Growing Neural Cellular Automata
Distill
Growing Neural Cellular Automata
Differentiable Model of Morphogenesis
Click or tap the image to erase the part of the pattern and see it regenerate.<br>Double clicking places a new seed cell on the grid.
Speed:
Step<br>( step/s)
Model type:
Growing
Persistent
Regenerating
Rotation [experiment 4]
Growing models were trained to generate patterns,<br>but don't know how to persist them. Some patterns explode, some decay,<br>but some happen to be almost stable or even regenerate parts!<br>[experiment 1]
Persistent models are trained to make the pattern stay for a prolonged<br>period of time. Interstingly, they often develop some regenerative<br>capabilities without being explicitly instructed to do so<br>[experiment 2].
Regenerating models were subject to pattern damages during<br>training, so their regenerative capabilities are much stronger,<br>especially in the central area. [experiment 3]
Try in a Notebook
Authors
Affiliations
Alexander Mordvintsev
Ettore Randazzo
Eyvind Niklasson
Michael Levin
Allen Discovery Center at Tufts University
Published
Feb. 11, 2020
DOI
10.23915/distill.00023
This article is part of the<br>Differentiable Self-organizing Systems Thread,<br>an experimental format collecting invited short articles delving into<br>differentiable self-organizing systems, interspersed with critical<br>commentary from several experts in adjacent fields.
Differentiable Self-organizing Systems Thread<br>Self-classifying MNIST Digits
Most multicellular organisms begin their life as a single egg cell - a<br>single cell whose progeny reliably self-assemble into highly complex<br>anatomies with many organs and tissues in precisely the same arrangement<br>each time. The ability to build their own bodies is probably the most<br>fundamental skill every living creature possesses. Morphogenesis (the<br>process of an organism’s shape development) is one of the most striking<br>examples of a phenomenon called self-organisation. Cells, the tiny<br>building blocks of bodies, communicate with their neighbors to decide the<br>shape of organs and body plans, where to grow each organ, how to<br>interconnect them, and when to eventually stop. Understanding the interplay<br>of the emergence of complex outcomes from simple rules and<br>homeostatic<br>Self-regulatory feedback loops trying maintain the body in a stable state<br>or preserve its correct overall morphology under external<br>perturbations<br>feedback loops is an active area of research<br>. What is clear<br>is that evolution has learned to exploit the laws of physics and computation<br>to implement the highly robust morphogenetic software that runs on<br>genome-encoded cellular hardware.
This process is extremely robust to perturbations. Even when the organism is<br>fully developed, some species still have the capability to repair damage - a<br>process known as regeneration. Some creatures, such as salamanders, can<br>fully regenerate vital organs, limbs, eyes, or even parts of the brain!<br>Morphogenesis is a surprisingly adaptive process. Sometimes even a very<br>atypical development process can result in a viable organism - for example,<br>when an early mammalian embryo is cut in two, each half will form a complete<br>individual - monozygotic twins!
The biggest puzzle in this field is the question of how the cell collective<br>knows what to build and when to stop. The sciences of genomics and stem cell<br>biology are only part of the puzzle, as they explain the distribution of<br>specific components in each cell, and the establishment of different types<br>of cells. While we know of many genes that are required for the<br>process of regeneration, we still do not know the algorithm that is<br>sufficient for cells to know how to build or remodel complex organs<br>to a very specific anatomical end-goal. Thus, one major lynch-pin of future<br>work in biomedicine is the discovery of the process by which large-scale<br>anatomy is specified within cell collectives, and how we can rewrite this<br>information to have rational control of growth and form. It is also becoming<br>clear that the software of life possesses numerous modules or subroutines,<br>such as “build an eye here”, which can be activated with simple signal<br>triggers. Discovery of such subroutines and a<br>mapping out of the developmental logic is a new field at the intersection of<br>developmental biology and computer science. An important next step is to try<br>to formulate computational models of this process, both to enrich the<br>conceptual toolkit of biologists and to help translate the discoveries of<br>biology into better robotics and computational technology.
Imagine if we could design systems of the same plasticity and robustness as<br>biological life: structures and machines that could grow and repair<br>themselves. Such technology would transform the current efforts in<br>regenerative medicine, where scientists and clinicians seek to discover the<br>inputs or stimuli that could cause cells in the body to build structures on<br>demand as needed. To help crack the puzzle of the...