Abstract: Described herein are technologies related to constructing supplemental content items that summarize electronic landing pages. A sequence to sequence model that is configured to construct supplemental content items is trained based upon a corpus of electronic landing pages and supplemental content items that have been constructed by domain experts, wherein each landing page has a respective supplemental content item assigned thereto. The sequence to sequence model is additionally trained using self critical sequence training, where estimated click through rates of supplemental content items generated by the sequence to sequence model are employed to train the sequence to sequence model.

Abstract: A computer-implemented technique uses one or more neural networks to identify at least one item name associated with an input image using a multi-modal fusion approach. The technique is said to be multi-modal because it collects and processes different kinds of evidence regarding each detected item name. The technique is said to adopt a fusion approach because it fuses the multi-modal evidence into an output conclusion that identifies at least one item name associated with the input image. In one example, a first mode collects evidence by identifying and analyzing regions in the input image that are likely to include item name-related information. A second mode collects and analyzes any text that appears as part of input image itself. A third mode collects and analyzes text that is not included in the input image itself, but is nonetheless associated with the input image.

Abstract: Systems and methods for dynamically modifying a cache associated with a neural network model of a natural language generator are described. In examples, a neural network model employs a beam search algorithm at a decoder when decoding output and generating predicted output candidates. The decoder utilizes caching techniques to improve a speed at which the neural network operations. When an amount of memory utilized by one or more caches of the neural network model is determined to exceed a threshold memory size, a layer-specific portion of a cache associated with a layer of the neural network model is identified. The identified layer-specific portion of the cache can be deleted when the amount of memory utilized by the cache of the neural network model exceeds the threshold memory size. In examples, data in the cache is deduplicated and/or deleted.

Abstract: A transformer-based neural network includes at least one mask attention network (MAN). The MAN computes an original attention data structure that expresses influence between pairs of data items in a sequence of data items. The MAN then modifies the original data structure by mask values in a mask data structure, to produce a modified attention data structure. Compared to the original attention data structure, the modified attention data structure better accounts for the influence of neighboring data items in the sequence of data items, given a particular data item under consideration. The mask data structure used by the MAN can have static and/or machine-trained mask values. In one implementation, the transformer-based neural network includes at least one MAN in combination with at least one other attention network that does not use a mask data structure, and at least one feed-forward neural network.

Abstract: Knowledge graphs can greatly improve the quality of content recommendation systems. There is a broad variety of knowledge graphs in the domain including clicked user-ad graphs, clicked query-ad graphs, keyword-display URL graphs etc. A hierarchical Transformer model learns entity embeddings in knowledge graphs. The model consists of two different Transformer blocks where the bottom block generates relation-dependent embeddings for the source entity and its neighbors, and the top block aggregates the outputs from the bottom block to produce the target entity embedding. To balance the information from contextual entities and the source entity itself, a masked entity model (MEM) task is combined with a link prediction task in model training.

Abstract: A computer-implemented technique uses a generative adversarial network (GAN) to jointly train a generator neural network (“generator”) and a discriminator neural network (“discriminator”). Unlike traditional GAN designs, the discriminator performs the dual role of: (a) determining one or more attribute values associated with an object depicted in input image fed to the discriminator; and (b) determining whether the input image fed to the discriminator is real or synthesized by the generator. Also unlike traditional GAN designs, an image classifier can make use of a model produced by the GAN's discriminator. The generator receives generator input information that includes a conditional input image and one or more conditional values that express desired characteristics of the generator output image. The discriminator receives the conditional input image in conjunction with a discriminator input image, which corresponds to either the generator output image or a real image.

Abstract: A taxonomy of categories, attributes, and values can be conflated with new data triplets by identifying one or more conflation candidates among the attribute-value pairs within a category of the taxonomy that matches the category of the data triplet, and determining a suitable merge action for conflating the data triplet with each conflation candidate. The task of determining merge actions may be cast as a classification problem, and may be solved by an ensemble classifier.

Abstract: Technologies are described herein that relate to identifying supplemental content items that are related to objects captured in images of webpages. A computing system receives an indication that a client computing device has a webpage displayed thereon that includes an image. The image is provided to a first DNN that is configured to identify a portion of the image that includes an object of a type from amongst a plurality of predefined types. Once the portion of the image is identified, the portion of the image is provided to a plurality of DNNs, with each of the DNNs configured to output a word or phrase that represents a value of a respective attribute of the object. A sequence of words or phrases output by the plurality of DNNs is provided to a search computing system, which identifies a supplemental content item based upon the sequence of words or phrases.

Abstract: Neural network-based categorization can be improved by incorporating graph neural networks that operate on a graph representing the taxonomy of the categories into which a given input is to be categorized by the neural network based-categorization. The output of a graph neural network, operating on a graph representing the taxonomy of categories, can be combined with the output of a neural network operating upon the input to be categorized, such as through an interaction of multidimensional output data, such as a dot product of output vectors. In such a manner, information conveying the explicit relationships between categories, as defined by the taxonomy, can be incorporated into the categorization. To recapture information, incorporate new information, or reemphasize information a second neural network can also operate upon the input to be categorized, with the output of such a second neural network being merged with the output of the interaction.

Abstract: A computer-implemented technique uses one or more neural networks to identify at least one item name associated with an input image using a multi-modal fusion approach. The technique is said to be multi-modal because it collects and processes different kinds of evidence regarding each detected item name. The technique is said to adopt a fusion approach because it fuses the multi-modal evidence into an output conclusion that identifies at least one item name associated with the input image. In one example, a first mode collects evidence by identifying and analyzing regions in the input image that are likely to include item name-related information. A second mode collects and analyzes any text that appears as part of input image itself. A third mode collects and analyzes text that is not included in the input image itself, but is nonetheless associated with the input image.

