I can’t believe it. I really can’t belive it…
There is no way it has already been one year since I managed to overcome my fears to pursue my dreams; one year since I took control of my own life; one year since I moved to London. But checking the calendar, there is no doubt: it really has been one year (and a few days, for the sake of accuracy) since I started my PhD at the Centre for Digital Music (C4DM) of Queen Mary University of London (QMUL).
It still feels like it was yesterday when I met Bob L. Sturm, my supervisor, for the first time. During this year, however, I suppose I have achieved quite a bit, both personally and professionally. Who knows… maybe commenting some of these achievements helps me assume this last year has really happened and, at its turn, paves the road for launching this long-delayed blog!
A large part of my efforts, or at least those that have led to “tangible” outcomes, have been devoted to conducting a Case Study. I know I still haven’t described what my research is about, but in rough terms I could say that the goal of my PhD is to devise ways to improve the validity, reliability and relevance of Music Information Retrieval (MIR) evaluation information. Quite unclear, I know. I promise I will explain it better in another post!
For now let’s say that with the case study we mainly attempt to address the relevance facet of my goal. Which information — beyond a bunch of performance metrics in a benchmark dataset — might help developers understand and improve their systems? For that purpose, we analysed in depth the scattering systems proposed by Andén and Mallat, which achieve high classification accuracies in GTZAN. We have presented different stages of our analysis in:
- ISMIR 2015 Late-Breaking Demo Session (Paper and Poster)
- DMRN+10 (Extended Abstract and Poster)
- ISMIR 2016 (Paper and Poster)
By performing system analysis and intervention experiments, we realised these systems exploit information below 20 Hz (inaudible for humans) to reproduce labels in the dataset. This information, which no one could guess by simply looking at the results reported in the original paper, will (probably) help the developers to improve the generalisability of their approach! For that reason, I presented our findings to the main author of the original paper in Princeton (Slides). In addition to that, I strongly believe this case study represents a magnificient example of how machine learning algorithms might exploit the most unexpected cues in our data to address certain tasks, so I also presented it at HORSE 2016 (Video and Slides).
Apart from this somewhat pragmatic (and visible) aspect of my research, I have also devoted large efforts to acquire expertise in the theoretical grounds my research is based on. In particular, I attended the course on Design of Experiments (MTH6116) lectured by my second supervisor, Hugo Maruri-Aguilar. This introduced me to a whole new field of research, and motivated me to investigate how evaluation/experimentation is conducted in other disciplines, such as medical or social sciences. For that purpose, I have read countless articles and books, and completed a handfull of online courses (which I hope I can review properly in devoted posts one of these days!).
All these efforts allowed me to successfully pass the Stages 0 and 1 of my PhD. But I’m no fool. I know now is when the real hard work starts!
Apart from the work directly related to my PhD, I have also spend some time in other tasks to train myself as the researcher I want to become. I peer-reviewed submissions to both ISMIR 2016 (for which I received the best reviewer award) and SMC 2016. I am also collaborating in the OpenMIC challenge of the COSMIR project. I expect to write quite a few posts discussing issues raised there. Finally, I have also worked as Teacher Assistant (now Student Demonstrator) for Digital Signal Processing (ECS602U, ECS707P), and Automata and Formal Languages (ECS421U) modules, and as Assistan Examiner for Introduction to Java Programming (EBU4201).
Well, now it really looks like a lot of work for one year, I guess…
Anyhow, I still feel I’m just starting… There is too much I want/need to learn! Hopefully, I will take advantage of this blog to organise, prioritise and consolidate all that!
PS: It’s happening!!! I’m publishing a blog post… FINALLY!!! I’m about to cry… :’)