Roadmap Almost: VAEs

 Posted on November 17, 2019  |  2 minutes  |  338 words  |  John Chen

The score function estimator: a single sample, DOES estimate the score function etc. The actual point of reinforce: not taking the derivative through the expectation, but taking a derivative through a random variable. And the big kicker is that we take it through the RV, SO that we assume no dependence (i.e. an rv X sampled has no dependence on the probability distribution governing X); and so the problem resorts down to maximum likelihood estimation (essentially); just trying to max/min the probability of something happening; as weighted by the reward function The two views of taking a derivative through a random variable. [Read More]

Vector Calc Primer

 Posted on November 16, 2019  |  2 minutes  |  424 words  |  John Chen

Vector calculus generally means multivariate calculus. A function f(x,y) = z denotes a surface in 3-dimensional space. IF we FIX z to a specific value, then note that we just have two variables now! Then, we are looking at a 2D-CURVE. This is equivalent to “looking at the surface from the top”, or fixing the height to a specific value, and seeing which points on the surface correspond to that height. [Read More]

Goldman Sachs

 Posted on November 11, 2019  |  1 minutes  |  78 words  |  John Chen

Recently completed the GS technical assessment. It was not bad! I liked the programming bits (in that I got the questions, and they were straightforward), but the math section was pleasantly challenging and refreshing. The math section provided an opportunity to review and refresh lots of topics I have learnt throughout university and high school, including calculus (multivariable), combinatorics and counting, linear algebra, and statistics. In particular, it was fun reminiscing about Lagrange Multipliers, and all that! [Read More]

Common Programming Interview Tips

 Posted on November 11, 2019  |  1 minutes  |  202 words  |  John Chen

Here are some tips I have found to be quite useful while doing programming interviews: Have a whiteboard nearby (really helps to draw out your thoughts!) Here are some technical tips: 1. A common theme is iterating down the columns of a 2D list of lists (without using numpy!) 2. Graphs! Many problems are just waiting to be solved using graphs. But the translation from the problem into the graph has some flexibility (which affects the runtime, and everything really): 1. [Read More]

Set

 Posted on November 11, 2019  |  1 minutes  |  78 words  |  John Chen

Here’s a question: how can you do membership look-up in O(1) time in Python? One way is via a dictionary. But what if you don’t have values to store with the keys? Then, you don’t really need that extra stuff. In that case, let’s use a set! A Python set is like a mix between a list and dictionary: it only supports the ‘keys’, like a list, while still allowing membership checking in O(1) amortized, like a dictionary. [Read More]

Lyft Offices

 Posted on November 11, 2019  |  3 minutes  |  462 words  |  John Chen

As you may know, I recently received a Lyft offer for software engineer intern in machine learning for summer 2020. Just got off the phone with a Lyft engineering manager! He gave me some very exciting and interesting news and information about the New York Lyft office. New York Lyft office has several teams including Marketplace Labs, as well as several product teams, like: Pick-up (all the data that gets sent when a rider is picked up by a driver), (Routing? [Read More]

Unet

 Posted on November 10, 2019  |  3 minutes  |  497 words  |  John Chen

The Unet architecture is monuemental, and understanding it introduces you to so many different areas of deep learning. So I’ve decided to dedicate an entire blog post to it. U-net is an architecture for semantic segmentation. Semantic segmentation is the following task: given an image, classify each pixel as belonging to a class. It is like regular segmentation: while regular segmentation probably looks at pixel intensities and other intrinsic image vision properties (edges, gradients, coherence, continuity, etc. [Read More]

Hugo New Considered Harmful

 Posted on November 10, 2019  |  2 minutes  |  234 words  |  John Chen

I’m trying to like Hugo. I really am. But more and more it seems to me like the only reason this static site generator exists is because, “hey, why not?”. In other words, I am finding that the overhead of using the SSG is quite high, and interfering with my ability to do what I want, leading to me wondering why I am using it at all. (For reference: as someone coming from a CS background, with some web dev experience, I could trivially make a simple and functional blog by myself, so the value prop for Hugo really seems limited) [Read More]

Topics in Deep Learning

 Posted on November 9, 2019  |  1 minutes  |  50 words  |  John Chen

There are some new topics and muses for my study:

  1. Reinforcement learning (off-policy, on-policy)
  2. Embeddings; just generally getting accustomed to the lingo!
    1. ex. a “question” embedding, or some other use of the word “embedding”
  3. Graph neural networks
  4. Really want to concretize MCTS + Neural networks; i.e. the Alpha Zero framework

C3

 Posted on November 8, 2019  |  3 minutes  |  437 words  |  John Chen

So I had my 1st round final interviews for c3.ai today! It was an OK experience. The interview consisted of 3 portions, and I thought it was a pretty good interview process! First interview: Let’s say you are working for Expedia. How would you build a model to predict which trips are being taken for business and which are being taken for leisure? Second interview: Questions on domain knowledge of ML. [Read More]