Projects

• Rainbow Arrays blog post series After a lot of work (and software engineering) I finished up the first two episodes in what is hopefully a longer series of illustrated blogs. The theme is multidimensional arrays and how we manipulate them conceptually. I know the deep learning crowd like to call them tensors, but this brazen theft of terminology from mathematical physics has some some big downsides. First, it makes everyone assume that array programming is just linear algebra, which it definitely is not. Second, it means that …
• Neuroevolution Paper gene expression, connectomes, and metalearning A few years into my mid-career pivot into science, I just landed my first published paper! What started as an idea I wrote on a napkin at the Cosyne neuroscience conference is finally out in Nature Communications: “Complex computation from developmental priors”. […] In the paper, my neuroscience collaborator Dániel and I describe how to cast neuroevolution as a metalearning problem, by treating his XOX model of connectome generation as a differentiable step in the two-stage …

• DLI: "Array Algebra" NumPy practical for the Deep Learning Indaba in Tunisia For the 2022 Deep Learning Indaba (held in Tunis), I contributed a practical entitled “Array Algebra”. The goal was to equip students with a better understanding of multidimensional arrays of various order, since array manipulation is the bread and butter of programming deep learning architectures. […] I used 2D and 3D illustrations of the fundamental operations of slicing, aggregation, and transposition, with a particular focus on manipulating image arrays as these illustrate …

• Quiver Geometry discrete geometry via graph theory Quiver geometry is an experiment in defining a kind of discrete geometry purely in terms of directed graphs with a particular structure of edge labels on them. I began thinking about this project in late 2020 when thinking about the abstract geometry induced by computational rewriting systems, having found nothing in the existing literature that seemed like a good “port” of the methods and theories of classical differential geometry to the setting of discrete geometry. I started to …

• MCTS in Swift AlphaZero and MCTS implementation for OpenSpiel This project is a work-in-progress fork of the OpenSpiel library. It adds an MCTS implementation that will form the basis of a simple AlphaZero trainer. The design leverages some of the powerful type system abilities of Swift to achieve a maximally composable design, which allows the underlying tree-building implementation to be shared between several algorithms (various variants of MCTS, as well as Alpha-Beta search), which additionally makes them very clear. The fundamental Swift constructs …

• DLI: "Build Your Own TensorFlow" TensorFlow practicals for the Deep Learning Indaba in Kenya For the 2019 Deep Learning Indaba (held in Nairobi, Kenya), I was privileged to co-organize the tutorial sessions with Jamie Allingham, with the extensive help and support of Avishkar Boopchand, Stephan Gouws, and Ulrich Paquet of DeepMind. We had more than 50 tutors this year, covering more 500 students across 2 parallel sessions each day. Our goal for this year was to expand the range of material that the tutorials covered, to address pain points we saw in previous years, and to further …
• Spieeltjie experiment with multi-agent RL on zero-sum differentiable games Spieeltjie is a single-file package for doing simple experiments with multi-agent reinforcement learning on symmetric zero-sum games. For more information see “Open-ended learning in Symmetric Zero-Sum Games” and “A Unified Game-Theoretic Approach to Multiagent Reinforcement Learning”. The name “spieeltjie” comes from the Afrikaans word for “tournament”. […] This first set of images shows trajectories when starting from a set of random …
• Wasserstein GAN DFL contributed to Depth First Learning course on Wasserstein GANs While tutoring at the 2019 Deep Learning Indaba, I got to know the multi-talented Cinjon Resnik, who is currently doing his PhD with Kyunghyun Cho at NYU. After the Indaba, Cinjon invited me to join an experiment he is running in distributed teaching and learning called Depth First Learning. One innovation that particularly resonated with me was the effort DFL makes to plot a path through “paper space”, as a way to explain a core idea or story. The story we chose as the backbone for …
• Biologically Plausible Backprop feedback alignment & activity perturbation in PyTorch SOAP (Second Order Activity Perturbation) is a package for experimenting with a computational neuroscience phenomenon called feedback alignment in PyTorch. It formed my project for the 3-week IBRO-Simons Computational Neuroscience Summer School in 2019. It relates to an important challenge in reconciling how learning works in artificial neural networks with what we know about how real neurons behave, a topic called biologically plausible back-propogation. […] When a neuron fires, its axon …