Abstract: A computer-implemented technique uses a generative adversarial network (GAN) to jointly train a generator neural network (“generator”) and a discriminator neural network (“discriminator”). Unlike traditional GAN designs, the discriminator performs the dual role of: (a) determining one or more attribute values associated with an object depicted in input image fed to the discriminator; and (b) determining whether the input image fed to the discriminator is real or synthesized by the generator. Also unlike traditional GAN designs, an image classifier can make use of a model produced by the GAN's discriminator. The generator receives generator input information that includes a conditional input image and one or more conditional values that express desired characteristics of the generator output image. The discriminator receives the conditional input image in conjunction with a discriminator input image, which corresponds to either the generator output image or a real image.

Abstract: A technique is described herein for processing a given query item in a latency-efficient and resource-efficient manner. The technique uses a first transformer-based encoder to transform the given query item into an encoded query item. In one case, the given query item is an expression that includes one or more query-expression linguistic tokens. The technique includes a second transformer-based encoder for transforming a given target item into an encoded target item. The given target item may likewise correspond to an expression that includes one or more target-expression linguistic tokens. A similarity-assessing mechanism then assesses the semantic similarity between the given query item and the given target item based on the encoded query item and the encoded target item. Each transformer-based encoder uses one or more self-attention mechanisms. The second transformer-based encoder can optionally perform its work in an offline manner, prior to receipt of the given query item.

Abstract: Technologies are described herein that relate to identifying supplemental content items that are related to objects captured in images of webpages. A computing system receives an indication that a client computing device has a webpage displayed thereon that includes an image. The image is provided to a first DNN that is configured to identify a portion of the image that includes an object of a type from amongst a plurality of predefined types. Once the portion of the image is identified, the portion of the image is provided to a plurality of DNNs, with each of the DNNs configured to output a word or phrase that represents a value of a respective attribute of the object. A sequence of words or phrases output by the plurality of DNNs is provided to a search computing system, which identifies a supplemental content item based upon the sequence of words or phrases.

Abstract: Described herein are technologies related to constructing supplemental content items that summarize electronic landing pages. A sequence to sequence model that is configured to construct supplemental content items is trained based upon a corpus of electronic landing pages and supplemental content items that have been constructed by domain experts, wherein each landing page has a respective supplemental content item assigned thereto. The sequence to sequence model is additionally trained using self critical sequence training, where estimated click through rates of supplemental content items generated by the sequence to sequence model are employed to train the sequence to sequence model.

Abstract: The present application describes a system and method for converting a natural language query to a standard query using a sequence-to-sequence neural network. As described herein, when a natural language query is receive, the natural language query is converted to a standard query using a sequence-to-sequence model. In some cases, the sequence-to-sequence model is associated with an attention layer. A search using the standard query is performed and various documents may be returned. The documents that result from the search are scored based, at least in part, on a determined conditional entropy of the document. The conditional entropy is determined using the natural language query and the document.

Abstract: The present application describes a system and method for converting a natural language query to a standard query using a sequence-to-sequence neural network. As described herein, when a natural language query is receive, the natural language query is converted to a standard query using a sequence-to-sequence model. In some cases, the sequence-to-sequence model is associated with an attention layer. A search using the standard query is performed and various documents may be returned. The documents that result from the search are scored based, at least in part, on a determined conditional entropy of the document. The conditional entropy is determined using the natural language query and the document.

Abstract: A computer-implemented technique is described herein for providing a digital content item using a generator component. The generator component corresponds to a sequence-to-sequence neural network that is trained using an adversarial generative network (GAN) system. In one approach, the technique involves: receiving a query from a user computing device over a computer network; generating random information; generating a key term using the generator component based on the query and the random information; selecting at least one content item based on the key term; and sending the content item(s) over the computer network to the user computing device.

Abstract: Systems and Methods for multi-modal or multimedia image retrieval are provided. Automatic image annotation is achieved based on a probabilistic semantic model in which visual features and textual words are connected via a hidden layer comprising the semantic concepts to be discovered, to explicitly exploit the synergy between the two modalities. The association of visual features and textual words is determined in a Bayesian framework to provide confidence of the association. A hidden concept layer which connects the visual feature(s) and the words is discovered by fitting a generative model to the training image and annotation words. An Expectation-Maximization (EM) based iterative learning procedure determines the conditional probabilities of the visual features and the textual words given a hidden concept class. Based on the discovered hidden concept layer and the corresponding conditional probabilities, the image annotation and the text-to-image retrieval are performed using the Bayesian framework.

Abstract: Images are encoded into multidimensional vectors in a high-dimensional space utilizing an image model and textual content utilizing a text model. At least one of the image model and/or the text model are tuned such that the point within the multidimensional space pointed to by a vector encoded from an image is proximate to, as determined within the context of that multidimensional space, a point pointed to by a vector encoded from correlated textual content. Received images and textual content are encoded into image vectors and text vectors, respectively, and stored in an image graph and text graph, respectively. An input image can then be encoded as an input image vector and utilized to find close vectors in both the image graph and the text graph, thereby enabling an input image to be utilized to search textual content without using classifiers to deduce textual content therefrom.

Abstract: A taxonomy of categories, attributes, and values can be conflated with new data triplets by identifying one or more conflation candidates among the attribute-value pairs within a category of the taxonomy that matches the category of the data triplet, and determining a suitable merge action for conflating the data triplet with each conflation candidate. The task of determining merge actions may be cast as a classification problem, and may be solved by an ensemble classifier.