• 2017

    • "Learning to high five: making computers understand our hands!"

      Muhammad Asad
      - 3 Minute Thesis Competition, City, University of London, UK

      Given the current technologies, if Terminator was to exist today he would have a very hard time understanding a lot of things in his surrounding environment. Specifically, he would not be able to understand how we use our hands for complex communication and as manipulation tools for interacting with the world around us. To make matters worse, the shape and size of a human hand varies quite rapidly across different people, while each person might have their own style variations of doing the same hand-based interaction. From a machine learning perspective, this makes the problem really hard. My research addresses this by proposing the use of multiple expert learners that are based on the idea that a skilled worker can excel at only a single skill. Similarly, the expert learners are trained on subsets of data, which enables them to excel only for that sub-domain. My research looks at how such models can be trained and proposes ways to incorporate several expert learners to work collectively to improve overall accuracy and robustness of a model. My research has an impact on future hand-based human-computer interfaces, where we would be able to interact with a virtual world similar to how we interact with the real world. Most important of all, it will enable future Terminator robots to recognize and understand our instructions based on our hand movements.

    • "Learning to see: how machines learn to understand images?"

      Muhammad Asad
      - Pint of Science, London, UK

      Recent years have seen significant advancements in Computer Vision algorithms, where computers can look at image and video data to understand a given scene. Be it face recognition or smart cars in real-world scenarios, computers are beginning to excel and, in some cases, surpass their human counterparts. This talk introduces the audience to the concept of machine learning, which enables computers to learn from known examples of a given task and apply their learned knowledge to a new, but similar, scenario.

    • "SPORE: Staged Probabilistic Regression for Hand Orientation and Pose Inference"

      Muhammad Asad, Greg Slabaugh
      - Data Natives, London, UK

    • "Regression including Advanced Non-linear Methods"

      Muhammad Asad, Greg Slabaugh
      - Computer Vision Group Team Meetings

  • 2016

  • 2015

  • 2014

  • 2013