News from Nullspace

Tim Cowlishaw's blog

About me | Twitter | Github

Weeknotes: 15th - 21st June 2015

Most of this week I spent trying to condense my thoughts on my general research direction into a more formal proposal that explains the motivation for my own work and its place in relation to the published literature.

As part of this, I revisited some of the papers that I’d encountered during my literature review which had most shaped my direction thus far - in particular Si├ón Lindley’s work on the relationship between users computing activities and their daily routines and habits. Going back through that paper, and the work it builds upon (in particular, Tye Rattenbury’s work on Plasticity), as well as Lindley’s later work on time and technology has helped me make a better case for the importance of activity tracking / life-logging to information retrieval. The idea of ‘plastic activities’ which expand and contract to fill the available time, and the way in which a user’s experience of the web can exist in multiple temporalities (for instance, when looking at a social media timeline, the difference between when a post was created and when it was viewed) both highlight the idea that users computing habits are tied up with the routines and rhythms of their everyday life, and that tools that assist with these habits need to model these rhythms effectively. I’m now pretty happy with the general direction my research is taking, and am going to spend the next week working on the specifics of a pilot experiment.

Otherwise:

  • I caught up with Marios and Jiyin about how we can all work together on the work that’s common to all our projects.
  • Started the process of setting up some office-wide informal interest-groups for both UCL and BBC folk with common interests, by creating and sending out a survey on everyone’s interests. I got plenty of interest in a Machine Learning group, and have organised a first meeting for the beginning of July.
  • Had a chat to some BBC people about a project I might help them out with over the next few months.
  • Listened to a few more episodes of the Talking Machines podcast, which has tons of interesting information and discussion. Highlights: David Blei talking about the role of Topic Modelling in the Humanities and a two part conversation between Geoffrey Hinton, Yoshua Bengio and Yann Lecun, as well as an interesting introduction to Determinantal Point Processes, which sound like they could be an interesting technique to use in my research as a way of optimising the diversity of a set of search results or recommendations.
  • Went to see Dan Deacon at Oval Space on Tuesday Evening

Interesting reading:

Music: This lovely live recording of Hauschka, Nick Cave and the Bad Seeds, Some Hyperdub Records stuff, Dan Deacon, Yellow Swans.

Our politics is utopian or it is nothing

It’s not often that I find anything Boris Johnson says to be particularly prescient or remarkable, but his denouncement of “Left-wing misery-guts” (within a risible bit of shilling for TTIP, with which I fervently disagree) struck a nerve.

Boris’s characterisation of us left-wingers as ‘misery-guts’ could be dismissed as a predictable bit of clichéd ad-hominem, where it not for the fact that, well, it can sometimes be true. And it’s not just the right who are pointing it out; Emma Burnell, writing for LabourList writes eloquently about how a Labour party which campaigns in the style of Harold Wilson’s ‘moral crusade’ might well alienate many of those it seeks to represent.

This isn’t just a question of public perception however, it is also, crucially, one of identity. The idea that being left-wing must necessarily mean being ascetic and humourless is directly opposed to my own experience of being of the left, and my reasons for identifying as so, and I’m certain that many of our comrades feel similarly!

One of my own favourite pieces of socialist writing, and one that was instrumental in my transition from seeing myself as a liberal social democrat to a socialist proper is William Morris’s News from Nowhere - a utopian sci-fi short story in which William Guest - an avatar of Morris - wakes up in a post-revoulutionary Marxist Britain, and sets out to learn all he can about this new society, in which toil, want, and war have been abolished. The only anger or disapproval Morris shows is directed towards iron bridges and other architectural carbuncles - his vision of the future society is one of unalloyed optimism. It was at this point that I realised that socialism isn’t about the miserly bean-counting task of allocating resources in the most equal way possible, but about the far more important struggle to ensure that every human has the right to happiness, health, leisure and creative fulfilment. This is echoed in Marx’s own analysis of the working day - which makes it clear that Marxism is, in the limit, the struggle for mastery over our own time.

