Delia M
- Research Program Mentor
MS at University of Edinburgh
Expertise
Artificial intelligence(ML/NLP/DL), Computer vision & graphics
Bio
I'm driven by research that connects deep technical work with real-world applications. My journey has taken me through roles in cybersecurity, NLP, recommendation systems, health tech, and robotics, across both start-ups and large tech companies. I completed my MSc in collaboration with Amazon, focusing on adversarial attacks on transformer models against SSM architectures, examining how memory works in these models. I have also done research on data extraction attacks focusing on low-resource languages. I currently work on the autonomous behavior system for a pet robot, designing its autonomy and perception. This involves cognitive modeling, computer vision, reinforcement learning, and multimodal fusion. Ideally, the robot should develop its own personality and perceive the state of the person with whom it is interacting. Outside of work, I enjoy going to art galleries and travelling to new places! I also love running and listening to music all the time. :)Project ideas
Create cartoons from real images
Create a program that will take a real image as an input, and use Computer Vision Tools to transform the image into a cartoon. You can have fun with removing/changing the colour hue, increasing the grain, and finding parameters that create your ideal cartoon image. The outcome in creating these cartoons could be used to create a cartoon comic for real life stories or even by 'cartoonifying' real-life videos. This will give practical experience in working with OpenCV tools that are used in real-world (both in university research and corporations that are incorporating computer vision technology). You'll learn about the process of what tools to use and in what order (how they work together). Specifically looking into techniques regarding smoothing, thresholding (how edges are handled), and filters that can be applied to the image for different effects.
Predict emotions from facial images
Create a classification and prediction algorithm that will be trained on a group of faces to recognise emotion. After training, when the model is shown a new face it should be able to predict what a new face is feeling. This can be applied to UI/UX testing to get a better understanding as to how a user feels about a particular product or feature. Given that questionnaires are not always reliable, this could help supplement the findings from these exams to provide more comprehensive and honest results. This will give practical experience in working with creating a classification prediction model and corresponding modern technologies for ML. This project will improve ML understanding and provide a good background for future work done to apply AI practically.
Portfolio Website
Create your own website for projects, links and personal information. This will require making a spec of what you want the site to look like and then implementing it through HTML (defines main elements), Javascript (define ways to make the site/elements interactive) and CSS which can add color, formatting and 3D graphic backgrounds to your site.
Alien Shooter Game
Create a simple alien shooter game where you can shoot at the aliens as they descend. Some aliens can hold powerups, bombs, or can shoot. Sound can also be added.