Walmart Apollo, LLC

Venkatesh Kandaswamy

Alessandro Magnani

Abon Chaudhuri

Shreyansh Prakash Gandhi

Abilash Amarthaluri

Behzad Ahmadi

Omer Ovenc

Abhinandan Krishnan

Samrat Kokkula

Theban Stanley

Karthik Deivasigamani

Ashish Kumar Gupta

Zuzar Fakhruddin Nafar

Rohit Deep

Anthony Soohoo

Qian Li

Ganesh Krishnan

Vijay Raghavendra

Charandeep Singh Sehgal

Suleman Ashfaq

Sreyash D Kenkre

Charles Myslinsky

Vikas Bhat

Harman Kochar

Arvind Senthil Kumaran

Arvind Chockalingam Senthil Kumaran

A system including one or more processors and one or more non-transitory computer-readable storage media storing computing instructions that, when executed on the one or more processors, cause the one or more processors to perform: training a neural network detection model with a training dataset comprising synthetic training images by: using a transformation algorithm to create the synthetic training images by appending edge case training images to one or more compliant images; receiving, at the neural network detection model, as trained, at least one image; and determining, using the neural network detection model, as trained, whether the at least one image comprises non-compliant content. Other embodiments are disclosed herein.

Systems, methods, and techniques for training neural networks and utilizing the neural networks to detect non-compliant content

Systems and methods including one or more processing modules and one or more non-transitory storage modules storing computing instructions configured to run on the one or more processing modules and perform acts of receiving attribute data comprising a set of unstructured attribute data and a set of structured attribute data, analyzing the set of unstructured attribute data by processing through a first set of one or more Long Short Term Memory (LSTM) layers, to obtain an unstructured semantic signature, analyzing the set of the structured attribute data by processing through a first set of one or more Convolutional Neural Network (CNN) layers, to obtain a structured semantic signature, analyzing the unstructured semantic signature and the structured semantic signature, and classifying the item in one or more item categories. Other embodiments are disclosed herein.

Systems and methods for classification using structured and unstructured attributes

A system can include one or more processors and one or more non-transitory computer-readable storage media storing computing instructions configured to run on the one or more processors and perform: generating a training dataset for training a neural network detection model; identifying, using the neural network detection model, as trained, the non-compliant content in the synthetic training images; receiving, at the neural network detection model, at least one image; and utilizing the neural network detection model to determine whether the at least one image comprises the non-compliant content. Other embodiments are disclosed herein.

A system can include one or more processors and non-transitory computer-readable media storing computing instructions configured to run on the one or more processors and perform a method for determining one or more substitutes for a product requested in an inquiry from a user. The method comprises one or more of the steps including receiving an inquiry from a user, which comprises a requested product; extracting one or more attributes of the requested product based on the inquiry; determining a context of the inquiry; and determining one or more substitutes for the requested product from an inventory database based on a distance between the requested product and each substitute of the one or more substitutes, as projected in a product graph relating to a domain of the requested product and the context of the inquiry. In many embodiments, the product graph relating to the domain is generated by a machine learning module configured to dynamically learn information about the domain from various sources, including one or more expert opinions, the inventory database, and/or public records; and dynamically build the product graph according to the information about the domain. An exemplary machine learning module can be configured to dynamically build the product graph by: creating one or more product categories in the domain; extracting one or more entity nodes from the one or more product categories based on the information about the domain; and build one or more connections, including an entity-entity connection between each pair of related entity nodes, a product-entity connection between each product node and each of one or more entity nodes associated with the each product node, and a product-product connection between each pair of related product nodes. Other embodiments are disclosed.

System and method for determining substitutes for a requested product

A system can include one or more processors and non-transitory computer-readable media storing computing instructions configured to run on the one or more processors and perform a method for determining one or more substitutes for a product requested in an inquiry from a user and the order of the one or more substitutes to be transmitted to the user. The method can comprise: receiving an inquiry from a user, wherein the inquiry comprises an image and a text describing a requested product; extracting one or more visual attributes of the requested product based on the image; extracting one or more textual attributes of the requested product based on the text; determining a first projected requested product, in a product graph relating to a domain of the requested product, based on the one or more visual attributes of the requested product; determining a second projected requested product in the product graph based on the one or more textual attributes of the requested product; and determining one or more substitutes for the requested product. In this embodiment, the one or more substitutes for the requested product can be determined by: determining one or more first candidates, each first candidate of the one or more first candidates existing in both the product graph and an inventory database and comprising a visual distance between the first projected requested product and the each first candidate in the product graph; determining one or more second candidates, each second candidate of the one or more second candidates existing in both the product graph and the inventory database and comprising a textual distance between the second projected requested product and the each second candidate in the product graph, the one or more substitutes comprising the one or more first candidates and the one or more second candidates; determining a score of a candidate substitute of the one or more substitutes based on a visual distance and a textual distance of the candidate substitute of the one or more substitutes; and sorting the one or more substitutes based on the score of the candidate substitute of the one or more substitutes. In this embodiment, after the one or more substitutes are determined and sorted, this method can further comprise: in response to receiving the inquiry from the user, automatically transmitting to the user higher ranked ones of the one or more substitutes. Additionally, in this embodiment, the product graph can be dynamically generated by a machine learning module based on information about the domain of the requested product; and the product graph can comprise: (a) one or more product nodes, the one or more product nodes comprising the one or more substitutes; (b) one or more entity nodes, the one or more entity nodes comprising the one or more visual attributes and the one or more textual attributes of the requested product; (c) one or more entity-entity connections between each pair of related entity nodes of the one or more entity nodes; (d) one or more product-entity connections between a product node of the one or more product nodes and an entity node of the one or more entity nodes, wherein the entity node is related to the product node; and (e) one or more product-product connection between a first product node of the one or more product nodes and a second product node of the one or more product nodes, wherein the first product node and the second product node are determined to be substitutable by or complimentary of each other. Other embodiments are disclosed.

