Arista Networks, Inc.

Amazon Technologies, Inc.

Uber Technologies, Inc.

Verizon Media Inc.

Oath Inc.

Vijay Mahadevan

Rahul Bhotika

Sachin Sudhakar Farfade

Ayyappan Arasu

Venkat Kumar Reddy Barakam

Ayman Kaheel

Adam James Sweeney

Rajesh Kumar Semwal

Stefano Soatto

Victor Wen

Kunwar Yashraj Singh

Jay Kiran Mahadeokar

Dileep Honsoge Ramesh

Mohammed Waleed Kadous

Meng Wang

Yusheng Xie

Amit Adam

Vasudev Parameswaran

Shai Mazor

Jesper Skriver

Prashant Srinivas

Jonathan Wu

R Manmatha

Aditya Sarawgi

Vishal Bandekar

Ramakrishnan Ganapathy Iyer

Hadar Averbuch Elor

Shahar Tsiper

Joaquin Zepeda Salvatierra

Timo Pekka Pylvaenaeinen

Gal Sabina Star

Mohammed El Hamalawi

Roee Litman

Pegah Massoudifar

Jonathan Berliner

Jan Kiran Mahadeokar

Erhan Bas

Max Xiao

Ryan Megathlin

Craig Lauer

Christopher Roche

Pengkai Zhu

Peng Tang

Bhavan Ashwinbhai Jasani

Guneet Singh Dhillon

Yuting Zhang

Gangadhar Payyavula

Viet Cuong Nguyen

Document type specific adapters of a document analysis system are used to provide for additional document types or document specialization when generating answers to user submitted questions targeting information included in a document image provided with the user submitted question. The document analysis system receives a visual question answering (VQA) prompt comprising a document image and a question defining information to be extracted from the document image, generates tokens based on the document image and the question, and adjusts encoding of the tokens using document type specific adapters of a transformer model to extract the information from the document image. A classifier of the document analysis system may determine whether the document image matches a document type supported by the document type specific adapters.

Document information extraction using visual question answering and document type specific adapters

Methods, computer-readable media, and systems are disclosed for coordinating host link status and installation of egress filters in coordination with a multi-chassis link aggregation group (MLAG) peer. In response to receiving a request from an MLAG peer to install an egress filter, a local network device provides an indication that the local device has entered a local egress filter installed state. Next, a filter-installed watermark value associated with the local network device is incremented. The local network device provides an indication that the egress filter has been installed. Finally, in response to determining that the MLAG peer network device is not indicating a request peer to install filter status, a local network device indicates a link down status and uninstalls its local egress filter. Alternatively, in response to determining that the local network device is in a link up status, the local device indicates that an MLAG peer should install an egress filter. In response to determining that the peer network device is in a filter request status, installing a local egress filter.

Coordinating host link status and associated egress filter

Systems and methods are provided herein for allocating the same ESI label on multihomed peers for a given ES. In some embodiments, each network device that provides multihoming to a host using an ES, advertises EVPN AD per ES routes to each other, wherein the EVPN AD per ES routes comprise an ESI label associated with the ES. Because the network devices advertise the same ESI label for the ES, a first network device generates a bitmap. The first network device uses the bitmap to include the advertised ESI label in replicated packets that the first network device forwards to the other network devices that provide multihoming to the host via the ES. The network devices that consider themselves non-DF devices will drop the packet. The network devices that consider themselves the DF device will not forward the packet to the host via the ES because of the ESI label.

Shared ethernet segment identifier label allocation for ethernet virtual private network multihoming

Techniques for configuring a logical network switch in label-switched networks are provided. In some embodiments, a first network device in a label-switched network is configured with a network address. A second network device in the label-switched network is configured with the same network address. The first network device is configured to use a set of labels for a set of virtual local area networks (VLANs). The second network device is configured to use the same set of labels for the same set of VLANs. The configured first and second network devices appear as a logical network device from the perspective of other network devices in the label-switched network.

Configuring logical network devices for label-switched networks

Techniques for creating machine learning models for direct homography regression for image rectification are described. In certain embodiments, a training service trains an algorithm on a source view of a training image and a homography matrix of the training image into a machine learning model that generates a normalized homography matrix for an input of the source view. The normalized homography matrix may then be utilized to generate a target view of an image input into the machine learning model. The target view of the image may be used in a document processing pipeline for document images captured using cameras.

