HOME
POSTS
It’s almost the New Year, so I feel like reflecting a bit on the past year, and thinking about the one that’ll start soon.
Some say that the first step is always the hardest, but it seems that it doesn’t apply to making blog posts often (at least for me). 3 posts per year is not that much, but if we want to be on the positive side - it’s an easy bar to raise next year :)
I guess that counts both as a reflection and a resolution, although quite a vague one. To make it more concrete, I’ll set a goal of making at least 12 posts here in 2019. One might think of it as 1 post per month, but I’ll leave a small loophole, allowing myself to post several times on some months, and not posting on others. Not too ambitious, but hey, slow and steady wins the race!
This should be made easier by the fact that there have been several things that I wanted to write about, but never had the proper mood to do so, including:
- moving from writing CV/ML/DL code on frontend to backend, and learning about some lower-level Python details, as well as some DevOps-related stuff
- taking part in Kaggle competition seriously for the first time: TGS Salt Identification Challenge, 11th place solution writeup
- doing some vision/deep-learning experiments
That’s a few posts already, and we’ve barely started.
Since taking part in Kaggle turned out to be a great learning experience in several areas (research, implementation, working under pressure, team collaboration), I will try to take part in at least 2 competitions, with corresponding write-ups here.
I could’ve set a goal of becoming Kaggle competition master next year (which should be possible after 2 competitions), but this depends on factors beyond my control (number and skill of other participants), so I will focus on doing my best each time, and won’t get distracted by anything else.
Apart from that, I feel like I should pay more attention to research in unsupervised and reinforcement learning areas. They seem to be less applicable to production right now (at least in projects that I work on), but I expect that to change as soon as research in data-rich areas will reach saturation and more attention will be drawn to approaches which lack labelled data.
It’s not easy to quantify “paying more attention”, so I’ll try to work on toy projects involving unsupervised and reinforcement learning. Still vague, but I don’t want to limit myself too much in choosing what it will look like.
I guess that’s already enough to work on, and instead of throwing more stuff into this “list”, I’d rather work more on what’s already mentioned above.
Twitter Facebook Google+