I interned at DoorDash in San Francisco this fall and also did a bit of traveling.
Interning at DoorDash, the market leader in food delivery, was so exciting. For my first project, I served a new carousel on our feed page, which took about six weeks. The carousel retrieved the user’s most recent order and recommended similar stores to the user. I created the entire system end-to-end, from the developing the machine learning algorithm and creating an ETL, to reading from recommendations from our search service database, and to finally serving the carousel from the feed service. Every week, I felt like I was leveling up.
One important lesson that I learned is that deep learning is not always the best solution. I spent about a week working on how to tag items with their food types. At first I came up with a deep learning solution using BERT and new pre-training tasks, but after manually evaluating a couple hundred samples of data myself, I realized that a rule-based approach is most appropriate.
Left, my desk at DoorDash; middle, not my trophy; right, the carousel I made.
Being back in the city again was fine, except this time, many of my Dartmouth and Blend friends, who were here over the summer, went back to school. From Sep 27 - Oct 31, I lived in the Winsor Hotel, a one-star temporary housing unit just one block away from the DoorDash office. The price was good for my own room with a full-sized bed, but the location was not so great.
Later, I moved back to the student homestay in Chinatown where I lived for the summer.
- Office visits: Flexport, Accenture @ Salesforce Tower; Robinhood, Thumbtack, Affirm, Palantir
- Bars: PianoFight
Left, my room at the Winsor Hotel; middle, 6th street; right, the Tenderloin.
My data augmentation paper was accepted to EMNLP, and so I and Chengyu, a co-worker from Protago Labs went to Hong Kong from November 3-7. The first day, I went to Deep Learning for NLP and Conversational AI tutorials, which were good. The second day, I went into the city to see the Tsim Tsa Tsui Promenade, the glamorous K11 mall, and Victoria’s Peak.
Some lessons that I learned from the conference:
- Some papers are marginal. They don’t take big risks or advance the field substantially but instead demonstrate small improvements and are technically sound, so reviewers often can’t find good reasons to reject them. Don’t work on these marginal papers. Dream bigger and take a risk. Even if you get rejected, you won’t regret it.
- Not all papers have to be on methodological innovations. [Example 1] [Example 2].
- There are so many unsolved problems in NLP, and many papers introduced new unsolved tasks for us to pursue.
On the third day of the conference, my friend June and I took a cable car across the island to see the Tian Tan Buddha. I loved the views from the cable car, and the Hall of One-Thousand Buddhas was beautiful. That evening, we went to Hong Kong Disneyland, for which the conference had rented out Tomorrowland exclusively for EMNLP participants. We enjoyed the buffet dinner and went on the Star Wars, Iron Man, Ant Man, and Small World rides.
The last day, I presented my poster during the last session. It seemed, however, that there was less interest than when I had previously presented at ICLR. I had also added a sign saying I was a prospective PhD student, but unfortunately no professors came to talk to me.
Throughout the conference, I frequently went to the hot tub and sauna at our hotel, which was a great place for me to get away from all the sounds of the city and spend some time alone with my thoughts.
Left, Victoria Peak; middle, Hong Kong disneyland with Chengyu; right, the Big Buddha.
I went to Vancouver for one day to attend the Machine Learning for Health Workshop at NeurIPS. It was pretty surreal seeing my brother, a high school student, present his work to over 500 people as a spotlight talk. The recording is here (starts at 56:03). I’m so proud of him.
I didn’t have enough time (i.e., wasn’t self-motivated enough) to do any projects by myself, but my brother’s paper was accepted to the machine learning for health workshop at NeurIPS.
Generative Image Translation for Data Augmentation in Colorectal Histopathology Images
Jerry Wei, Arief Suriawinata, Louis Vaickus, Bing Ren, Xiaoying Liu, Jason Wei, Saeed Hassanpour
NeurIPS ML4H Workshop 2019
Moving forward, I will try to see research as a creative art rather than a technical exercise, and approach it with an attitude of fun.
Health & Fitness
I originally had hoped to hit 2 plates on the bench press by the end of the year, but I didn’t even come close. I worked out at 24-hour fitness, sometimes with my friend Ashar, but was set back by various things.
Meniscectomy. Last December, I had hurt my knee from squatting, which led to an IT band injury. After playing tennis while I was home, I felt more pain, so I went to see James Chen, an orthopaedic surgeon in SF. I got a knee MRI and Dr. Chen told me I had a sizable tear in my meniscus. On November 22, I had a partial meniscectomy, a simple surgery. The recovery, however, was not as easy as I thought it would be. I bought an ice therapy unit, which I used three times a day for five days, and worked from home for the next week. After exactly two weeks on crutches and some quad-strengthening exercises, I was able to walk again and was feeling better. I was not very productive throughout the whole process, and I realized that I take good mobility for granted.
Nose surgeries. I saw Dr. Jacob Johnson, an ENT in SF. I had a turbinate reduction, a small procedure to help with my breathing/snoring. So far, it hasn’t felt like anything changed. I had also scheduled a non-invasive balloon septoplasty for the last week, but I ended up getting sick so we couldn’t do the procedure.