News from Nowhere was written over a century ago; and over that time socialism has evolved both in substance, and perception, and not always in a way that has been desirable. It is obvious, I think, that as a movement, we could not but emerge from the century that gave us McCarthy, Stalin and Thatcher, unscathed. Much has changed over those 125 years, and our ideas and tactics must therefore change too. What is a constant, however, is the fact that, fundamentally, we’re fighting for the happiness, health and self-determination of every person, and that this is a fundamentally optimistic vision. When the right talks about aspiration, we must make it clear that this is not synonymous with acquisition, which is only a means to an end. To her credit, Liz Kendall partly makes this case, even though her statement that “most people want […] somewhere to live, something to do, something to look forward to and someone to love” fails to mention the freedom to spend our time as we please, and and to assert all these desires as universal rights. (It should be obvious that I fundamentally disagree with her proposed means of achieving this, too).

As socialists, we must continue to make this case, and do so in an optimistic, utopian way, and in so doing, reclaim the language of freedom, security and a good life for everyone as the preserve of the left.

Weeknotes: 8th - 14th June 2015

This week, I started work on the task logging software that’s going to be the apparatus for the first proper study of my research project. It’s a little bit of Windows software (for ease of recruiting participants in the study, primarily), I’m writing it in F#, and so far, it’s been an incredibly pleasant experience. F# is a great language, and the Visual Studio tooling is also very pleasant to use. I was particularly impressed with the ease of calling out to low level C APIs from within F# - it’s as easy as writing a type signature for the foreign function, and annotating it with the location of the .dll file with its definition. I’m doing the majority of this work directly in Visual Studio inside a VM (with VsVim without which I think I might have found the whole experience frustrating). However, I’d like to somehow set up VS to watch the filesystem for changes and build automatically on the VM, so that I can edit files and do the rest of my work natively on my machine. I still haven’t worked out how to do this though.

I had a productive weekly review with Emine, in which we looked over the draft research proposal I’m preparing. My direction looks broadly positive, but she advised me to start with a ‘big picture’ research question for the whole of my PhD and then drill down into specific experiments, rather than beginning with specifics and generalising out; my research proposal outlines a series of specific experiments, and while I think I’ve got a good idea of the direction they’re heading in, I’ve not been great at articulating it. This’ll be my first priority in the coming week.

Emine also alerted me that afternoon to a guest lecture by Abigail Sellen, whose work has featured prominently in my research so far. This was really interesting, and also very useful in helping me articulate my more general research ideas. I’ve written up a more complete set of notes on the lecture.

Jiyin and I had a further chat about organising a joint study, and I started writing up our application to the Ethics Committee - my first taste of University bureaucracy since my application, and definitely not the last!

Outside of work, I returned to Grasmere school to cover the code club I used to run while Frank, my replacement, had a week of. It was a really fun hour that went all too quickly. It was great to see all the kids and see how they’d progressed - Frank’s done some astonishing work with them in my absence. I also caught up over a beer or two with Sam and Bestie, with whom I used to work at Cambridge Healthcare, and finished the week with a couple of days in Derbyshire with Vicky and my parents - visiting family, running around Calke Park and visiting the National brewery centre.

Interesting links:

  • This episode of the Talking Machines podcast is great, with an interesting, intuitive explanation of the Dirichlet process, some thought-provoking debate about the role of human neurology as a model for machine learning, and a discussion with Zoubin Ghahramani about his “Automated Statistician” project.
  • An insightful analysis of the economic incentives for Facebook’s real names policy, and a call for decommodified digital social space which echoes my own thoughts on digital public space.
  • Zoe Williams, on fine, righteously angry form, writing about the need for a politics that isn’t ashamed of leisure.
  • A Marxian critique of the emergence of state-subsidised capitalism.
  • A fascinating critical comparison of the ‘Rich Kids of Instagram’ and 17th and 18th century painting traditions, and the role of conspicuous consumption in both.
  • A Henri Lefevbre reading list, which I’ve bookmarked for later use.
  • A critique of ‘ruin porn’, and the role of photography in illustrating social crises.

