Today was, again, quite busy. However, it didn’t seem nearly as overwhelming as yesterday. I got through a good portion of the CITI training this morning and I hope to finish up tomorrow morning. It isn’t as time-consuming as I was anticipating and I feel that I am learning a great deal. I especially liked the modules on authorship and mentoring because I was kind of confused on how to approach both of those subjects.

During the late afternoon, we learned more about qualitative analysis and needs assessments. Though often overlooked, these processes are very important for identifying needs in a community so that technology can be developed and actually used. I think that I would be good at interviewing people and coding the transcripts to find overarching themes. The only thing that I am nervous about with some of the projects involving qualitative analysis is that I would spend the majority of my summer trying to recruit participants since the graduate students are in the beginning phase of their research.

Ben Jelen also talked to us more about related work. We were told to read an article by one of the faculty mentors and write another 5 sentence summary on it. I am very interested in Dr. Natarajan’s work in machine learning, so I looked up some of the papers he has authored. Work involving statistics is a lot of fun for me; AP Statistics was one of my favorite classes in high school. My interest in machine learning, in particular, was piqued this spring when I created a hangman AI using a Naive Bayes model. When I saw that Dr. Natarajan created a new version of the Markov Model, I was hooked. The citation for the article and the 5-sentance summary appear below.

Natarajan, Sriraam, Hung H. Bui, Prasad Tadepalli, Kristian Kersting, and Weng-Keen Wong. 2008. “Logical Hierarchical Hidden Markov Models for Modeling User Activities.” In Inductive Logic Programming, 192–209. Berlin, Heidelberg: Springer Berlin Heidelberg. doi:10.1007/978-3-540-85928-4_17.

This paper details the development of Logical Hierarchical Hidden Markov Models (LoHiHMMs), which doesn’t have to be completely unrolled (conserving space) and involve particle filtering (conserving time). Researchers compared the effectiveness of the LoHiHMM with the normal HMM over the course of 50 runs of 2 simulated domains: the Doorman Domain (which involved a gridworld ) and the Kitchen Domain (a more real-world scenario). The comparison showed that, with the particle filter, the LoHiHMM was only slightly less accurate than a normal HMM and didn’t require nearly as many ground states to make inferences. By incorporating hierarchies and allowing conditional transitions,  LoHiHMM has the potential to become an extremely efficient learning algorithm. I am interested to see how LoHiHMM performs in real-world domains, which is where this study plans to focus next.

To learn more about Markov Models and machine learning, I plan to read:

•  Introduction to Statistical Relational Learning
• “The Hierarchical Hidden Markov Model: Analysis and Applications”
• “Logical Hidden Markov Models”

After lunch, we played with Arduino. I worked with Gabby since my computer is still being finicky about connecting to Lilypad. The coding for this device was really simple and we were able to do some neat things with the LED lights, buzzer, a light detector, and a sound controller. During class, we copied in the music for the Harry Potter theme and made a program that would change which LED was on depending on the frequency of the note. The program activated when the light sensor didn’t detect any light.You can watch a recording of that here.

Gabby working on the Final Jeopardy! code

For homework, we were tasked with doing something cool with the Arduino involving music, input, and output. We coded the Final Jeopardy!  theme, which was tedious because we had to input each note’s frequency and duration. We then created a pattern for which the LEDs to light up corresponding to the notes played. This one looks a lot cooler than the Harry Potter version. Like our previous program, this turns on when the lights turn off. I feel that this could be useful when someone needs to make a difficult decision because they will focus on the lights and the music will create a sense of urgency. If you are interested, you can watch a video of the project or read the Press_Release I wrote up.

At the end of the day, Fernando gave his project talk. He is working online to do needs assessments and product testing for people with HIV. I think he is doing really good work with populations who are vulnerable and isolated due to their health conditions.

Unknown Animal of the Day:  Pink Fairy Armadillo

This Argentinean desert-dweller is the smallest armadillo species. Both its shell and thin fur helps with regulating its body temperature.

Conservation Status – Data Deficient, but quickly declining in number due to expanding suburbs