A Two Minute Papers YouTube episode by Dr. Károly Zsolnai-Fehér summarizing a Distilled Journal paper about a cellular automaton that is trained (via a formulation analogous to deep neural networks) to grow a prescribed organism from a single cell. Key differences from prior CA work: cells have three states (empty, growing, mature) and the training setup mirrors deep learning, which yields surprising emergent behaviors — some instances decay or stop growing, while others exhibit untrained regeneration and robustness to rotation. The video notes the paper is interactive in the Distilled Journal and highlights the first author Aleksandr Mordvintsev and a sponsorship by Weights & Biases.
Chapters
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Today we are going to play with a cellular automaton. You can imagine these automata as small games where we have a bunch of cells and a set of simple rules that describe when a cell should be full and when it should be empty. These rules typically depend on the state of the neighboring cells. For instance, perhaps the most well known form of the cellular automaton is, is John Horton Conway's Game of Life, which simulates a tiny world where each cell represents a little life form. The rules again depend on the neighbors of this cell. If there are too many neighbors, they will die due to overpopulation. If too few, they will die due to underpopulation. And if they have just the right amount of neighbors, they will thrive and reproduce. So why is this so interesting? Well, the cellular automaton shows us that a small set of simple rules can give rise to remarkably complex life forms, such as gliders, spaceships, and even John von Neumann's universal constructor, or in other words, self replicating machines. I hope you think that's quite something. And in this paper today, we are going to take this concept further, way further. This cellular automaton is programmed to evolve a single cell to grow into a prescribed kind of life form. Apart from that, there are many other key differences from other works, and we will highlight two of them today. One, the cell state is a little different because it can either be empty, growing or mature. And even more importantly, two, who. The mathematical formulation of the problem is written in a way that is quite similar to how we train a deep neural network to accomplish something. This is absolutely amazing. Why is that? Well, because it gives rise to a highly useful feature, namely that we can teach it to grow these prescribed organisms, but wait, over time, some of them seem to decay, some of them can stop growing, and some of them will be responsible for your nightmares. So from this point on, proceed with care. In the next experiment, the authors describe an additional step in which it can recover from these undesirable states. And now hold on to your papers because this leads to one of the major points of this paper. If it can recover from undesirable states, can it perhaps regenerate when damaged? Well, here you will see all kinds of damage and then this happens. Wow. The best part is that this thing wasn't even trained to be able to perform this kind of regeneration. The objective for training was that it should be able to perform its task of growing and maintaining shape. And it turns out some sort of regeneration is included in that it can also handle rotations as well, which will give rise to a lot of fun and, as noted a moment ago, some nightmarish experiments. And note that this is a paper in the Distilled Journal, which not only means that it is excellent, but also interactive. So you can run many of these experiments yourself right in your browser. If Aleksandr Mordvintsev, the name of the first author, rings a bell, he worked on Google's Deep Dreams approximately five years ago. How far we have come since My goodness, loving these crazy non traditional research papers and I'm looking forward to seeing more of these. This episode has been supported by Weights and Biases. Here they show you how you can visualize the training process for your boosted trees with XGBoost using their tool. If you have a closer look, you'll see that all you need is one line of code, weights and biases provides tools to track your experiments in your deep learning projects. Their system is designed to save you a ton of time and money and it is actively used in projects at prestigious labs such as OpenAI, Toyota Research, GitHub and more. And the best part is that if you're an academic or have an open source project, you can use their tools for free. It really is as good as it gets. Make sure to Visit them through wnb.com papers or just click the link in the video description and you can get a free demo today. Our thanks to Weights and Biases for their long standing support and for helping us make better videos for you. Thanks for watching and for your generous support and I'll see you next time.
