Research
Patent filings and peer-reviewed research spanning Extended Reality, Generative AI Videos, Dialogue Systems, and Reinforcement Learning, in industry & academia.
Methods and Systems for Estimating and Updating a Skill Score
Details will be furnished once the application is published.
A Method and A System for Transforming a Video Clip presented in an Environment
Details will be furnished once the application is published.
Method and System for Providing Device Notifications in a Video
A method and a system for embedding device notifications in a video is disclosed. The method may include receiving a device notification associated with an electronic device, detecting one or more objects across a plurality of frames of the video, and selecting at least one frame set based on that detection. It then determines the least relevant objects in each frame, obtains a personalized notification object to replace the least relevant object, and produces the corresponding modified frames of the video.
Methods and Systems for Transforming Real-World Entities' Shadow and Reflection in an Extended Reality Device
Disclosed are systems and methods for transforming the shadow and reflection of real-world entities in an extended reality (XR) device. Semantic information associated with the field of view (FOV) of the XR device is first determined, followed by a plurality of attributes for generating one or more virtual appearances corresponding to the shadow and the reflection of the real-world entities. The virtual appearances are then generated and overlaid on the image of the FOV, such that the shadow and reflection of the real-world entities are transformed.
System and Method for Displaying Adaptive Contextual Calm Information during Non-Active States of Electronic Devices
Disclosed is a method for displaying adaptive contextual calm information during a non active state of an electronic device. The method includes obtaining input data from one or more smart devices. Further, the method includes generating, based on the obtained input data, prioritized context vectors associated with the one or more smart devices. Further, the method includes generating, based on the generated prioritized context vectors, contextual images associated with the one or more smart devices. Furthermore, the method includes calculating a cognitive load and a comprehension index for each of the generated contextual images. Furthermore, the method includes generating a set of images for each contextual image among the generated contextual images. Furthermore, the method includes displaying the adaptive contextual calm information as an image among the generated set of images based on a type of the non-active state.
Methods and Systems for Dynamic Context and Response Generation of a Virtual Assistant System
A method and system for dynamic customization of virtual assistant systems. Context is extracted from a unified representation of individual features, where features and previous context are combined in a two-step process to calculate the final context and its priority. This extracted context is incorporated into the user input to obtain better entity features, leading to context-aware decision making for a dialog manager. The methodology incorporates context from multiple sources and continuous feedback generated from the user response and a backend process analysis, which is continuously analyzed to intimate the user of any future abnormality in the functioning of a device or an event.
Systems and methods for adjusting camera configurations in a user device
Improved camera settings through Machine Learning.
This patent describes a way to automatically adjust camera settings based on the content of the image. By analyzing objects in the scene and comparing them to a database of ideal images and their settings, the camera can be optimized for capturing the best possible shot.
BEAR: Reinforcement Learning for Throughput Aware Borrowing in Energy Harvesting Systems
Energy Borrowing (EB) aided Energy harvesting (EH) systems provide a greener alternative to self-sustaining electronic devices in a complex, unprecedented environment by borrowing energy from a supplementary source to regulate the data transmission flow. We propose a reinforcement learning-based algorithm for energy scheduling policy which jointly optimizes the EB and utilizes harvested energy for efficient data transfer at every time instant. As the exact pattern of harvested energy and channel conditions at any time slot is unknown, the proposed algorithm, BEAR (Borrowing Energy with Adaptive Rewards), based on actor-critic architecture, learns the optimal power allocation policy for the transmission node. Our designed reward function accommodates the concept of adaptive penalty to punish the transmission node for selecting unfavourable actions. Our simulations show that the BEAR algorithm providing efficient energy management with a focus on throughput maximization yields a 35.45% enhancement in sum throughput over a typical non-borrowing system. Lastly, nontrivial design insights are outlined via numerical results to quantify the practical efficacy of BEAR for EH systems.
Debunking Fake News by Leveraging Speaker Credibility and BERT Based Model
The exponential growth in fake news and its role in deteriorating general public trust and democratic standards certainly calls for some counter combat approaches. The prediction of chances of news to be fake is deemed to be hard task since most of the deceptive news has its roots in true news. With a minor fabrication in legitimate news, influential fake news can be created that can be used for political, entertainment, or business-related gains. This work provides a novel intuitive approach to exploit data from multiple sources to segregate news into real and fake. To efficiently capture the contextual information present in the data, Bidirectional Encoder Representations from Transformer (BERT) have been deployed. It attempts to further enhance the performance of the deceptive news detection model by incorporating information about the speaker profile and the credibility associated with him/her. A hybrid sequence encoding model has been proposed to harvest the speaker profile and speaker credibility data which makes it useful for prediction. On evaluation over benchmark fake news dataset LIAR, our model outperformed the previous state-of-the-art works. This attests to the fact that the speaker’s profile and credibility play a crucial role in predicting the validity of news.