Music: Lawrence English’s newly re-released ‘The Peregrine’, Quietus editor John Doran’s spoken-word collaborations, Teeth of the Sea side-project Hirvikolari, Perc, The Wicker Man OST (RIP Christopher Lee) and Acid Mother’s Temple, who I’m also going to see playing in a ruined hospital on an island in a few weeks!

Designing Computer Systems that See - Notes on a lecture by Abigail Sellen

Abigail Sellen, from Microsoft Research, came to give a guest lecture at UCL yesterday. Sellen’s work has featured fairly prominently in my research so far, in particular her paper with Steve Whittaker investigating the use of lifelogging technology, and laying out some principles for the design of lifelogging systems, and her work with Sîan Lindley et. al. on the ways in which use of the web is situated within everyday life and routines. While the topic of her talk was ostensibly the design of systems with some computer vision component, a lot of the insights she shared from her experiences with this are very applicable to the design of intelligent systems more generally, and were very useful in helping me to examine and articulate some aspects of my own research interests. I’ll briefly share my notes on the lecture here.

The main theme of the talk was symbiosis in the design of computer systems, which Sellen described as a continuously-evolving process whereby human needs shape the design of a technology system, which in turn shapes the behaviour of the humans interacting with it, in response. Giving examples from her own work at MSR, she elaborated on how this symbiosis manifests itself in a system’s behaviour, as well as ways of designing for symbiosis in intelligent systems.

Principally, it seems to me that the main challenges for the design of such systems are ones of perception and intelligibility. The system must be able to perceive the relevant information it needs from its environment, and in turn, its output must be intelligible by the users interacting with it. This might seem like a simple point, but, as Sellen shows, both of these processes can be hampered in ways that are sometimes subtle. For instance, in a medical visualisation system controlled by gestures (for use in a sterile operating theatre environment), the gesture-detection algorithm was often be confused by two people dressed in identical scrubs, stood in front of each other. It was discovered by the MSR team developing this system that this problem was solved by enabling a feature originally intended for debugging, whereby the surgeon was given feedback, in the form of a stick-man-like graphic overlaid on the video feed, showing how the algorithm perceived the people in its environment. That way, they could quickly identify and remedy the source of any interference.

This use of perception-revealing feedback from the algorithm to the user was a recurring theme in the talk, and designing ways of visualising this feedback seems to be one of the central challenges. For instance, in a system designed to assist in the diagnosis of Multiple Sclerosis using computer vision, it was discovered that clinicians were mistrustful of the output of the algorithm, due to the fact that it’s decisions were opaque. Presenting a visualisation of the amount of tremor detected in the patient’s limbs over time (one of the principal features in the classification algorithm) to the clinician along with the automatic classification alleviated this, without the need to change the way the system worked internally - illustrating the reasons why an algorithm has arrived at a given output is more effective in making it intelligible to the user, than attempting to explain how it did so.

Another interesting example, which is especially relevant to my own interests, (and for which I can’t currently find a reference or link), concerned a mobile application allowing users to curate collections of real-life objects, by photographing and tagging them. The application attempted to automatically identify the photographed object, and the three most-likely classifications were presented to the user, encouraging them to select the appropriate one, or to provide a more accurate classification, from which the object classification algorithm would learn in future. Sellen pointed out that the misclassifications presented here also played a valuable role in revealing the system’s perception of the world to the user, enabling them to understand how the system arrived at its judgements. In this way, they functioned as a feedback mechanism in two ways - explicitly soliciting feedback on classifications, while also implicitly shaping use of the system by revealing the algorithm’s perception of the world.

Sellen also described another interesting consequence of this feedback - users were also curious about how the algorithm would perceive various objects, and would experiment with photographing things in order to see the suggested classifications. In this way, the misclassifications also afforded a kind of serendipity in the way the users themselves viewed the world and used the application - this is particularly interesting to me as I’ve been thinking about how active learning techniques such as uncertainty sampling (which by design present the most-uncertain classifications to the user) might also be used to model some sort of serendipity in information retrieval.

