Abstract: Ultra-low-latency video encoding and rate control methods and systems are configured to minimize latency in applications such as cloud gaming and real-time communication. The methods and systems enhance complexity estimation and buffer management. A dynamic rate controller architecture adjusts encoding based on complexity and buffer status, ensuring consistent bitrate and quality. Deep learning models predict optimal encoder instances and quantization parameter (QP) distributions, improving encoding efficiency. System architectures, pseudocode for bitrate and QP calculations, and training and testing phases for deep learning models are also provided.

Abstract: Ultra-low-latency video encoding and rate control methods and systems are configured to minimize latency in applications such as cloud gaming and real-time communication. The methods and systems enhance complexity estimation and buffer management. A dynamic rate controller architecture adjusts encoding based on complexity and buffer status, ensuring consistent bitrate and quality. Deep learning models predict optimal encoder instances and quantization parameter (QP) distributions, improving encoding efficiency. System architectures, pseudocode for bitrate and QP calculations, and training and testing phases for deep learning models are also provided.

Abstract: Systems and methods are provided for optimizing reengagement with a content item by leveraging advanced memory-related techniques. At a first time, a user is determined to be disengaged from consuming a content item at a first position within the content item. A reengagement cue is determined by selecting a plurality of portions of the content item, wherein the plurality of portions of the content item are positioned in the content item prior to the first position within the content item. A respective memorability score is determined for each of the plurality of portions. A subset of the plurality of portions is identified based on the respective memorability scores. The reengagement cue is generated based at least part of each of the portions in the subset. At a second time, based at least in part on receiving a request to reengage with the content item, the reengagement cue is presented.

Abstract: Systems and methods are disclosed herein for protecting and managing usage of user data for artificial intelligence (AI) model training. More particularly, systems and related methods are disclosed for collecting user data by tracking interactions with a user interface of a user device in accordance with user defined consent parameters. Further, the collected data may be stored in a training database and leveraged to train an AI model for the user. However, upon receiving an input to delete data from the training database, either explicitly or from deleting an application installed on (or otherwise accessed via) a user device, the data may be removed from the training database and new AI model weights are calculated and provided to the AI model to fine tune the AI model.

Abstract: Systems and methods are disclosed for modifying content items based on eye tracking data, which may be indicative of saccadic suppression of a user. A plurality of content items is displayed on a device, including a first content item displayed at a first region. In an example, based on eye tracking data of the user, the onset of a saccade of the user is detected. An end location and duration of the saccade may be predicted based at least in part on the eye tracking data. A saccadic suppression window is identified based on the predicted duration. Based at least in part on the predicted end location, the display of the first content item is modified at a time that is within the saccadic suppression window.

Abstract: Systems and methods are disclosed for modifying content items based on eye tracking data, which may be indicative of saccadic suppression of a user. A plurality of content items is displayed on a device, including a first content item displayed at a first region. In an example, based on eye tracking data of the user, the onset of a saccade of the user is detected. An end location and duration of the saccade may be predicted based at least in part on the eye tracking data. A saccadic suppression window is identified based on the predicted duration. Based at least in part on the predicted end location, the display of the first content item is modified at a time that is within the saccadic suppression window.