Adobe, Inc.

Tung Mai

Anup Bandigadi Rao

Ryan A Rossi

Matvey Kapilevich

Viswanathan Swaminathan

Shivakumar Vaithyanathan

Georgios Theocharous

Eunyee Koh

Saayan Mitra

Nedim Lipka

Michele Saad

Somdeb Sarkhel

Charles Menguy

Margarita Savova

Moumita Sinha

Thomas Greger

Kirankumar Shiragur

Antonio Cuevas

Kelly Quinn Nicholes

Nathan Ng

William Brandon George

Wei Zhang

Fan Du

Sungchul Kim

Gang Wu

Ritwik Sinha

Iftikhar Ahamath Burhanuddin

Subrata Mitra

Nikhil Sheoran

Tong Yu

Shunan Guo

Yasin Abbasi Yadkori

Said Kobeissi

Vijeth Lomada

Sapthotharan Krishnan Nair

Yeuk-Yin Chan

Zhao Song

Jordan Walker

Harleen S Sahni

Lakshmi Shivalingaiah

Prithvi Bhutani

Alexandru Ionut Hodorogea

Thomas W Jacobs

Ghetia Siddharth

Jatin Varshney

Vikas Maddukuri

Laxmikant Mishra

Yeshwanth Vijayakumar

Aravind Reddy Talla

Enayat Ullah

Jiong Zhu

Liron Goren Snai

Tyler Rasmussen

Samuel Nephi Grigg

Abhisek Trivedi

Alireza Farhadi

Systems and methods for training a neural network are described. One or more embodiments of the present disclosure include training a neural network based on a first combined gradient of a loss function at a plurality of sampled elements of a dataset; receiving an insertion request that indicates an insertion element to be added to the dataset, or a deletion request that indicates a deletion element to be removed from the dataset, wherein the deletion element is one of the plurality of sampled elements; computing a second combined gradient of the loss function by adding the insertion element to the dataset or by replacing the deletion element with a replacement element from the dataset; determining whether the first combined gradient and the second combined gradient satisfy a stochastic condition; and retraining the neural network to obtain a modified neural network based on the determination.

Machine unlearning and retraining of a machine learning model based on a modified training dataset

Systems and methods for dynamic user profile management are provided. One aspect of the systems and methods includes receiving, by a lookup component, a request for a user profile; computing, by a profile component, a time-to-live (TTL) refresh value for the user profile based on a lookup history of the user profile; updating, by the profile component, a TTL value of the user profile based on the request and the TTL refresh value; storing, by the profile component, the user profile and the updated TTL value in the edge database; and removing, by the edge database, the user profile from the edge database based on the updated TTL value.

Dynamic user profile management

Embodiments provide systems, methods, and computer storage media for a Nonsymmetric Determinantal Point Process (NDPPs) for compatible set recommendations in a setting where data representing entities (e.g., items) arrives in a stream. A stream representing compatible sets of entities is received and used to update a latent representation of the entities and a compatibility distribution indicating likelihood of compatibility of subsets of the entities. The probability distribution is accessed in a single sequential pass to predict a compatible complete set of entities that completes an incomplete set of entities. The predicted complete compatible set is provided a recommendation for entities that complete the incomplete set of entities.

Online inference and learning for nonsymmetric determinantal point processes

Graphic visualizations, such as charts or graphs conveying data attribute values, can be generated based on natural language queries, i.e., natural language requests. To do so, a natural language request is parsed into n-grams, and from the n-grams, word embeddings are determined using a natural language model. Data attributes for the graphic visualization are discovered in the vector space from the word embeddings. The type of graphic visualization can be determined based on a request intent, which is determined using a trained intent classifier. The graphic visualization is generated to include the data attribute values of the discovered data attributes, and in accordance with the graphic visualization type.

Learning-based method for generating visualizations via natural language

Techniques are provided for training graph neural networks with heterophily datasets and generating predictions for such datasets with heterophily. A computing device receives a dataset including a graph data structure and processes the dataset using a graph neural network. The graph neural network defines prior belief vectors respectively corresponding to nodes of the graph data structure, executes a compatibility-guided propagation from the set of prior belief vectors and using a compatibility matrix. The graph neural network predicts predicting a class label for a node of the graph data structure based on the compatibility-guided propagations and a characteristic of at least one node within a neighborhood of the node. The computing device outputs the graph data structure where it is usable by a software tool for modifying an operation of a computing environment.

Graph neural networks for datasets with heterophily

Systems and methods for dynamic user profile projection are provided. One or more aspects of the systems and methods includes computing, by a prediction component, a predicted number of lookups for a future time period based on a lookup history of a user profile using a lookup prediction model; comparing, by the prediction component, the predicted number of lookups to a lookup threshold; and transmitting, by a projection component, the user profile to an edge server based on the comparison.

