Abstract: Aspects of the present disclosure are directed to generating media element(s) using a multimodal scene graph. A scene manager can process visual information, such as video, images, and/or a recorded artificial relay scene, and generate a multimodal scene graph that comprises components and metadata generated via the processing. The scene manager can utilize the multimodal scene graph to generate social media elements, such as images, video, and/or artificial reality scenes. For example, a video of a user can be converted to a multimodal scene graph, which can be used to generate one or more images (e.g., memes, animated images, stickers, etc.), such as an image that represents the user via an avatar of the user. This generated media can be shared with other social platform users, and the stored multimodal scene graph can be accessed by the others to generate variations of the media.

Abstract: Aspects of the present disclosure are directed to composing an editable artificial reality scene. Implementations include a composer tool for generating editable artificial reality scenes. For example, a scene creator can generate an artificial reality scene and publish the scene such that it can be accessed and/or edited by other users. The artificial reality scene can include scene components, such as an avatar, virtual object, motion dynamics, background, camera perspective, and others. Implementations of a privilege manager can manage access and/or edit privileges for stored artificial reality scenes. For example, the scene creator can define the scope of users that can access and/or edit the original scene and/or define the components of the scene that are editable. In some implementations, an original artificial reality scene can be edited multiple times by multiple users, thus generating multiple edited versions of the original scene.

Abstract: An online system actively and randomly selects content items to be labeled for training a classifier. An online system receives content items from client devices of users and selects sets of the content items to be labeled by human labelers. The randomly selected content items are selected at random from the received content items, and the actively selected content items are selected based on the classifier's confidence in accurately predicting the classification of the content items. The online system may use a histogram of content items to actively select content items. The online system assigns the content items to bins of the histogram based on priority scores and selects content items with priority scores of the highest percentile. The online system provides the selected content items to human labelers for labeling. The labeled content items are then used for training the classifier.

Abstract: An online system actively and randomly selects content items to be labeled for training a classifier. An online system receives content items from client devices of users and selects sets of the content items to be labeled by human labelers. The randomly selected content items are selected at random from the received content items, and the actively selected content items are selected based on the classifier's confidence in accurately predicting the classification of the content items. The online system may use a histogram of content items to actively select content items. The online system assigns the content items to bins of the histogram based on priority scores and selects content items with priority scores of the highest percentile. The online system provides the selected content items to human labelers for labeling. The labeled content items are then used for training the classifier.

Abstract: In one embodiment, a method may include receiving, from a client system of a first user of an online social network, a search query for event listings where the search query specifies one or more criteria, identifying first event listings that match the query, wherein each first event listing is associated with a respective first event and further associated with one or more predetermined categories selected from a plurality of categories in a hierarchical category index and where each category associated with a first event listing is determined using a machine-learning model that analyzes the first event listing, identifying second event listings based on the first event listings, wherein each second event listing is associated with a predetermined category from an adjacent category in the hierarchical category index, and sending, to the client system, instructions for presenting a search-results interface comprising references to the first and second event listings.