Machine learning models for direct homography regression for image rectification

Embodiments described herein relate to techniques for designated forwarder (DF) elections, which may include: obtaining DF candidates that are part of a supplementary broadcast domain (SBD), wherein the DF candidate is one of the plurality of DF candidates for the SBD; performing a SBD DF election process to determine an SBD DF winner from among the DF candidates; making a first determination that the DF candidate is not the SBD DF winner; making second determination that a first broadcast domain (BD) provisioned on the DF candidate is not provisioned on the SBD DF winner; excluding the first BD from a set of BDs that are also provisioned on the SBD DF winner; performing additional DF election processes for each BD of the set of BDs; and processing multicast traffic based at least in part on the SBD DF election process and the additional DF election processes.

Modified designated forwarder election process for non-overlapping broadcast domains

Techniques for document rectification via homography recovery using machine learning are described. An image rectification system can intelligently make use of multiple pipelines for rectifying document images based on the detected type of device that generated the images. The image rectification system can provide high-quality rectifications without requiring human cooperation, multiple views of the document in multiple images, and/or without being constrained to only be able to process images from one source context.

Document rectification via homography recovery using machine learning

A method and apparatus of a device that starts an address resolution service on a network element after a boot-up of this network element is described. In an exemplary embodiment, the network element sends an indication of the boot-up. The network element further sends a request for an address resolution table and receives a reply with the requested address resolution table. In addition, the network element starts the address resolution service using the requested address resolution table.

System and method of persistent address resolution synchronization

Disclosed are systems and methods for improving interactions with and between computers in content searching, generating, hosting and/or providing systems supported by or configured with personal computing devices, servers and/or platforms. The systems interact to identify and retrieve data within or across platforms, which can be used to improve the quality of data used in processing interactions between or among processors in such systems. The disclosure provides a novel, computerized framework for automatically selecting the most definitive, precise and high-quality content files corresponding to POIs. The disclosed systems and methods utilize the performance of visual comparisons with a set of definitive content files of a given POI, and by incorporating visual aesthetic features as a factor of such comparisons, a search result is identified that down-weights imprecise and poor quality content files of a given POI, and ensures that only high quality, accurate content files are selected or identified.

Computerized system and method for automatically identifying and providing digital content based on physical geographic location data

A system trains and applies a machine learning model to label maps of a region. Various data modalities are combined as inputs for multiple data tiles used to characterize a region for a geographical map. Each data modality reflects sensor data captured in different ways. Some data modalities include aerial imagery, point cloud data, and location trace data. The different data modalities are captured independently and then aggregated using machine learning models to determine map labeling information about tiles in the region. Data is ingested by the system and corresponding tiles are identified. A tile is represented by a feature vector of different data types related to the various data modalities, and values from the ingested data are added to the feature vector for the tile. Models can be trained to predict characteristics of a region using these various types of input.

Determining geographical map features with multi-sensor input

Disclosed are systems and methods for automatic selection of canonical digital images from a large corpus of digital images, such as the corpus of digital images available on the web, for an entity, such as and without limitation a person, a point of interest, object, etc. The automated, unsupervised approach for selecting a diverse set of high quality, canonical digital images, is well suited for processing a large corpus of digital images. A set of canonical digital images identified for an entity can be retrieved in response to a digital image request for digital images depicting the entity.

Automatic canonical digital image selection method and apparatus

A structured document analyzer that associates keys and values in structured documents based on key, value, and key-value container bounding boxes. A trained machine learning model analyzes images of structured documents to determine bounding boxes for keys, values, and key-value containers in the images with confidence scores for the classifications. For each image, duplicate bounding boxes are removed, and then a set of key-value containers are selected and sorted based on the confidence scores. For each key-value container, a best key and value are determined for the container based on overlap of the key and value bounding boxes with the container bounding box and the confidence scores. Optical character recognition may be performed on the image to determine text for the keys and values.

Structured document analyzer

Techniques for multiple object tracking in video are described in which the outputs of neural networks are combined within a Bayesian framework. A motion model is applied to a probability distribution representing the estimated current state of a target object being tracked to predict the state of the target object in the next frame. A state of an object can include one or more features, such as the location of the object in the frame, a velocity and/or acceleration of the object across frames, a classification of the object, etc. The prediction of the state of the target object in the next frame is adjusted by a score based on the combined outputs of neural networks that process the next frame.

Multiple object tracking in video by combining neural networks within a bayesian framework

A media item comprising a set of frames is received by a feature extraction system. A frame predictor is executed on each frame of the set of frames. An error representation is extracted for each frame of the set of frames during the execution of the frame predictor. An error-based feature vector is generated from the error representations associated with each frame of the set of frames. A seed media item is identified having a first error-based feature vector. A similarity score is determined among the first error-based feature vector and each error-based feature vector of a set of error-based feature vectors. A subset of error-based feature vectors, hence a subset of corresponding media items, is selected based on similarity score.

Characterizing content with a predictive error representation

Growing community of inventors

San Francisco, CA, United States of America

Vijay Mahadevan