The correspondence between Sellen’s work and active learning techniques seems fairly natural to me, in a way that’s pretty fundamental. Sellen talked about how she sees the role of intelligent systems as being to work in partnership with humans, enabling rather than replacing them, and active learning is an obvious technique to facilitate this goal in general.

A couple of other shorter observations, noted here for posterity:

Sellen highlights the value in intelligent systems that perceive the world in a different way to humans, rather than emulating them. This is a useful thing to keep in mind. I can imagine many ways in which this difference could manifest itself however: As well as the obvious one of sensing different stimuli as input, this difference could manifest itself in terms of different ways of agglomerating this input into higher level features and output actions, different lengths and accuracies of recall of perceived stimuli over time, or simply presence in the world at a different time or space to the user. The central thing here is to look at both the needs and capabilities of the user and to design a system around both.

A question from the audience pointed out the correspondence between Sellen’s work and the HCI literature on Distributed Cognition. This is an area I’ve come across before and not examined fully yet, and so will definitely now do so as a priority.

Weeknotes: 1st - 7th June 2015

PhD

My efforts to organize my workload a bit better last week have gone, broadly speaking very well, however, I’ve not had as much time to dedicate to my work as I’d like this week as, along with the rest of my group, I’ve been spending much of my time marking Information Retrieval exam papers. This was an eye-opening exercise: both in terms of revising some basics of the subject with which I was a little rusty, and in terms of seeing first hand how others approach exams, and how they are marked.

In addition, I also wasted a little bit of time at the beginning of the week due to the fact that my Mendeley database managed to corrupt itself, losing a lot of the organization I’d done of my research paper library. This seemed like a good excuse to ditch Mendeley entirely (I’ve wanted to move away from using it since I realised they were owned by Elsevier, who are shits.) Thankfully, the open-source Zotero does everything I need (along with the Zotfile plugin) and so far is easier to use than Mendeley. It was also pretty painless to transfer my library across (With the exception of the folder structure I’d created, but this only took an hour or so of dragging and dropping in front of the TV one evening to recreate), and crucially, it offers a solution to the problem I described last week of keeping track of relationships between papers, as it allows you to ‘link’ related publications.

However, I did also found the time to review some of the literature on task identification, as well as beginning my research plan for the next few months. In advance of doin g this, I also did some revision from my research methods textbook, which was a useful refresher - it helped me to think more formally about the hypothesis I was testing, the dependent and independent variables, likely confounds that I would have to eliminate or control for, and possible approaches to analysiing the results. The whole process of writing this plan is helping me focus my thoughts in fact - by setting out my proposal as a series of concrete, testable hypotheses, I feel I’ve got a much clearer understanding of what I’m trying to achieve, and am much more able to articulate it.

I also met Marios, this week, he’s a masters student who’s going to be working with us on an activity tracking project related to my own.

Elsewhere

I also quickly rustled up a write-to-mp derivative to assist in the process of soliciting nominations for Jeremy Corbyn’s campaign for the Labour leadership. We’ve had 760 emails sent out in 5 days through the site, which I think is pretty respectable.

On Saturday Vicky and I took a wander around the E17 Art Trail - a great local initiative consisting of two weeks of exhibitions, events and open studios around Walthamstow. We hardly scratched the surface of everything on offer, but saw some fantastic work - in particular Owen Booth and David Southwell’s Stories from the Forest is a brilliant, erie psychogeographic exploration of Epping Forest, and Gavin Coyle makes some wonderfully beautiful and inspired furniture, and also gave us a really warm welcome to his workshop.

On Wednesday evening, we went to see The Chap at The Lexington, who were as archly witty, fun and interesting as ever.

Interesting links and reading

Listening

Had a listen to the new FFS tracks, enjoyed them greatly, and subsequently overdosed on older Sparks records.