Dynamic user profile projection

Techniques are provided for analyzing user actions that have occurred over a time period. The user actions can be, for example, with respect to the user's navigation of content or interaction with an application. Such user data is provided in an action string, which is converted into a highly searchable format. As such, the presence and frequency of particular user actions and patterns of user actions within an action string of a particular user, as well as among multiple action strings of multiple users, are determinable. Subsequences of one or more action strings are identified and both the number of action strings that include a particular subsequence and the frequency that a particular subsequence is present in a given action string are determinable. The conversion involves breaking that string into a sorted list of locations for the actions within that string. Queries can be readily applied against the sorted list.

Determining patterns within a string sequence of user actions

Multi-modal machine-learning model training techniques for search are described that overcome conventional challenges and inefficiencies to support real time output, which is not possible in conventional training techniques. In one example, a search system is configured to support multi-modal machine-learning model training. This includes use of a preview mode and an expanded mode. In the preview mode, a preview segment is generated as part of real time training of a machine learning model. In the expanded mode, the preview segment is persisted as an expanded segment that is used to train and utilize an expanded machine-learning model as part of search.

Multi-modal machine-learning model training for search

A cluster generation system identifies data elements, from a first binary record, that each have a particular value and correspond to respective binary traits. A candidate description function describing the binary traits is generated, the candidate description function including a model factor that describes the data elements. Responsive to determining that a second record has additional data elements having the particular value and corresponding to the respective binary traits, the candidate description function is modified to indicate that the model factor describes the additional elements. The candidate description function is also modified to include a correction factor describing an additional binary trait excluded from the respective binary traits. Based on the modified candidate description function, the cluster generation system generates a data summary cluster, which includes a compact representation of the binary traits of the data elements and additional data elements.

Trait expansion techniques in binary matrix datasets

In implementations of item transfer control systems, a computing device implements a transfer system to receive input data describing types of requested items and corresponding quantities of the types of requested items to receive at each of a plurality of destination sites and types of available items and corresponding quantities of the types of available items that are available at each of a plurality of source sites. The transfer system constructs a flow network having a source node for each of the plurality of the source sites and a destination node for each of the plurality of the destination sites. An integral approximate solution is generated that transfers the corresponding quantities of the types of requested items to each of the plurality of the destination sites using a maximum flow solver and the flow network. The transfer system causes transferences of the corresponding quantities of the types of requested items to each of the plurality of the destination sites based on the integral approximate solution.

Item transfer control systems

The present disclosure relates to systems, methods, and non-transitory computer-readable media that estimate the overlap between sets of data samples. In particular, in one or more embodiments, the disclosed systems utilize a sketch-based sampling routine and a flexible, accurate estimator to determine the overlap (e.g., the intersection) between sets of data samples. For example, in some implementations, the disclosed systems generate a sketch vector&#x2014;such as a one permutation hashing vector&#x2014;for each set of data samples. The disclosed systems further compare the sketch vectors to determine an equal bin similarity estimator, a lesser bin similarity estimator, and a greater bin similarity estimator. The disclosed systems utilize one or more of the determined similarity estimators in generating an overlap estimation for the sets of data samples.

Generating overlap estimations between high-volume digital data sets based on multiple sketch vector similarity estimators

The present disclosure relates to method for responding to a query requesting an intersection being performed. The method includes receiving a query referencing a first set, a second set, and a desired quantile related to the first set from among a plurality of quantiles; generating a data structure including a bottom-k sketch of user identifiers (ids) of the first set and corresponding numerical values of the first data; partitioning the data structure into a plurality of sketches to correspond to the quantiles, respectively; determining an intersection of one of the sketches associated with the desired quantile and a sketch of the second set; and responding to the query based on the intersection.

Efficient set operation execution on streaming data using sketches

In some embodiments, a model training system trains a sample generation model configured to generate synthetic data entries for a dataset. The sample generation model includes a prior model for generating an estimated latent vector from a partially observed data entry, a proposal model for generating a latent vector from a data entry of the dataset and a mask corresponding to the partially observed data entry, and a generative model for generating the synthetic data entries from the latent vector and the partially observed data entry. The model training system trains the sample generation model to optimize an objective function that includes a first term determined using the synthetic data entries and a second term determined using the estimated latent vector and the latent vector. The trained sample generation model can be executed on a client computing device to service queries using the generated synthetic data entries.

Query-oriented approximate query processing based on machine learning techniques

Embodiments of the present invention provide systems, methods, and computer storage media for determining an increased matching for large graphs in which an increased matching is generated for the graph by leveraging an initial matching for a small fraction of edges of the large graph. An initial matching for a random subset of edges of an input graph is leveraged to generate alternating paths based on the initially matched edges and the remaining edges, not included in the random subset. An increased matching for the entire graph includes the alternating paths without the initial matched edges, thus increasing the number of matched edges in the increased matching by at least one for every initially matched edge. Graph-based tasks may then be triggered based on the increased matching.

Single-pass matching in large data streams

Growing community of inventors

San Jose, CA, United States of America

Tung Mai