Sapere Aude
I ran the Chicago marathon some weeks ago, setting a new PB. However, I feel bittersweet about it. My primary goal was to run a sub 2:55 in order to qualify for Boston (at least on paper). However, I ran a 2:57:04 (previous best was 2:59:56). I will go into details soon. This is now my 4th marathon and I feel I’m starting to have enough data to compare to previous performances and get some nice statistics/plots. Let’s see how it goes!
My brother and I recently tried to transfer some data from MacOS to Windows using an external hard drive. Easy peasy, right? Simply copy the data MacOS -> HDD -> Windows. First step was smooth. However, after inserting HDD into Windows a popup appeared. It said more or less something like “The data on your HDD looks corrupted. Do you want Windows to fix it?”. Withouth thinking much about it, we were like yes, sure. Putting the only backup copy of your data in the hands of Windows: sure, what can go wrong?
Assume that we have a matrix $M$ where each row represents some observations from a random variable. How do we calculate the correlation matrix, i.e., a matrix where entry $i,j$ gives the correlation between row $i$ and row $j$? Note that this matrix is symmetric, since the correlation between row $i$ and row $j$ is the same if we invert the indices.
I was recently trying to profile a simple program in c++. I then found out about flamegraphs. I’ll write down here for my future reference the steps I followed to generate one.
Listening to random music today on Spotify, I encountered this:
I want to create a series of posts about music and interesting facts related to it.
I am creating this post because Neurips 2020 wants us to upload an image of our paper. The best image we could come up is the one of the regret bound of our algorithm :D
Given a real symmetric matrix $H \in \mathbb{R}^{n\times n}$, we can show that:
If not, check this!
I think so, but how to efficiently display it?
You should, because of this. Actually, that was the Woodbury formula. As a special case, we have the Sherman-Morrison update, which we here implement in Python:
Multi-armed bandit algorithms are becoming more and more important in the field of machine learning (at least to me, since I started a PhD on this topic :D). This funny name derives from the one-armed bandit, a name used for a lever operated slot machine (and apparently also for a Belgian rock album).
Variance? Whaaaat?