Advait Sarkar

Project ideas for Part II / Part III / MPhil ACS

Here are some ideas for projects that could culminate in a final year dissertation. If you are a Part II / Part III / MPhil ACS student interested in carrying out any of the below, contact me. Some of these ideas involve substantial new research; if you are a researcher and are interested in collaborating on one of these ideas, please contact me.

  1. Data visualisation for mobile devices

    Pretty self-explanatory. Mobile devices are the dominant form of computing today. However, we still use data visualisations developed for the printed page (bar charts, line charts, histograms, etc). Can you develop a set of guidelines that helps developers of small-screen visualisations make best use of the space? Can you invent new, mobile-first data visualisations that are more conducive to small mobile screens, and with the advent of wearable devices, screens that have extremely limited graphical capabilities or are smaller than one inch?

    The ideal candidate for this project will have prior experience in graphics/UI programming.

  2. Intelligent visual LaTeX editor

    Let's face it, writing LaTeX sucks. You have to look up obscure commands and parameters all the time, and the source code text only bears a faint resemblance to the final document. There do exist some interactive visual editors such as Overleaf and LyX. However, they leave a lot to be desired in terms of usability and transparency. Can you think of ways to visualise and augment a standard text editor to make aspects of the LaTeX editing experience better? Your editor could include: code autocompletion/suggestion based on LaTeX language modelling, mixed source/WYSIWYG editing, contextual help, and simplified layout viewers.

    The ideal candidate for this project will have prior experience in graphics/UI programming, and a lot of experience writing different types of LaTeX documents.

  3. Rethinking the command line as a visual network

    We interact with various command lines every day, be it performing filesystem operations in bash, statistical manipulations in R, or data processing in a Python REPL. The command line experience is literally that — interacting with one line at a time. However, our mental model of a command line session contains a rich history of the things we are trying to achieve and the sequences of commands we have invoked in order to achieve them. This project will investigate visualising a command line session as a network of invoked commands, providing a REPL user with a better sense of context and bringing some of the advantages of direct manipulation (rapid, incremental, reversible actions) to command line interaction.

    The ideal candidate for this project will have prior experience in graphics/UI programming.

  4. The artificial pianist

    MIDI files often sound dull and lifeless because they do not contain the delicate imperfections in volume, timing and duration that a real performance has. Introducing random noise into MIDI signals improves things slightly, but it is still far off from what we might expect of a real performance. This project will involve modelling imperfections in a live-performed MIDI sequence by training various machine learning models such as neural networks on actual performances, and then applying these models to enliven a synthetic MIDI file.

    The ideal candidate for this project will have prior experience in MIDI programming and will be an experienced and competent musician.

  5. New visualisations for (big) music data

    We have a limited repertoire of data visualisations for music: traditional staff notation, piano roll notation, tablature, audio amplitude waveforms, and spectrograms. However, these visualisations focus very much on the small picture. How do we gain insights about large volumes of musical data? Can we visualise an artist's entire body of work and immediately identify recurring themes and outliers? This project will focus on building new visualisations for large datasets of music.

    The ideal candidate for this project will have prior experience in MIDI programming, digital signal processing, graphics programming, and will be an experienced and competent musician.

  6. The secrets to success in scientific publication

    What does it take to succeed in scientific publication? Previous research has shown that the truth is far from the ideal that a paper succeeds purely on the merit of its academic content. In the real world, the success of a paper is influenced by, among other things, how early it was published, how long it is, how catchy its title is, and where its authors are from. This project will involve analysing a large corpus of published papers and identifying factors that should not affect a paper's chance of success, but do. The project might culminate in the production of a hypothetical "perfect" paper that has been primed for success, based on all we learn.

    The ideal candidate for this project will be confident with machine learning and statistics.

© Advait Sarkar