Demos
Small interactive pages that run entirely in the browser, made for talks and for playing. Each one works offline once loaded and can be presented fullscreen on a projector.
The defence, slide by slide
The slides of my public PhD defence, every one of them live: the audience plays rule 90 with their hands, the Game of Life turns into rules on networks, a quarter of a million rules shrink to 27 without a single simulation, and one stubborn person flips a whole network. Step through the builds here, present it with a clicker, or take the whole talk offline as one file.
The thesis, chapter by chapter
My PhD thesis as twelve tabs: how much of a cellular automaton’s behaviour can be read off its rule, how the wiring changes the answer, and why a cellular automaton is a neural network with its weights written down by hand. One live figure per chapter, from defect spreading and Lyapunov spectra to finding the nodes that matter in a network.
Life on graphs
Conway’s Game of Life taken onto networks, where one node has two neighbours and the next has twenty. Build any rule on the interval diagram and read its mean-field curve, pick 1 rule out of 262,144 that synchronises a firing squad, turn it into a majority vote, and follow a single flipped node as it heals or spreads on lattices, small worlds and random networks.
Five ways to bend a cellular automaton
The families of cellular automata beyond the elementary one: cells that update out of step, rules that roll dice, more states, wider neighbourhoods, different rules for different cells. Each family in a tab with its definition and variants, and the exact identities between them run side by side and checked cell by cell, down to the 39-digit rule number of the paper.
The automaton is the network
A cellular automaton written as a convolutional neural network with its weights set by hand: forty weights for one rule, one more layer when every cell has its own. Take an update apart, watch the paper's figures come out cell for cell, train the weights from scratch in your browser, and see what the benchmarks did and did not show.
Is the network cheating?
Neural networks sort cellular automata by behaviour almost perfectly, but every picture also spells out the rule that drew it. Scan a diagram for its rule table, run a network that reads the rule with no training at all, then a small one built so that it cannot, and see how far altered pictures push it towards the behaviour.
Where the virus travelled
Did it matter where Belgians travelled in 2020, in a country this small and this connected? Pick any of the 43 arrondissements and see its commuters and the excess deaths side by side, week by week; watch a noisy curve of hospital admissions turn into a fan of 100 fits; and slide one province's epidemic over another's to see who ran ahead. All on public data, computed in the browser.
Finding echoes
What a gravitational-wave echo from a not-quite-black-hole would look like, where it would come from, and how to dig one out of detector noise without knowing its shape. Four animations from my MSc thesis.