Getting sick. I got sick with a runny nose, sore throat, and cough about ten days after my surgery, which turned out to be Moraxella catarrhalis. I felt a bit better after taking Augmentin antibiotics, but then after I worked out hard at the gym, my cough came back even worse and I still don’t feel well.
Left, pre-surgery; middle, post-surgery; right, meniscus tear.
The Last Lecture. Randy Pausch, a computer science professor at CMU, gave a “last lecture” after being diagnosed with pancreatic cancer. I admire Pausch’s communication so much. Here are some of the pieces from his book and their takeaways (Chapter #):
- Pouring soda in the backseat (15). People are more important than things.
- I’m on my honeymoon, but if you need me… (23). Time needs to be managed, like money.
- Dream big (28). Pausch grew up watching man land on the moon.
- Don’t complain, just work harder (32).
- Be the first penguin (39). Take big risks.
- The lost art of thank-you notes (41).
- The friday night solution (43). Q. What was your secret to getting tenure early? A. “It’s pretty simple, call me any Friday night in my office at ten o’clock and I’ll tell you.”
- Tell the truth (48). All the time.
- No job is beneath you (51).
- All you have to do is ask (55).
- Make a decision: Tigger or Eeyore (56). “I’m dying and I’m having fun… there’s no other way to play it.”
His final remarks, which he addressed to his family, touched me dearly.
Everything is F*cked: A Book about Hope. I was excited to read this book because I liked The Subtle Art of Not Giving a F*ck so much. I had a hard time getting through the first half, which talked about this struggle between your feeling brain and your thinking brain (i.e., doing what feels good versus what you know is good in the long run). In the second half of the book, however, Manson brought to light many important concepts.
- The Uncomfortable Truth. You will one day die, and beyond a small group of people for an extremely brief amount of time, nothing you do matters.
- The Formula of Humanity. Manson loves Immanuel Kant, who says that the supreme value in the universe is the thing that conceives of value itself (i.e., the humanity is the greatest value). And Manson’s formula of humanity therefore states, “Act that you use humanity, whether in your own person or in the person of any other, always at the same time as an end, never merely as a means.” Do something because it’s the right thing to do, not beacuse of the outcome.
- Meditation. Manson’s account of Quang Duc moved me, and he describes meditation as “observing the interiors of your mind and heart, in all their horror and glory.”
- Adulthood versus Adolescence. While adolescents use principles as a means to get pleasure, adults value principles as ends in and of themselves. Manson defines adulthood as “the realization that sometimes an abstract principle is right and good for it’s own sake, that even if it hurts you today, even if it hurts other, being honest is still the right thing to do.”
I plan to read this book again in the future and re-evaluate it.
Who are my heroes?
- Randy Pausch says, “Tell the truth. All the time.”
- Mark Manson has a law of F*ck Yes or No, which says that if you’re in the grey area, you’ve already lost (i.e., if you have to think about it, the answer is no).
- I loved Lady Gaga in A Star Is Born.
- Julianne Moore is great in Kingsman: The Golden Circle.
- Most tennis players have an idol that they want to play like (usually Federer, but I like Djokovic more). Well, I code, and my coding idol is my mentor from Blend, Danny Hermes.
- uuuuuuuukewithme’s singing and persona are so great.
- Paul Graham writes excellent blog posts.
This term, I learned a lot about software at DoorDash, did a bit of traveling, and got through a surgery. Compared with other terms, however, I did not do as much. And even more importantly, I didn’t give recruiting or PhD applications their due diligence in effort. I also didn’t get to spend any time on research. I really dropped the ball.
I liked this Paul Graham quote:
The most dangerous liars can be the kids’ own parents. If you take a boring job to give your family a high standard of living, as so many people do, you risk infecting your kids with the idea that work is boring. Maybe it would be better for kids in this one case if parents were not so unselfish. A parent who set an example of loving their work might help their kids more than an expensive house.
You probably think it’s silly for me to post the details of job offers that I got and didn’t get, but I don’t care. I’m an advocate for transparency, and new grad offers are usually pretty standard, and so here it is:
|Company||Salary||Signing Bonus||Stock Options||Offer Date||Deadline||My prediction|
|Blend||$125k||$20k||55k shares/4 yrs||Sep 6||Nov 5||Unsure|
|Thumbtack||$125k||$15k||~$100k RSU/4 yrs||Oct 24||Nov 15||No|
|Company||Reject Date||My prediction||Interviews|
|Neeva||Aug 27||Unsure||Coding challenge, onsite in Mountain View|
|Figma||Sep 17||Unsure||Phone screen, phone interview|
|Benchling||Sep 20||Yes||Phone screen|
|Coda||Sep 24||No||Coding challenge, phone screen, onsite in Bellevue|
|Robinhood||Oct 8||Yes||Coding challenge, phone screen, onsite in Menlo Park|
|Affirm||Nov 5||Yes||Two phone screens, onsite in SF|
|Palantir||Dec 16||No||Phone screen, onsite in Palo Alto|
While we’re here, I am working on being more comfortable with rejection:
I want to visit Cappadocia, Turkey.