologies, the amount of time required to identify objects in videos can be greatly reduced. Motion information for a video, such as motion vectors, are extracted during the encoding or decoding of the video. The motion information is used to determine whether there is sufficient motion between frames of the video to warrant performing object detection on the frames. If there is insufficient movement from one frame to a subsequent frame, the subsequent frame will not be processed to identify objects contained therein. In this way, object detection will not be performed on video frames that have changed minimally as compared to a previous frame, thereby reducing the amount of time and the number of processing operations required to identify the objects in the video.

Efficient identification of objects in videos using motion information

Techniques are described for automatically scaling (or 'auto scaling') compute resources&#x2014;for example, virtual machine (VM) instances, containers, or standalone servers&#x2014;used to support execution of service-oriented software applications and other types of applications that may process heterogeneous workloads. The resource requirements for a software application can be approximated by measuring 'worker pool' utilization of instances of each service, where a worker pool represents a number of requests that the service can process concurrently. A scaling service can thus be configured to scale the compute instances provisioned for a service in proportion to worker pool utilization, that is, compute instances can be added as the fleet's worker pools become more &#x201c;busy,&#x201d; while compute instances can be removed when worker pools become inactive.

Automatically scaling compute resources for heterogeneous workloads

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

Devices, systems and methods are disclosed for validating an electronic publication and determining a source of identified errors in a rendering of the electronic publication. The rendering may be captured as a rendered image and rendered data may be extracted from the rendering. The rendered data may be compared to actual input data to the renderer used to generate the rendered image. If errors are visible in the rendering, a source of the errors may be identified based on the comparison between the extracted rendered data to the actual input data. If errors are not visible in the rendering, the rendering may be validated.

Validating digital content rendering

Disclosed are techniques and systems to detect a layout of a source document. A process may include receiving content from a first page and a second page of the source document, designating sections in each page along a first direction of the page, and assigning similar sections to a group. For the group, the process may proceed by dividing sections for each page into discrete portions associated with 2D coordinate areas, and identifying sets of 2D coordinate areas for the discrete portions that contain content. The number of times each portion contains some content may be compared to a threshold to determine a layout of the group of sections.

Detection of layouts in electronic documents

A method of generating a reflowable content file from a physical text source is described. An image of the physical text source is segmented into a plurality of glyphs and a character and font is determined for each of the glyphs. The font for each of the plurality of glyphs is determined based on two or more of the glyphs.

Digital conversion of imaged content

A platform for generating a first character recognition-based work including a first plurality of automatically-made edits, each edit being characterized by a Unicode and a confidence score. The platform may identify at least one edit as being of questionable accuracy based on the confidence score, may determine a unique character signature of the edit, and may receive a manual correction made to the edit. The platform may also store the manual correction in association with the character signature and the Unicode, such that the manual correction is configured for use in generating a second plurality of automatically-made edits in a second character recognition-based work different than the first work.

Cohort-based learning from user edits

Some implementations include using a trained classifier to identify page-turn events in a video. The video may be divided into multiple segments based on the page-turn events, with each segment of the multiple segments corresponding to a pair of adjacent pages in a book. Exemplar frames that provide non-redundant data compared to other frames may be chosen from each segment. The exemplar frames may be cropped to include content portions of pages. The exemplar frames may be aligned such that a pixel is located in a same position in each frame. Optical character recognition (OCR) may be performed on exemplar frames and the OCR for exemplar frames in each segment may be combined. The exemplar frames in each segment may be combined to create a composite image for each pair of adjacent pages in the book, and OCR may be performed on the composite image.

Creating an electronic book using video-based input

Growing community of inventors

Tamil Nadu, India

Vasant Manohar