Resolving ML Complexes

Machine Learning (ML)

• In research and development, there is a sense of complex about not using advanced techniques such as Machine Learning and Deep Learning.
  • There is a feeling that most people have not done anything more advanced than “using AI services or models provided by APIs” (blu3mo).
• However, this might just be due to an outdated image of high barriers set during high school.
  • I was just starting to learn calculus, let alone linear algebra.
  • I struggled a lot in the Master of Information Science program, which may have created an image in my mind that certain things are difficult for me.
• Now, there is a possibility that I could do some of those advanced things.
  • I have taken courses like Linear Algebra, AI (COMSW4701), and Computational Aspects of Robotics COMS W4733.
    • Thanks to that, I am becoming less intimidated by formulas involving calculus and matrices.
  • I am starting to understand how to read research papers.
  • In fact, I learned about the workings of Neural Networks in Tokyo University 1S Information α, and even played around with CNNs and GANs.
    • Automatic generation of ambigrams
• Goals
  • To gain a correct understanding of “what I can and cannot do.”
    • This will increase the things I recognize as “things I can do,” which makes me happy.
• Things I can do
  • Read interesting papers that use known ML techniques with enthusiasm.
    • In the field of HCI, this is often not considered core, so I have tended to skim through them based on the atmosphere.
      • There is a theory that if I try to read them properly, I can understand them.
      • I also want to try implementing them myself.
    • Ideas that come to mind
      • Placement Retargeting of Virtual Avatars to Dissimilar Indoor Environments
      • Researchers combining ML and HCI
        • https://rikky0611.github.io/
  • Read famous papers.
    • I want to understand at least Transformer, GPT, and Diffusion Model.
      • I think these areas have well-established highways of knowledge.
  • Engage in hands-on activities.
    • Rather than “building LLAMA/GPT from scratch,” I want to do more practical tasks to understand the extent of what I can do.