Projects

Sign Language Recognition and Translation

An end-to-end pipeline for real-time sign language recognition and translation using cutting-edge deep learning algorithms.
  • Tab content image

    Key Contributions:

    - Surveyed, evaluated, and classified a wide range of sign language translation systems, from rule-based approaches to deep transformers.
    - Explored the potential of LLMs (e.g., ChatGPT) to process or assist sign language translation tasks.
    - Created the MedASL dataset for American Sign Language in medical contexts.
    - Developed and optimized pipelines integrating video backbones with transformer-based architectures.
    - Contributed to the creation of VisioSLR, a vision-driven YOLO-based framework to enhance real-time sign language recognition.

Efficient Sequence Learning

Novel transformer variants for low-resource, sequence-related tasks like sign language translation. The project aims to balance training efficiency, temporal awareness, and translation accuracy.
  • Tab content image

    Models Developed:

    - ADAT (Adaptive Transformer):
    A time-series-aware architecture that achieved +14% improvement in training efficiency over baselines by dynamically adapting to temporal variations in sign sequences.
    - GLoT (Gated-Logarithmic Transformer):
    A gated-logarithmic mechanism to reduce computational complexity and improve generalization on long input sequences.

Accessible AI Systems

Investigation on the human-AI interaction in the UAE's Deaf community, including accessibility gaps in voice assistants.
  • Tab content image

    Key Contributions:

    - Performed experiments with IoT devices and intelligent assistants (e.g., Amazon Alexa) to evaluate usability and interaction patterns for Deaf users.
    - Conducted user-centric evaluations to identify linguistic, interface, and accessibility gaps.
    - proposed recommendations and design improvements for inclusive AI systems in smart home environments.

AI for Mental Health

A proactive framework for event-based emotion detection using ML and NLP, targeting mental health scenarios with social media data.
  • Tab content image

    Key Contributions:

    - Proposed a proactive event-based framework that detects public emotions.
    - Collected and annotated a twitter dataset with Plutchik’s Wheel of Emotions.

Submission Successful

The form has been successfully submitted.