Table of Contents
Following previous month’s summary and resolution, here is the February update.
First, a picture from a trip to a 2-star restaurant, Momofuku Ko.
Delicious bread from Momofuku Ko
If you had a cursory look at this website beforehand you’ll know I’m very much into food. I’ve started reading Modernist Cuisine this month and will give you a quote from the author Nathan Myrvold from the foreword to the book. On one hand it is motivating, but on the other hand you’ll be sure to question your skills and whether whatever you are doing has any value compared to people such as Nathan.
My interest in cooking was so strong that I might have become a chef, had my interest in other things-particularly math and science-not intervened. I was very good at school and often skipped grades, to the point that I started college at 14. Every topic related to math and science fascinated me, so by the time I was finished with school, I had a quite a collection of degrees: a Ph.D. in mathematical physics, a master’s degree in economics, another master’s degree in geophysics and space physics, and a bachelor’s degree in mathematics. By that point I was 23 years old. My next step was to become a postdoctoral fellow at Cambridge University, where I worked with Dr. Stephen Hawking on the quantum theory of gravitation. My career in science was off to a roaring start.
Starting university at 14, a whole array of degrees complete with a PhD by 23, research work with Stephen Hawking, CTO of Microsoft, now a food researcher, and I’m sure a multitude of achievements in between. Let that sink in.
Now for the updates.
New Content
Programming
I moved from Div2 (the noob/beginner division) to Div1 (advanced) in Topcoder (1182 -> 1257 from last SRM competition). I wasn’t sure whether to include this here as Blue (1200-1499 points) is still a low score for someone who writes code (and hence problem solves) professionally. For those that are not familiar with it, Topcoder is an online coding competition. You get three problems of increasing difficulty and 75 minutes to solve them. You have to solve them exactly right or you get no credit and drop in ranking. There are two divisions as explained above, Div2 and Div1.
It is really discouraging many times to read the Div2 Hard problem description, scratch your head for thirty minutes, come up with nothing that would reasonably pass all the tests, or nothing that you can reasonably implement 100% correctly in the remaining time, and then question your abilities. The reason why it took me so long (more than twenty rated competitions) is in hindsight obvious. I never did any practice outside of the 75 minute competitions, so obviously every single round went the same: solve the first two problems, fail on the third. Upon seeing a problem I couldn’t solve within around 30 minutes, I would give up, and go to something easier, whereas all anecdotal evidence I’ve heard says to stick for hours to one problem if you can’t solve it, and try to solve problems above your current skill level to grow.
A great way to never improve is to just do the same thing over and over, which is exactly what I was doing, and a trap I still fall for very often. If there is one thing I could have changed with my time at university, it would be to compete and properly train in competitive programming, ideally in a dedicated class and/or a team where I feel progress and motivation would be optimized for. I’ve seen it first hand how being good at this pretty much guarantees you will pass any software engineering interview, making it one less thing to worry about.
Some have asked me why I do these competitions and exercises. A couple of reasons. One, it’s not that I even want to get much better at problem solving (I do), but I don’t want to atrophy. It’s really a terrible feeling seeing a problem that you know you could solved when you were much younger, but your skills have atrophied. Two, I feel I suffer from Impostor syndrome and quantifying my problem solving ability with a rank number gives me some empirical evidence to suggest I’m not a impostor. Three, it’s a fantastic feeling solving a problem correctly that you struggled with a lot. If you do weightlifting, it’s like hitting a new PR for your squat.
During the rest of this year, I will set my goal to reach Yellow (1500-2199) in TC via dedicated training. If you have tips for me, want to share your experiences, or even train together, let me know. Some pitfalls I’m foreseeing that might prevent me from reaching said goal, in order of probability:
- My interests are varied and I might just not care enough to make it happen, in relation to other interests. I might want the end goal (Yellow), but might not fully understand the dedication and work to reach it and hence quit the goal.
- I might care enough and spend the required time to get to the goal, but mis-appropriate it by training on problems on my current skill level and hence not growing. Back where I started.
- I might somehow convince myself this whole thing is useless and stop altogether in exchange for another activity.
Reading
Total: 5
2016 total: 15 (goal 52)
- Crucial Conversations: Tools for Talking When Stakes are High (235 pages) (3/5) (Finished on Feb 28 2016)
- Momofuku (304 pages) (3/5) (Finished on Feb 27 2016)
- Stay Hungry (141 pages) (4/5) (Finished on Feb 8 2016)
- Why We Get Fat: And What to Do About It (272 pages) (2/5) (Finished on Feb 7 2016)
- The Technological Singularity (272 pages) (5/5) (Finished on Feb 2 2016)
Technical Papers Read
- Practical Bayesian Optimization of Machine Learning Algorithms
- Not easy to read and I don’t fully understand it, limited by math knowledge. The gist of it: Many machine learning algorithms rely on good hyperparameter selection to work well (for example k in k-nearest-neighbours or regularization parameters in various regression methods). The usual way is to do something brute-force, such as grid-search your way through the various parameter values, do cross-validation and figure out which set of hyperparameter values is the best. Obviously this is slow. This paper goes to show how you can use Bayesian reasoning to create a prior for the space of hyperparameter values using Gaussian Processes, and hence ‘lead’ the hyperparameter search in the right direction. A further contribution is showing how you can parallelize this - normally, for time
t, you would need your posterior computed for timet-1, so this doesn’t look very parallelize able. The paper shows how you can evaluate the acquisition function under all possible outcomes of currently pending evaluations to parallelize this.
