P2P No. 40 — Prototyping in Academia
Another piece of advice that academia can learn from start-ups
The fabric of the world of makers and start-uppers is permeated with slogans such as fail fast. Though all quips (including mine) sacrifice nuance on the altar of conciseness, they are not to be dismissed. Fail fast emphasizes the importance of prototyping, which is not archetypical in academia—that does not mean it is not there. However, it receives little attention, making it harder to incorporate into the canon (if and when appropriate).
In short succession, I encountered prototyping three times in academia; thus, it made me reconsider how, why, and when academia could adopt the prototyping mindset.
The idea took me three strikes to sink in. Two of them were rather specific, and one more general. I will go top-down.
The Prototyping Mindset
is a professor of engineering at Harvey Mudd College and the writer of . He writes about structuring his life as a professor in terms of experiments, i.e., trying out new ideas and practices on a small scale. Since everything changes, including you, adopting this approach can help move your expectations from settling into a comfortable but temporary optimum into continuous experimentation.If you focus on the process of figuring out what works best, then the inevitable failure will become a new data point about what is not working (as Edison famously said), and not a devastating strike to your ego.
Scale down for a better understanding
In his keynote at the ELLIS Doctoral Symposium in Helsinki, Amir Zamir
, a professor of computer science at EPFL argued for starting testing new algorithms on a small scale. Amir is working with neural networks and acknowledged that though large-scale models will probably work much better (they are more powerful), it will be bordering on the impossible to understand why they do. Maybe your tweak helped, but maybe it was another of the myriad knobs you (accidentally) turned that gave better results.
On the other hand, playing around with toy models is not very exciting (which I am painfully aware of), but your chances of understanding are higher.
Simulate your proofs
My professional living habitat is a cozy place with a pen, keyboard, and paper. However, developing theoretical arguments can fall prey to the curse of knowledge (that nasty thing again!). Even if your proof seems to be good, it might miss some details.
An invaluable piece of advice I got from my supervisors is to simulate the proof. That is, to create a computer program that models the same setting. There, you need to tell the computer every detail, so it forces you to make your implicit assumptions explicit. Thus, it is a perfect antidote against the curse of knowledge.
Shameless marketing
If you are passionate about neural networks, robotics, or intelligent systems and looking for a Ph.D. in Europe, I cannot help but recommend both graduate programs I am partaking in: the International Max Planck Research School for Intelligent Systems (IMPRS-IS) and the European Laboratory for Learning and Intelligent Systems (ELLIS).
Most of the insights I write about are related to my experience as an IMPRS-IS and ELLIS scholar. In the following weeks, more is coming.
Important dates and resources
IMPRS-IS application deadline: November 15, 2023, 23:59 CET. Find out more on the IMPRS-IS website.
ELLIS application deadline: November 15, 2023, 23:59 CET (yes, they are—not coincidentally—the same). Find out more on the ELLIS website.