I originally planned on writing out a full explanation of the research in the paper that we submitted to The International Conference on Machine Learning (ICML) last Friday. However, since we uploaded to Arxiv at 10pm on Friday, and the verification process is slower over the weekends, I still don’t have a nice arxiv link to put in the post. I also have way too much work right now to make this technical post, and my blog also still doesn’t have Latex functionality.
Instead, I’m going to write a Life Update, the second half of which will kinda be about the paper because it was the thing that I worked on and thought about in the week before the Jan 27 (AOE midnight) deadline.
I’ve also set up a mailing list, so if you enter your email into the “subscribe to our newsletter” form on the main page I’ll email you each time I post. I’m also probably going to start putting the date into blog post titles and just ignoring bluehost’s dates.
Week 3: Sun 1/16 – Sat 1/22
Hours: 9:40 (U), 10:00 (M), 9:30 (T), 12:35 (W), 10:00 (R), 13:00 (F), 10:10 (S)
For the first half of the week, I was still somewhat recovering from getting sick (which you can read about it in my previous life update). The period from Monday to Thursday was a delightful return to normal. After being sick for a week, and suffering through the accompanied decrease of QOL, it was amazing being back to normal. I promised myself that I would appreciate my regained health, and I did. All in all, I lost ten pounds (mostly water) through the process. However, getting sick was probably a net positive on my mental health. It was a forced break where I stopped overthinking things, and all of my thoughts were single and easy. I even considered becoming religious.
Beginning on Friday though, the ICML paper grind began. That Friday night was really busy, I had classes, winter rowing, ran physics club, made an amazing poster w/ DH, and then spent the rest of the night trying to solve the fact that I did not have a full convergence proof for our ICML submission. On Saturday, the only thing I did was work on the paper, and this mainly took the form of me rereading papers. Throughout the entire process, I mainly referenced three papers, and I’ll reveal them in my full post about the paper. 🙂
Week 4: Sun 1/23 – Sat 1/29
Hours: 10:00 (U), 14:20 (M), 12:00 (T), 13:30 (W), 13:35 (R), 9:10 (F), 8:30 (S)
The only times in my logs (and probably ever) when I worked more than 14 hours in one day were 2/8/2021 (14:30) to finish a history podcast, 3/1/2020 (14:00) to do USACO on a Monday, 6/17/2021 (14:40) for AP CSA, MIMO, & MOP math. Hitting 14 hours is a momentous occasion, and will always be an occasion 14 hours + 8 hours of sleep + 2 hours of life essentials is the entire day. Of course, I can’t hit 100% efficiency and the 14 hours are usually achieved by sleeping less.
The period from Monday to Thursday when I made the most progress on the paper. I made an observation that put me in the right direction on Thursday at 10:30am, and then made the crucial observation at around 11:30am while in a SAC study room with DH. Thankfully, by some miracle, I didn’t have too much homework this week. Or maybe since I was sleeping less and had on average 13 hours of working time on my hands each day the homework load seemed really light.
I also think that I’ve mastered the art of handling midnight AOE times as an East coast person. Thursday at 11:59pm in AOE time is equal to Friday at 7am in EST time, so the technique I’ve used in my past two paper deadlines is to sleep from ~2:30 AM – ~5:00 AM. This is the perfect timing because you get a long block of time until 2:30 AM, and then wake up with a fresh mind, sync up with your collaborators, and then do a final 1.5 hours of grammar checking and higher quality edits since you’re slightly less sleep-deprived.
On Friday after submitting at 7am, I slept from 7:00-8:30 since I didn’t have classes until 9:20 AM. I got through classes and sports with only one person, HD, remarking about me looking tired, which I think is a great endorsement of the 2:45am – 5:15am nap as opposed to a full all-nighter. I took Friday night off, watched movies with C, went to the Lunar Banquet, and got home to Easton around 10pm, where I worked on the arxiv version until 12:30am and then went to sleep. I slept in until noon on Saturday, and was decently efficient but didn’t get that much done because I woke up so late.
Looking Forward
While I lucked out and there wasn’t too much homework last week, this week is not so lucky. In the first two days of the week, I have to do a presentation on the Froude number, write a history essay, work on an english essay, and write my personal narrative for my college counselor. I’m pretty excited though, if I can stay as efficient as I was in the week before the ICML deadline, I should be able to handle everything. Contestwise, I skipped USACO this weekend because I just didn’t have enough time, and AIME is terrifyingly only two weeks away. I really should get back to doing my homework, I managed to write this post in 1 hour (which is very short by my standards), so there were also probably a lot of grammar errors.
I can’t wait for the long weekend next week when we get Monday and Tuesday off. The long weekend will be a great time to write out a full explanation of the paper, so my next post will hopefully be an accessible explanation of our model of distributed machine learning which is the first instance where the aggregation of local models is nonlinear. (The actual paper title is “Achieving Efficient Distributed Machine Learning Using a Novel Non-Linear Class of Aggregation Functions”)
