🌳 Personal Growth and SGD
I've long believed that you don't always need an end goal; you just need a direction. At any given moment, you need to be clear about what you're doing: weighing different options, gathering data, figuring out what you want.
College
I'm in college right now, and I don't know exactly what career path I want to pursue. But I'm very consciously trying to narrow it down. I always have a direction that comes from searching for answers to different questions:
- Last year I asked myself, "Do I want to pursue machine learning or quantum computing?" So I did everything I could to answer that question. I talked to research labs in both fields; after reading papers on quantum error correction and transformer fine-tuning, I decided that quantum computing wasn't for me.
- This year, I had a new question: "Do I want to pursue a PhD in ML?" To answer this new question, I've been taking as many ML courses as I can. Joining a research lab and applying to the BS+MS program at my university are other ways I've been collecting even more data.
By iteratively asking and answering questions like these, I'll eventually reach my destination. I don't know exactly where it's going to be, but I know I'm going to love it.
Stochastic Gradient Descent
Intentionally answering your questions like this is similar to following a gradient descent algorithm. The SGD algorithm doesn't find the optimum directly; it takes plenty of bad steps. On average, however, the good steps cancel out the poor ones—and SGD finds an optimal set of parameters.
Similarly, we can't directly figure out what will make us the happiest and most fulfilled; we make a lot of mistakes. But as we learn and continue to take deliberate steps to answer our questions, we will hopefully converge on an optimal set of ways to live our lives. That's why I keep asking myself questions, hoping I'll soon converge on a lifestyle that's optimal for me.