• Deep Learning for Mathematica high-level, flexible, friendly deep learning From 2016 to 2018, I led development of the neural network framework, which encompassed 50+ functions and objects. Since version 10.3 it has been part of the core Mathematica language. It covers a wide range of state-of-the-art deep learning functionality, in an interface that is extremely easy for scientists and engineers to use, rivaling and exceeding the ease-of-use of libraries like Keras. One notable use of the framework was to replace the filter-bank based gravitional wave detectors used …

• Cryptography for Mathematica user-friendly cryptography API
• Semantic Import for Mathematica automated import and recognition of rich tabular data
• Text Search for Mathematica easy-to-use lucene-backed indexing engine

• Hierarchical Data in Mathematica DataFrame-like construct for data science workflows For Mathematica 11, I contributed the the new Dataset construct in Mathematica, which acts as a flexible container for tabular data. It involved a lot of interesting challenges around automated layout and deduction of types from unstructured data. The underlying technology scales to quite complicated nested use of Mathematica’s two native data structures, being lists $\{v_1,v_2,...\}$ and associations $\left<\left| k_1 \to v_1,k_2 \to v_2,... \right|\right>$. Because Mathematica is a …

• Facebook Data Analysis analysis and visualization for popular blog by Stephen Wolfram For this project, I helped Stephen Wolfram analyze a dataset of Facebook profiles we collected from volunteers who engaged with a previous project I worked on. We looked at the influence of the friendship paradox, correlations between various demographic variables, performed topic modeling on people’s facebook posts, and did cluster analysis on their friend graphs. The friendship paradox summarizes that, on average, your friends have more friends than you do, because popular people are …
• Wikispider tool for breadth-first crawling of matching Wikipedia articles

• Wolfram|Alpha Facebook Analytics free personal analytics for your Facebook profile
• Wolfram|Alpha Pro automatic analysis of user-uploaded datasets

• Sonifying Shakespeare experiment in sonifying Shakespeare's plays While instructing at the NKS summer school, I helped students from diverse backgrounds to implement their ideas in Mathematica. One of the my favorite projects was working with the artist Elizabeth Latta to visualize and computationally explore the famous Shakespeare play The Tempest. Among the many promising ideas we investigated, two in particular were interesting enough that I’d like to show them off. The beautiful drawings are Elizabeth’s work, and I planned and wrote the code. …

• Spatial Game Theory package for experiments in evolutionary spatial game theory SGT is a Mathematica package to unlock effortless explorations in evolutionary spatial game theory. Using SGT, you can setup, conduct and visualize experiments using a small number of functions and symbolic expressions to specify the initial distribution of agents, the topology of the grid, the strategies the agents employ, the payoff structure of the games they play, the rules used to update the agents and apply natural selection, whether and how mutation occurs, whether agents have access to …

• iThemba Labs GUI for programming FPGA particle physics detection trigger circuits I spent a few months over my 3rd year vacation period interning at the iThemba Laboratory for Accelerator-Based Sciences in Cape Town. The facility had a dorm, where I stayed with other interns. During the day we worked on different software projects. By night, we roamed the (fortunately disabled) proton beam lines. […] There was no one there on weekends, but two things kept it from being dull. First, the facility bordered the Khayelitsha township, so we could enjoy cheap beer from the …

• Floatworld evolutionary sim of RNN-controlled agents on a gridworld floatworld is an C++ library and associated GUI that serves as a laboratory for a multi-agent RNN-based simulation in which agents compete for space and energy on a two-dimensional grid. I wrote floatworld, along with a short paper about it, during my undergraduate degree. Each agent is controlled by a recurrent neural network and harvests energy in order to reproduce. Time and space are discretized. Agents obtain a variety of data from their environments that is fed to their neural network …

• RLC Diagnostics program to use sound card to measure RLC circuit resonance I completed this project while studying my final year of high school. I was in the process of building and tuning an electromagnetic coilgun, and as part of that effort I needed to accurately measure the inductance of my prototype coils. Unfortunately, I couldn’t afford the expensive, professional equipment that could make such measurements. Fortunately, I read about the techniques that this equipment actually used, and I realized I could use my computer’s sound card to do much the …

• PenGUIn from-scratch cross-platform GUI and windowing system