Wednesday, May 22, 2019

That's a wrap

Today was my last day of my Spanish class. I'm the only person that started at the beginning of the course last fall and continued uninterrupted through the year until now. Everyone in this picture:

started sometime later in the year. Moritz (German) is in front and across the back from left to right are Valentina (Italy), me (US), Marisa (Spain, teacher), Laura (Italy), Mike (UK), and Lisa (Canada). The class will continue for a few more weeks without me. I decided to stop now to focus a bit on my own work and to accommodate upcoming travel and visitors. I enjoyed the class. For me, it was a low-key way to get some practice with the language.


Laura and I went back to the Bulebar Cafe for another night of "Pint of Science". This evening someone from the University of Sevilla talked broadly about big data:

It was in pretty fast Spanish, so I was only getting about 20% of it. I did manage to get an IPA from the US. It was pretty good and only 3.50 euros. The availability of different kinds of beer here in southern Spain is pretty limited, so you have to strike when you can:


In Algorithms to Live By the authors talked about "overfitting". I've been impressed with their ability to map these ideas from data science to real life. In this case, overfitting is the problem of tracking the known data so closely that you are unable to handle new situations. IRL this is equivalent to developing a strong habit or reaction to a set of circumstances and then reacting poorly to a similar, but in some way novel situation. While there are many such examples, the extreme example they provided was in the training of FBI agents. In training, a common lesson they practice is taking a gun away from an assailant. In training the student may do this exercise many times in a row and at the end of each round they hand the weapon right back to the instructor. In one instance, this habit became so ingrained that an agent, after disarming an actual assailant, immediately handed the weapon back to the assailant--a distressing case of overfitting.