Abstract: Various embodiments classify one or more portions of an image based on deriving an “intrinsic” modality. Such intrinsic modality acts as a substitute to a “text” modality in a multi-modal network. A text modality in image processing is typically a natural language text that describes one or more portions of an image. However, explicit natural language text may not be available across one or more domains for training a multi-modal network. Accordingly, various embodiments described herein generate an intrinsic modality, which is also a description of one or more portions of an image, except that such description is not an explicit natural language description, but rather a machine learning model representation. Some embodiments additionally leverage a visual modality obtained from a vision-only model or branch, which may learn domain characteristics that are not present in the multi-modal network.

Abstract: A search system provides search results with images of products based on associations of primary products and secondary products from product image sets. The search system analyzes a product image set containing multiple images to determine a primary product and secondary products. Information associating the primary and secondary products are stored in a search index. When the search system receives a query image containing a search product, the search index is queried using the search product to identify search result images based on associations of products in the search index, and the result images are provided as a response to the query image.

Abstract: An interactive search session is implemented using an artificial intelligence model. For example, when the artificial intelligence model receives a search query from a user, the model selects an action from a plurality of actions based on the search query. The selected action queries the user for more contextual cues about the search query (e.g., may enquire about use of the search results, may request to refine the search query, or otherwise engage the user in conversation to better understand the intent of the search). The interactive search session may be in the form, for example, of a chat session between the user and the system, and the chat session may be displayed along with the search results (e.g., in a separate section of display). The interactive search session may enable the system to better understand the user's search needs, and accordingly may help provide more focused search results.

Abstract: Systems and methods for data augmentation are provided. One aspect of the systems and methods include receiving an image that is misclassified by a classification network; computing an augmentation image based on the image using an augmentation network; and generating an augmented image by combining the image and the augmentation image, wherein the augmented image is correctly classified by the classification network.

Abstract: A content analysis system provides content understanding for a content item using an attention aware multi-modal model. Given a content item, feature extractors extract features from content components of the content item in which the content components comprise multiple modalities. A cross-modal attention encoder of the attention aware multi-modal model generates an embedding of the content item using features extracted from the content components. A decoder of the attention aware multi-modal model generates an action-reason statement using the embedding of the content item from the cross-modal attention encoder.

Abstract: Systems and methods for image processing are described. Embodiments of the present disclosure receive a reference image depicting a reference object with a target spatial attribute; generate object saliency noise based on the reference image by updating random noise to resemble the reference image; and generate an output image based on the object saliency noise, wherein the output image depicts an output object with the target spatial attribute.

Abstract: The technology described herein receives a natural-language sequence of words comprising multiple entities. The technology then identifies a plurality of entities in the natural-language sequence. The technology generates a masked natural-language sequence by masking a first entity in the natural-language sequence. The technology retrieves, from a knowledge base, information related to a second entity in the plurality of entities. The technology then trains a natural-language model to respond to a query. The training uses a first representation of the masked natural-language sequence, a second representation of the information, and the first entity.

Abstract: This disclosure describes one or more implementations of systems, non-transitory computer-readable media, and methods that regularize learning targets for a student network by leveraging past state outputs of the student network with outputs of a teacher network to determine a retrospective knowledge distillation loss. For example, the disclosed systems utilize past outputs from a past state of a student network with outputs of a teacher network to compose student-regularized teacher outputs that regularize training targets by making the training targets similar to student outputs while preserving semantics from the teacher training targets. Additionally, the disclosed systems utilize the student-regularized teacher outputs with student outputs of the present states to generate retrospective knowledge distillation losses.

Abstract: Content creation techniques are described that leverage content analytics to provide insight and guidance as part of content creation. To do so, content features are extracted by a content analytics system from a plurality of content and used by the content analytics system as a basis to generate a content dataset. Event data is also collected by the content analytics system from an event data source. Event data describes user interaction with respective items of content, including subsequent activities in both online and physical environments. The event data is then used to generate an event dataset. An analytics user interface is then generated by the content analytics system using the content dataset and the event dataset and is usable to guide subsequent content creation and editing.

Abstract: Embodiments are disclosed for training a neural network classifier to learn to more closely align an input image with its attribution map. In particular, in one or more embodiments, the disclosed systems and methods comprise receiving a training image comprising a representation of one or more objects, the training image associated with at least one label for the representation of the one or more objects, generating a perturbed training image based on the training image using a neural network, and training the neural network using the perturbed training image by minimizing a combination of classification loss and attribution loss to learn to align an image with its corresponding attribution map.

Abstract: In some embodiments, techniques for extracting high-value sequential patterns are provided. For example, a process may involve training a machine learning model to learn a state-action map that contains high-utility sequential patterns; extracting at least one high-utility sequential pattern from the trained machine learning model; and causing a user interface of a computing environment to be modified based on information from the at least one high-utility sequential pattern.