- Not easy to read and I don’t fully understand it, limited by math knowledge. The gist of it: Many machine learning algorithms rely on good hyperparameter selection to work well (for example k in k-nearest-neighbours or regularization parameters in various regression methods). The usual way is to do something brute-force, such as grid-search your way through the various parameter values, do cross-validation and figure out which set of hyperparameter values is the best. Obviously this is slow. This paper goes to show how you can use Bayesian reasoning to create a prior for the space of hyperparameter values using Gaussian Processes, and hence ‘lead’ the hyperparameter search in the right direction. A further contribution is showing how you can parallelize this - normally, for time
Interesting articles read and notable links
- I no longer understand my PhD
- I’m not sure why but when I read that I just thought it was incredibly hilarious and laughed out loud. I imagined someone spending years of their life writing up this document, their PhD on functional analysis, looking at it couple years later, and then proclaiming “The notation was alien. I even had to scour the examiner’s report to direct me to the key results.”. I can only imagine the level of disappointment for the author when coming back to something he spent a good chunk of adult life on, and having trouble understanding what he has done and how. There are good arguments for why this is ok and why it’s not ok, many of them in the HN comments here. I see where the author is coming from when he says “My research was in an area of Pure Mathematics called Functional Analysis which, in short, meant it was self-motivated and void of tangible real-world application.”. I claim that for some curious and smart people, a tangible or useful result is completely irrelevant to the work they are doing, they just need an interesting challenge and a PhD in pure math presents such a challenge. Some of the most captivating things I’ve done, things that put me in Flow the most and presented themselves with a high intrinsic reward are things that had nearly no extrinsic value and were just an intellectual challenging exercise. Examples include:
- Reading some technical paper and implementing what it presents and then seeing it work (example - Sudoku Solving with Dancing Links; it has no extrinsic value as there a number of other implementations already)
- Solving a tricky programming challenge (which others have already solved), hitting the submit button and seeing your solution approved
- Readings textbooks for subjects I didn’t take and I had almost zero practical need to know about.
- I’m not sure why but when I read that I just thought it was incredibly hilarious and laughed out loud. I imagined someone spending years of their life writing up this document, their PhD on functional analysis, looking at it couple years later, and then proclaiming “The notation was alien. I even had to scour the examiner’s report to direct me to the key results.”. I can only imagine the level of disappointment for the author when coming back to something he spent a good chunk of adult life on, and having trouble understanding what he has done and how. There are good arguments for why this is ok and why it’s not ok, many of them in the HN comments here. I see where the author is coming from when he says “My research was in an area of Pure Mathematics called Functional Analysis which, in short, meant it was self-motivated and void of tangible real-world application.”. I claim that for some curious and smart people, a tangible or useful result is completely irrelevant to the work they are doing, they just need an interesting challenge and a PhD in pure math presents such a challenge. Some of the most captivating things I’ve done, things that put me in Flow the most and presented themselves with a high intrinsic reward are things that had nearly no extrinsic value and were just an intellectual challenging exercise. Examples include:
- FBI and CMU Tor Attack Collaboration
- It’s confirmed that Carnegie Mellon University collaborated with the FBI in a sybil attack combined with a timing attack on the Tor network to deanonymize users of Silk Road 2. It’s disheartening to see university security researchers using their knowledge to put people in prison. To elaborate, even though this attack was on Silk Road 2.0, we can’t be certain it wasn’t used (or that they aren’t working on new exploits now) to deanonymize others who for a number of reasons (such as an oppressive country they reside in), rely on the expected privacy and anonymity guarantees the Tor service is supposed to provide.
- “Relay early” confirmation attack (technical details of above)
- Technical details of the above attack.
- VICE Interview with Martin Shkreli
- I see a lot of people online demonizing Martin Shkreli for the price hikes he introduced to an HIV drug he bought the rights to. It’s always better to get the view from both sides, and this short interview was helpful in that. Watch and form your own opinion. Martin’s points are that all the big pharmaceutial companies are doing it and no one says anything there. He claims their percentage allocation of revenue to research is one of the highest in industry, and further claims that if anyone can’t actually afford the drug, he will it give it to them for free if they contact him. From that video Martin definitely gives off a weird vibe, but maybe that’s one of the reasons why he is very successful, and that you cannot deny. Currently age 32, he has co-founded a hedge fund, a biotech company and another pharmaceutical company, and has a net worth of around $100M.
- When you Burn off your fat, where does it go?