System and method for determining substitutes for a requested product and the order to provide the substitutes

Systems and methods including one or more processors and one or more non-transitory storage devices storing computing instructions configured to run on the one or more processors and perform acts of receiving a digital image comprising multiple items; determining an embedding for the digital image using a machine learning algorithm trained on one or more combined digital images, the combined digital image comprising one or more annotated digital images; identifying an item of the multiple items in the digital image; and facilitating an alteration of a GUI on an electronic device in response to identifying the item in the digital image.

Systems and methods for identifying items in a digital image

Systems and methods including one or more processors and one or more non-transitory storage devices storing computing instructions configured to run on the one or more processors and perform acts of receiving one or more digital images from a repository of digital images; annotating the one or more digital images from the repository of digital images; digitally altering the one or more digital images, as annotated, from the repository of digital images; digitally combining the one or more digital images, as annotated and digitally altered, with at least one or more portions of one or more other digital images of the repository of digital images to create one or more combined digital images; training a machine learning algorithm on the one or more combined digital images; and storing the machine learning algorithm, as trained, in the one or more non-transitory computer readable storage devices.

Systems and methods including one or more processing modules and one or more non-transitory storage modules storing computing instructions configured to run on the one or more processing modules and perform acts of: generating a training dataset comprising synthetic training images for training a neural network detection model to identify non-compliant content in images; executing a training procedure that utilizes the synthetic training images to train the neural network detection model to identify the non-compliant content; receiving, at the neural network detection model, at least one image; and utilizing the neural network detection model to determine whether the at least one image includes the non-compliant content. Other embodiments are disclosed herein.

Systems and methods including one or more processing modules and one or more non-transitory storage modules storing computing instructions configured to run on the one or more processing modules and perform acts of transmitting an action command to each of a first server and a second server, the action command comprising one or more operation statements, transmitting an undo operation statement to each of the first server and the second server with each of the one or more operation statements, and receiving a fail notification from the second server indicating that an action associated with the action command has failed if one of the operation statements has failed on the second server. The operation statements can include instructions to create, modify, or delete a referential relationship. The undo operation statements can include instructions to undo operations associated with the operation statements.

Systems and methods for creating and maintaining referential integrity of data across multiple server systems

Systems and methods including one or more processing modules and one or more non-transitory storage modules storing computing instructions configured to run on the one or more processing modules and perform acts of: receiving an image at a first-level analysis component comprising a first neural network structure; analyzing, using the first neural network structure of the first-level analysis component, the image to determine an image category associated with the image; selecting at least one second-level analysis component that is associated with the image category to analyze the image; analyzing, using a second neural network structure associated with the at least one second-level analysis component that was selected, the image to determine whether the image includes non-compliant content; and in response to determining that the image includes non-compliant content, executing a corrective measure. Other embodiments are disclosed herein.

Systems and methods for detecting content in images using neural network architectures

Some embodiments can include a system. In some embodiments, a system can comprise one or more processors and one or more non-transitory storage devices storing computing instructions configured to run on the one or more processors and perform acts. In many embodiments, the acts can comprise dividing a computer file into two or more portions; calculating a portion hash value for each portion of the two or more portions, wherein each portion hash value for each portion of the two or more portions can be different than other portion hash values for other portions of the two or more portions; creating two or more computer file portion replicas of each portion of the two or more portions, wherein each of the two or more computer file portion replicas for a corresponding portion of the two or more portions can comprise a same portion hash value as the corresponding portion of the two or more portions; sending the two or more computer file portion replicas to a distributive data storage cluster, the distributive data storage cluster comprising two or more storage nodes; storing each of the two or more computer file portion replicas in the two or more storage nodes. Other embodiments are disclosed herein.

Systems and methods for distributive data storage

In some embodiments, a method can comprise dividing an image into two or more portions and calculating a portion hash value for each portion of the two or more portions, each portion hash value for each portion of the two or more portions is different than an other portion hash value for other portions of the two or more portions. The method can further comprise creating one or more image portion replicas of each portion of the two or more portions by replicating each portion of the two or more portions by a replication factor, the one or more image portion replicas comprising the same portion hash value as each portion of the two or more portions. The method can further comprise sending the one or more image portion replicas to a distributive data storage cluster, the distributive data storage cluster comprising two or more storage nodes, and storing each of the one or more image portion replicas in the two or more storage nodes. Other embodiments of related methods and systems are also provided.

Growing community of inventors

San Ramon, CA, United States of America

Venkatesh Kandaswamy