Taro Logo
1

How Do You Implement New Knowledge in Practical Projects?

Profile picture
Mid-Level Software Engineer at Taro Community21 days ago

I am currently learning Machine Learning and AI as they are essential for my ongoing project.

I've found these topics particularly challenging to master on my own due to their heavy reliance on mathematics—or at least that's how it seems to me with machine learning.

Moreover, I tend to study thoroughly, aiming to understand everything from the basics to advanced concepts before I attempt to develop practical software applications.

Could you share any strategies or advice on how to effectively learn and apply new technical knowledge in project settings?

34
2

Discussion

(2 comments)
  • 1
    Profile picture
    Founding ML Engineer @ Lancey (YC S22)
    21 days ago

    Hey OP, are you able to share more on how the ML is being applied? In most cases you don't need to spend too much time understanding the low level math for ML unless you're dealing with customizing models.

    I would focus on understanding

    1. How does this model work -- the intuition behind it
    2. What are the assumptions this model makes for e.g. some models assume the data is normally distributed
    3. What are the pros of using this model
    4. What are the cons of using this model

    If you understand this you should be good for most cases.

    And if you do need to actually understand how the model works in depth I suggest only learning the parts you need to as you study

    E.g. you are learning about Convolutional Neural Networks - CNNs

    • Understand how CNNs work intuitively
    • Then you want to know what a convolution is so you dive deep into the math/code
    • Then you realize you need to understand what a dot product is, so then look at that
    • Keep repeating till you understand everything, almost like a DFS tree search

    Other things that help:

    • Look at minimal code replicas of models. Every model you want will have some version of minimal version of it (even LLMs)
    • Grab the code and walk through what's happening. For CNN you might look at how the image is before and after a convolution is applied. Step through the code and understand what is happening.
    • Start small, work with really small data that you can reproduce by hand. For CNNs this might be an "image" that's 5x5 0/1 and you can do the math on paper on what each step should look like

    Feel free to DM me on Slack (@Sai B) if you want to chat more

    • 1
      Profile picture
      Tech Lead/Manager at Meta, Pinterest, Kosei
      14 days ago

      Such a good point about not needing to understand the low-level math for almost all ML jobs.

      Be careful not to get "nerd-sniped". It may be intellectually interesting to dive into how a library or model works, but your primary job is to deliver a business outcome -- focus on that!