- The way we lose fat is by exhaling it as carbon dioxide.
- Haskell: Introduction to Lenses
- A nice six-part introduction to what lenses in Haskell are.
Movies watched
- Incendies (February 29th 2016) (139 min)
- If it has an 8.2 on IMDB then it’s probably worth watching. The story was well told, captivating, but also emotionally draining. It’s not a ‘feel-good’ movie, so leave it for a day when you are more cheerful than average.
- Jiro Dreams of Sushi (February 27th 2016) (81 min)
- To dedicate your whole life to perfecting one thing, in this case sushi making, is something. Jiro calls himself a ‘shokunin’, an artisan, someone who will repeat the same thing over and over every day, improving all the time. He is 90 years old now, and still goes to his restaurant every day to prepare sushi and hates going on holidays. He is still planning to be in the restaurant at least until the Tokyo Olympics in 2020, so I still have four years to visit.
- Sideways (February 20th 2016) (126 min)
- Cloud Atlas (February 13th 2016) (172 min)
- Julie & Julia (February 12th 2016) (123 min)
- It was good, it could have been better, but I can’t exactly figure out how to improve it. I do find the storyline impressive. Setting seemingly insurmountable goals, such as cooking every single of the more than hundred recipes from Julia Child’s cookbook, makes the main character grow a lot and opens a bunch of doors for her.
- Headhunters (February 10th 2016) (100 min)
- A fun action movie, you need one from time to time.
- Seven Years in Tibet (February 6th 2016) (136 min)
- The Lobster (February 1st 2016) (118 min)
- ‘The Lobster’ takes place in a dystopian future where every male and female must be matched up in a relationship and no one can live alone. If for any reason you lose your partner (for example they die) you are sent to a holding hotel where within forty-five days you must find a partner. If you don’t find a partner you will be turned into an animal of your choice on the last day of your stay. Yes, the premise of the movie is very weird and the delivery is even weirder. This is the second movie I’ve seen by Yorgos Lanthimos, the other being the just-as-weird Dogtooth. The director has a very peculiar style. It just slightly reminds me of Wes Anderson, except more dry, slower and much more difficult to derive enjoyment from. The narrative structure and a frequent reuse of same thematic thud also reminds me of Gaspar Noe’s I Stand Alone. I didn’t enjoy Dogtooth and I can’t say that I enjoyed this either but I can imagine there are people that will like this particular style. The concept is interesting and I am a big fan of dystopian movies, but this just didn’t do it for me. I will give 6/10 for the interesting first part of the movie.
Arts
- Adam Gyorgy plays Liszt - Carnegie Hall
- February 21st 2016, New York City, NY
- My first time hearing live a couple of pieces I like by Liszt. The selection included Hungarian Rhapsody No. 2 and La Campanella. Definitely great and much different to hearing it on your headphones.
- Budapest Festival Orchestra - Carnegie Hall
- February 18th 2016, New York City, NY
- I have been in NYC for more than a year now and embarrassingly, this was the first time I went to Carnegie Hall (While in London, I regularly went to Royal Albert Hall, I didn’t keep track but in 4 years it was no less than 25 times. One performance, “Classical Spectacular”, I liked so much I saw it a total of five times). The Budapest Festival Orchestra was a great first time at Carnegie Hall. I got a front row seat, unfortunately slightly more to the right so didn’t have a great view of the pianist during the piano concerto but otherwise it was great and I highly recommend you sit at least once in the front row to get a really good look at the string instruments.
Restaurants
+3 Michelin Star count (total 52)
- Momofuku Ko
- February 27th 2016, New York City, NY, 2 stars, ($244pp spent)
- Gramercy Tavern
- February 15th 2016, New York City, NY, 1 star ($80pp spent)
Conclusion
It was a slower month. I don’t really have an excuse except laziness. I’m currently on a cutting phase. A dietary restriction of 500 calories (down to 1800 from 2300 per day) together with a most uninteresting set of things to eat (poached chicken, rice, vegetables, some cheat meals from time to time) has some noticeable negative impact on my mental capacity. I’m still trying to figure out what’s the cost-benefit of weight training taking into account it’s a very demanding activity if you want some results rather than just maintain your current physique.
I have some more thoughts to add on weight training and in general about exercise. My claim is that, if someone cares about weight lifting, or running or swimming or whatever form of exercise, their performance there will be positively correlated with their performance in other parts of life such as professional and personal successes. Any form of consistent exercise is demanding in terms of motivation, dedication (it takes a long time to achieve anything), time management, goal setting, planning, goal tracking, and importantly, correct execution. All these traits are applicable and reusable in other contexts. You will find examples of highly successful people that care about some form of exercise and also are succesful there. What I claimed above also implies the following: If you do decide to care about exercise and don’t see marked progress over time, it’s likely whatever is causing it (be it laziness, lack of commitment, inability to set goals) will transfer to all other areas of life until you consciously decide to fix it. Feel free to express your opinion on this in the comments.