We have just been informed that the paper “Evaluating Decision Makers over Selectively Labelled Data: A Causal Modelling Approach”, by Riku Laine, Antti Hyttinen, and myself, was chosen for the Best Paper award at the 23rd International Conference on Discovery Science (DS 2020).
We have a few new members in our research group: Yanhao Wang (PhD, NUS 2020) joins as postdoc, while Ananth Mahadevan and Sachith Pai (MSc, Aalto 2020) join as doctoral students.
Moreover, Francesco Fabbri, doctoral student at UPF Barcelona, has joined us for an internship.
Welcome to all!
These days I listen to podcasts quite regularly. Here are two that I’ve found particularly interesting:
The Talking Machines. (Link) This is a podcast about machine learning, which I heard of from Clemens. From a few episodes that I’ve listened to so far, their content falls under the following broad categories. (a) Description of a machine learning technique. This part is usually fairly technical. Not that they’ll have anyone recite complex formulas, but understanding them requires familiarity with machine learning. (b) Personal stories. During this part, a machine learning expert shares details of their own career, e.g., how they chose to do machine learning, what problems they like to work on and why, etc. This is the part that I like the most. (c) General discussion about the field – e.g., the impact on machine learning on society.
More Perfect. (Link) This is a new podcast about the United States Supreme Court. It’s a spinoff from Radiolab. Each episode discusses a case tried before the Supreme Court. Their content is quite accessible and does not require expertise in law. I like particularly how their discussion brings up the individual circumstances and personal aspects of each case; and how it blends with a discussion about why the case was important for an entire society.
Besides these two, I listen to the occasional episode by Freakonomics (example episode link), Radiolab (episode link) and sometimes New Yorker. I’m also eagerly looking forward Dan Carlin’s next episode. Regarding the latter, here’s another suggestion, if you like history: listen to his series on the Mongols.
Consider a circle; an equilateral triangle inscribed on it; and a chord drawn at random. What is the probability that the chord is longer than the side of the triangle? This question is known as Bertrand’s paradox – here’s the Wikipedia entry. I found it fascinating and spent some time to make this ipython notebook about it.
I take as a given that whatever data I share online will circulate among companies, partners, services, and other entities who take my privacy very seriously, but I’ve also come to expect that that happen in a discreet way that doesn’t make me aware of the fact. Well, that’s not always the case, apparently…
While texting someone a few days ago (June 11), I got this suggestion on my keyboard.
I typed “…leave the conference”.
I’m pretty sure I had not typed the phrase “leave the euro” before on my phone, so I don’t think the prediction was based on my typing patterns. However, I had allowed the keyboard app to parse text from a couple of personal online accounts — one was Gmail and the other one was Evernote, where I had saved a few articles about the euro crisis. So it’s possible that the prediction was based on text contained there.
In any case, I got curious about how the app made that prediction, so I started playing around: for a few days I would type the same text now and then, just to see what the suggestions would be. At first, I would get suggestions like the one below — the keyboard app predicted I wanted to type what I actually typed that first time.
Then on June 15, things got more interesting.
“Leave the… 60”? That didn’t make sense immediately, so I followed the keyboard’s suggestions to complete the sentence. Here is what I got.
As you might imagine, I was surprised and a little shocked that my smart keyboard would think I meant to say such a thing. I searched for the phrase on Google.
The predicted phrase appeared verbatim in the book you see above. Why would the keyboard predict that phrase? Here are a three clues that I think are relevant: First, I had bought that book on May 30th and added it on that same day on my Goodreads account. Second, I use Facebook to sign in Goodreads. Third, I also use Facebook to sign in the keyboard app. So, my guess is that using Facebook for login allowed the keyboard to match me with that book… Or something of that sort. Well, at least it didn’t give away any spoilers.