Videos: Two Minute Papers
Excellent video summaries of newly published AI research
The bleeding edge of machine learning is revealed by research papers far before you see applications in the real world. But knowing the best publishers and where to find the research is barely half the battle. The real challenge is distilling any study (many of which are dense in mathematical notation and AI terminology) down to the takeaways.
Károly Zsolnai-Fehér is a research scientist in Vienna, and his ‘Two Minute Papers’ do this for you. In his words:
“In this series, I try to explain the intuition behind exciting research results in an understandable way, a couple minutes at a time. We are mostly exploring the area of computer graphics and machine learning with a hint of philosophy and physics.”
He also starts each video with his classic greeting “Dear Fellow Scholars…” that makes you feel like you’re about to watch an episode of the Magic School Bus. Odd at first, but I’ve come to find it rather endearing.
How I’m Using Two Minute Papers (TMP)
A. Discovery: Subscribing to the YouTube channel surfaces compelling papers frequently that I may not have come across ordinarily. Károly does a good job of stating the research referenced in the video and includes several links in the description so you can explore the sources.
B. Context-Setting: Watching the video before you read the paper gives you a head start. You’ll likely understand the entire paper much faster and can dive deeper into a highlighted finding from the video. I also particularly enjoy how Károly references previous research and videos on related topics to give you a sense for the trajectory and speed of progress in a given field.
Note: Don’t confuse these videos for a total replacement of academic research — you should still be reading things directly from the source (e.g. Google’s Research Blog). Instead, think of TMP as a helpful concierge to guide your learning.
Five Great Video Examples
A Closing Thought…
If business and academic institutions are serious about ‘growing the pie’ in ML and getting exposure for their research, they could benefit a great deal from developing simple video content that functions as abstracts for their experiments and findings.
Organizations like DeepMind clearly understand the need to summarize their findings in engaging mediums — their WaveNet post is a great example of this. Video content could function similarly, both as an explainer (watch a Vox video) or as an intro to more detailed explanations.
Clearly, there are challenges around ensuring research is properly represented, with limitations and further required research acknowledged; however, these seem more like important considerations than immovable barriers.
Serious research peers will still read papers through to the references, but there is a growing community (people like me, business leaders, media) who are looking for accessible ways to approach AI research. Video is increasingly the medium through which we are all introduced to new ideas, and I believe TMP’s example is one that should be followed by research groups with the means and motivation to share their ideas to a wider audience.