The content of this blog is my personal opinion only. Although I am an employee - currently of Imagination Technologies's MIPS group, in the past of other companies such as Intellectual Ventures, Intel, AMD, Motorola, and Gould - I reveal this only so that the reader may account for any possible bias I may have towards my employer's products. The statements I make here in no way represent my employer's position, nor am I authorized to speak on behalf of my employer. In fact, this posting may not even represent my personal opinion, since occasionally I play devil's advocate.

See http://docs.google.com/View?id=dcxddbtr_23cg5thdfj for photo credits.

Wednesday, August 16, 2017

Does Vision Research Drive Deep Learning Startups? | Chris Rowen | Pulse | LinkedIn

I think that I would like this conference format.  In preference to a multitrack conference where there are multiple papers I want to watch in the same timeslot.

Half of the reason to attend a conference in person is to observe the audience questions.

I really like the idea of poster sessions for all presented papers.

Does Vision Research Drive Deep Learning Startups? | Chris Rowen | Pulse | LinkedIn:

So many teams submit worthwhile papers [to CVPR]  that it has adopted a format to expose as many people to as many papers as possible. Roughly 2500 papers were submitted this year, of which only about 30% are accepted. Even so how can people absorb more than 750 papers? All the accepted papers get into poster sessions, and those sessions are unlike any poster session I’ve seen before. You often find two or three authors surrounded by a crowd of 25 or more, each explaining the gist of the talk over and over again for all comers. Some notable papers are given a chance to be shared a rapid-fire FOUR minute summary in one of many parallel tracks. And even smaller handful of papers gets much bigger exposure – a whole TWELVE minute slot, including Q&A!   
Remarkably the format does work – the short talks serve as useful teasers to draw people to the posters. A session of short talks gives a useful cross section of the key application problems and algorithmic methods across the whole field of computer vision. That broad sampling of the papers confirms the near-total domination of computer vision by deep learning methods. 

'via Blog this'

Sequences sunburst

Sequences sunburst:

'via Blog this'

I love data visualization. I love AI.

I want to love this visualization (of papers at the CVPR conference), but my wanna love is outweighed by my "I hate data viz that is too low bandwidth".

You can only "understand" this visualization interactively, by flying over the pie chart slices to see the labels. For somebody like me, who tends to absorb a whole picture all at once, this is much slower than presenting the piechart with labels or a legend.  Even if the legend is just color coded, because the labels don't fit.  But if the labels can be spatially attached, by proximity or by arrows, all the better.

I don't really have a photographic memory, but I have a cartoon-ographic spatial memory.  I remember pictures, usually highlighting the important areas. Sometimes my visual memory is augmented by time domain "zooming" into areas of interest.

This flyover graphic requires time domain sequential memory just for the first level absorption.  At the very least it is slower; but I suspect it also displaces the use of time domain memory for deeper understanding.

I am impelled to blog this because it is so bloody easy to make this visualization higher bandwidth.

It's a nested pie chart.  There is room for the labels for most slices of the innermost pie ring.  Even on the outermost pie ring.   I.e. without even changing the graphic most of the segments could be labelled, with interaction to dive into the segments too small for such trivial labelling.

And if you have the ability to explode sections - dynamically redraw - even more so.

I have probably made an enemy here, if this ever gets back to the author. (If you are the author, I would love talk to you.  If only to thank you for the raw data wrapped in this visualization.)

I have probably also dated myself, because video presentations are more and more the fashion.

Yes: I am the sort of person who hates watching videos, because I can read papers or slidesets faster than videos.   When I watch videos, I like fastplay and fast forward. I especially like video players that  recognize slide boundaries, so that I  can jump from slide to slide - and then only backtrack when it seems likely to have interesting discussion.

  • We can talk faster than we can type
  • But we can read faster than we can listen. Or watch video.

So, which is more important?