Abstract: In implementations of systems for generating images for virtual try-on and pose transfer, a computing device implements a generator system to receive input data describing a first digital image that depicts a person in a pose and a second digital image that depicts a garment. Candidate appearance flow maps are computed that warp the garment based on the pose at different pixel-block sizes using a first machine learning model. The generator system generates a warped garment image by combining the candidate appearance flow maps as an aggregate per-pixel displacement map using a convolutional gated recurrent network. A conditional segment mask is predicted that segments portions of a geometry of the person using a second machine learning model. The generator system outputs a digital image that depicts the person in the pose wearing the garment based on the warped garment image and the conditional segmentation mask using a third machine learning model.

Abstract: Methods and systems are provided for generating and providing insights associated with a journey. In embodiments described herein, journey data associated with a journey is obtained. A journey can include journey paths indicating workflows through which audience members can traverse. The journey data can include audience member attributes (e.g., demographics) and labels indicating journey paths traversed by audience members. A set of audience segments are determined that describe a set of audience members traversing a particular journey path. The set of audience segments can be determined using the journey data to train a segmentation model and, thereafter, analyzing the segmentation model to identify patterns that indicate audience segments associated with the particular journey path. An indication of the set of audience segments that describe the set of audience members traversing the particular journey path can be provided for display.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media for generating trained neural network with increased robustness against adversarial attacks by utilizing a dynamic dropout routine and/or a cyclic learning rate routine. For example, the disclosed systems can determine a dynamic dropout probability distribution associated with neurons of a neural network. The disclosed systems can further drop neurons from a neural network based on the dynamic dropout probability distribution to help neurons learn distinguishable features. In addition, the disclosed systems can utilize a cyclic learning rate routine to force copy weights of a copy neural network away from weights of an original neural network without decreasing prediction accuracy to ensure that the decision boundaries learned are different.

Abstract: Machine-learning based multi-step engagement strategy modification is described. Rather than rely heavily on human involvement to manage content delivery over the course of a campaign, the described learning-based engagement system modifies a multi-step engagement strategy, originally created by an engagement-system user, by leveraging machine-learning models. In particular, these leveraged machine-learning models are trained using data describing user interactions with delivered content as those interactions occur over the course of the campaign. Initially, the learning-based engagement system obtains a multi-step engagement strategy created by an engagement-system user. As the multi-step engagement strategy is deployed, the learning-based engagement system randomly adjusts aspects of the sequence of deliveries for some users.

Abstract: Systems and methods for machine learning are described. Embodiments of the present disclosure receive state information that describes a state of a decision making agent in an environment; compute an action vector from an action embedding space based on the state information using a policy neural network of the decision making agent, wherein the policy neural network is trained using reinforcement learning based on a topology loss that constrains changes in a mapping between an action set and the action embedding space; and perform an action that modifies the state of the decision making agent in the environment based on the action vector, wherein the action is selected based on the mapping.

Abstract: Generating a machine learning model that is trained using retrospective loss is described. A retrospective loss system receives an untrained machine learning model and a task for training the model. The retrospective loss system initially trains the model over warm-up iterations using task-specific loss that is determined based on a difference between predictions output by the model during training on input data and a ground truth dataset for the input data. Following the warm-up training iterations, the retrospective loss system continues to train the model using retrospective loss, which is model-agnostic and constrains the model such that a subsequently output prediction is more similar to the ground truth dataset than the previously output prediction. After determining that the model's outputs are within a threshold similarity to the ground truth dataset, the model is output with its current parameters as a trained model.

Abstract: Systems, methods, and computer storage media are disclosed for predicting visual compatibility between a bundle of catalog items (e.g., a partial outfit) and a candidate catalog item to add to the bundle. Visual compatibility prediction may be jointly conditioned on item type, context, and style by determining a first compatibility score jointly conditioned on type (e.g., category) and context, determining a second compatibility score conditioned on outfit style, and combining the first and second compatibility scores into a unified visual compatibility score. A unified visual compatibility score may be determined for each of a plurality of candidate items, and the candidate item with the highest unified visual compatibility score may be selected to add to the bundle (e.g., fill the in blank for the partial outfit).

Abstract: In implementations of systems for generating images for virtual try-on and pose transfer, a computing device implements a generator system to receive input data describing a first digital image that depicts a person in a pose and a second digital image that depicts a garment. Candidate appearance flow maps are computed that warp the garment based on the pose at different pixel-block sizes using a first machine learning model. The generator system generates a warped garment image by combining the candidate appearance flow maps as an aggregate per-pixel displacement map using a convolutional gated recurrent network. A conditional segment mask is predicted that segments portions of a geometry of the person using a second machine learning model. The generator system outputs a digital image that depicts the person in the pose wearing the garment based on the warped garment image and the conditional segmentation mask using a third machine learning model.

Abstract: Techniques described herein extract form structures from a static form to facilitate making that static form reflowable. A method described herein includes accessing low-level form elements extracted from a static form. The method includes determining, using a first set of prediction models, second-level form elements based on the low-level form elements. Each second-level form element includes a respective one or more low-level form elements. The method further includes determining, using a second set of prediction models, high-level form elements based on the second-level form elements and the low-level form elements. Each high-level form element includes a respective one or more second-level form elements or low-level form elements. The method further includes generating a reflowable form based on the static form by, for each high-level form element, linking together the respective one or more second-level form elements or low-level form elements.