Dropbox, Inc.

Other

Jongmin Baek

David J Kriegman

Thomas R Berg

Peter Neil Belhumeur

Nils Peter Welinder

Ying Xiong

Simon Kozlov

Jiarui Ding

Ermo Wei

Daniel R Horn

Scott McCrae

Ranjitha Gurunath Kulkarni

Xingyu Xiang

Sarah Andrabi

Anatoly Yakovenko

Daniel Wagner

Yuyang Guo

Austin Sung

Neeraj Kumar

Andraz Kavalar

Kimberly Watkins

David Lichtenberg

Shi Feng

Kristian Concepcion

Ivan Traus

Harrison Engel

Michael Loh

The present disclosure is directed toward systems and methods to quickly and accurately identify boundaries of a displayed document in a live camera image feed, and provide a document boundary indicator within the live camera image feed. For example, systems and methods described herein utilize different display document detection processes in parallel to generate and provide a document boundary indicator that accurately corresponds with a displayed document within a live camera image feed. Thus, a user of the mobile computing device can easily see whether the document identification system has correctly identified the displayed document within the camera viewfinder feed.

Live document detection in a captured video stream

The present disclosure relates to systems, methods, and non-transitory computer-readable media that utilize machine learning models to generate identifier embeddings from digital content identifiers and then leverage these identifier embeddings to determine digital connections between digital content items. In particular, the disclosed systems can utilize an embedding machine-learning model that comprises a character-level embedding machine-learning model and a word-level embedding machine-learning model. For example, the disclosed systems can combine a character embedding from the character-level embedding machine-learning model and a token embedding from the word-level embedding machine-learning model. The disclosed systems can determine digital connections between the plurality of digital content items by processing these identifier embeddings for a plurality of digital content items utilizing a content management model. Based on the digital connections, the disclosed systems can surface one or more digital content suggestions to a user interface of a client device.

Utilizing machine-learning models to generate identifier embeddings and determine digital connections between digital content items

The present disclosure is directed toward systems, methods, and non-transitory computer-readable media for performing a data search for a search request by utilizing a directed acyclic graph. For example, the disclosed systems receive a search request at a search engine of a content management system. In addition, the disclosed systems determine (e.g., in response to the search request) a node path from a directed acyclic graph that includes a plurality of interconnected nodes defining computer operations. Further, the determined node path includes a set of nodes that corresponds to the search request. Moreover, the disclosed systems perform the data search for the search request by executing operations defined by nodes along the node path within the directed acyclic graph.

Directed acyclic graph framework for executing search function operations

Computer-implemented techniques encompass using distinct machine learning sub-models to score respective types of candidate content for the purpose of providing personalized content suggestions to end-users of a content management system. The relevancy scores generated by the distinct sub-models are mapped to expected end-user interaction scores of the candidate content scored. Content suggestions are provided at end-users' computing devices where the suggested content is selected from the candidate content based on the expected end-user interaction scores of the candidate content. For each distinct sub-model, a normalizing mapping function is solved using an optimizer that maps the relevancy scores generated by the sub-model for the candidate content to expected end-user interaction scores for the candidate content. The expected end-user interaction scores are comparable across the distinct sub-models and can be used to rank content suggestions across the distinct sub-models.

Cross-model score normalization

Techniques for learning and using content type embeddings. The content type embeddings have the useful property that a distance in an embedding space between two content type embeddings corresponds to a semantic similarity between the two content types represented by the two content type embeddings. The closer the distance in the space, the more the two content types are semantically similar. The farther the distance in the space, the less the two content types are semantically similar. The learned content type embeddings can be used in a content suggestion system as machine learning features to improve content suggestions to end-users.

Content type embeddings

The present disclosure is directed toward systems and methods that efficiently and effectively generate an enhanced document image of a displayed document in an image frame captured from a live image feed. For example, systems and methods described herein apply a document enhancement process to a displayed document in an image frame that result in an enhanced document image that is cropped, rectified, un-shadowed, and with dark text against a mostly white background. Additionally, systems and method described herein determine whether a stored digital content item includes a displayed document. In response to determining that a stored digital content item does include a displayed document, systems and methods described herein generate an enhanced document image of a displayed document included in the stored digital content item.

Enhancing documents portrayed in digital images

The present disclosure is directed toward systems, methods, and non-transitory computer readable media for generating and suggesting content collections for user accounts of a content management system using combinations of content-based features such as textual signals and visual signals. In some embodiments, the disclosed systems select a seed content item from among a plurality of content items associated with a user account within a content management system. From the seed content item, the disclosed systems can determine one or more germane topics and can cluster additional content items in relation to the germane topic(s). In addition, the disclosed systems can select one or more content items from a content cluster to provide as a suggested content collection.

Seeding and generating suggested content collections

The present disclosure is directed toward systems, methods, and non-transitory computer readable media for generating and providing team-specific collections of collaborative content items. For example, the disclosed systems can determine relationships between teams of user accounts and various collaborative content items. Based on the determined relationships, the disclosed systems can identify which collaborative content items are germane to which teams and can provide the collaborative content items to user accounts accordingly. As part of determining relationships between collaborative content items and user accounts (or teams of user accounts), the disclosed systems can determine various information pertaining to the collaborative content items, including access patterns, sharing patterns, activity information, and geographic information.

Generating and providing collections of collaborative content items to teams of user accounts

The present disclosure relates to systems, methods, and non-transitory computer-readable media that can leverage a natural language model to determine a most probable candidate sequence of tokens and thereby generate a predicted user activity. In particular, the disclosed systems can tokenize activity event vectors to generate a series of sequential tokens that correspond to recent user activity of one or more user accounts. In addition, the disclosed systems can, for each candidate (e.g., hypothetical) user activity, augment the series of sequential tokens to include a corresponding token. Based on respective probability scores for each of the augmented series of sequential tokens, the disclosed systems can identify as the predicted user activity, a candidate user activity corresponding to one of the augmented series of sequential tokens associated with a highest probability score. Based on the predicted user activity, the disclosed systems can surface one or more suggestions to a client device.

Utilizing a natural language model to determine a predicted activity event based on a series of sequential tokens

Growing community of inventors

San Jose, CA, United States of America

Jongmin Baek