Spectra Research Writing Competition results announced for Spring 2022

By Nico Jimenez
We now announce the winners of the Spectra writing competition. Thanks to everyone who participated!

(#1) Event Camera: The Next Generation of Visual Perception System by Kaichao You

I had never heard of event cameras until reading this article so I found this article fascinating. Compared to frame by frame cameras, event cameras are much faster (no pixel synchronization needed), have much lower power consumption (if frames don’t change nothing happens), and also much higher dynamic range (easier to measure logarithmic change in intensity than absolute intensity). That is an insane combination of features! Read the article to learn more. Thank you Kaichao for sharing this exciting research direction with the Mathpix research community. Event cameras are not a new idea and there are existing products on the market but it does seem like now is a good time to work on them. There are many exciting applications for this technology, from satellites to self-driving cars.
Event cameras are not only cool from an engineering / entrepreneurial perspective, they are interesting from an evolutionary perspective. As someone who dabbled in computational neuroscience as an undergraduate, I was always fascinated by the event driven nature of neural systems, which use spiking as their primary mechanisms for information communication. So it’s exciting to see engineered systems become more like natural ones, without trying to do so intentionally (in fact event cameras are sometimes called neuromorphic cameras). The mathematics of energy and information are hard constraints on the universe, and so we expect evolution to favor efficient information processing, regardless of whether the computational substrate is silicon or spiking neurons. Some convergence is expected, but it’s still amazing to see it happen.
Of course, event cameras may have event driven neural networks behind them. One can imagine neurons a few layers deep that can capture high level information about an image, for example hot dog / not hot dog. Such a neuron would not fire until a hot dog either enters the field of vision or exits it. Such event driven mechanisms at all layers of visual processing could drastically reduce the amount of energy required for visual monitoring, and may have unexpected side benefits as well .
Such neural networks will probably be extremely difficult to train, and won’t be CUDA friendly at all, but that’s the beauty of engineering: you often need to take a step backwards to take two steps forwards.
We think that event cameras are an exciting direction, and we hope some of you find this topic interesting and end up working on it. There is a lot of work that needs to be done on hardware and software to make this technology ubiquitous. It will get there for sure.

(#2) Probing the Full Monty Hall Problem by Erica Mock

“Suppose game-show contestants are shown three doors, and there is a prize behind one of them. Contestants are asked to pick a door, giving them a one in three chance of winning. The host then opens one of the losing doors and asks contestants if they would like to switch to the remaining door.” (Mock)
What should you do? A fun article about perhaps the most notorious statistical puzzle of all time, complete with simulations, mathematical generalizations of the fundamental problem, and trivia. There is something very pure about the author sharing her love of mathematics and curiosity with the world. Thank you! With your help, we can be as smart as a pigeon:
“In his book The Power of Logical Thinking, cognitive psychologist Massimo Piattelli Palmarini writes: “No other statistical puzzle comes so close to fooling all the people all the time and even Nobel physicists systematically give the wrong answer, and that they insist on it, and they are ready to berate in print those who propose the right answer.” Pigeons repeatedly exposed to the problem show that they rapidly learn to always switch, unlike humans.” (Wikipedia)

(#3) A Review of Trustworthy Graph Learning by Jintang Li

Graph neural network techniques have become pervasive in areas from finance and computational chemistry to recommendation algorithms for social networks. Unlike data such as images or text, graph data is difficult to visualize. This means that robustness estimates and evaluation metrics are significantly harder to obtain in this domain. In addition, there are privacy and equity issues that must be considered when dealing with graph neural networks on social network graphs. This paper summarizes research directions in this field which aims to build complex graph neural networks that humans trust.

(#4) High Dimension Data Analysis - A tutorial and review for Dimensionality Reduction Techniques by Srishti Saha

A nice tutorial and review on dimensionality reduction for high dimensional data. Nowadays, it’s easy to become lost in a dizzying array of complex techniques, and forget about simple variance and correlation calculations. When data quality is low, these methods are essential.

Honorable mentions

What’s next?

All eligible submissions were given Mathpix Pro for 6 months free of charge. We will maintain this policy moving forward but will change this to 12 months for qualifying submissions.
Our next technical writing competition is now officially open! It will end on October 1st, 2022. We encourage prospective authors who want to share their opinions and knowledge with the Mathpix research community to submit an article now to get Mathpix Pro for a year and also automatically enter into the next competition. More competition details can be found here.
From now on, all cash rewards for competition winners will be $2000, and we will cut the delay between article submission and cash reward to under 2 weeks. This way authors can be compensated immediately for their contributions. There will be a minimum of 3 winners but we expect and hope for more. The great submissions are obvious at the very beginning and therefore we don’t need to wait until more submissions come in to give out awards. Answers to frequently asked questions can be found here.

Thank you!

We also want to thank our Snip users that upgraded to Mathpix Pro, we are working on some awesome product upgrades for you. Stay tuned!