Animated research Watch the mathematics move. Twelve cinematic explainers, grouped by what they are. Five animate results from my own papers on proof-of-learning, model watermarking, blockchain-enhanced ML and rare-event detection. One comes from the hard-real-time systems I engineer. One is an unpublished architecture experiment connecting AI inference to real-time execution. Three animate foundational results that belong to other people, credited as such. Two are open questions I am still working on, labelled so they are not mistaken for reviewed work. The films distinguish published results from teaching models and proposed experiments, with sources and assumptions stated on their pages.
From my published research Each of these animates a result from one of my own papers, with the mechanisms and the measured numbers the paper reports.
01 Proof-of-Learning (SecurePoL) A loss trajectory used as evidence of training provenance: inexpensive to record, harder to reproduce under the stated attack model. IEEE Access 2025 and my dissertation. 02 Blockchain-Enhanced ML Consensus that trains instead of hashing, contracts that pay for the loss you removed, and what the prototypes actually measured. IEEE Access 2023, my most-cited paper. 03 Model Heist Detector A watermark spread across many weights: a Gaussian Z-test illustration whose detection depends on calibration and attack assumptions. IEEE Access 2024. 04 Watermarking Models A side-by-side comparison of ML model watermarking architectures and their robustness. Auxiliary-head analysis from my dissertation. 05 Rare-Event Detection Why a 99%-accurate detector is wrong most times it speaks, and what an agglutinative language does to your features. ICISSP 2021. From the systems I engineer Hard real time as it is actually regulated and measured, drawn from my work on Level D full-flight simulators. No employer design is disclosed.
06 Determinism at 60 Hz A deadline is never met on average: the 150 ms qualification gate, 864,000 frames a session, and the straggler host that owns the frame. Architecture experiments Executable design studies with explicit assumptions. Proposed architectures, not claims of deployment or qualification.
07 AI Meets the Deadline Where should inference run? Test asynchronous proposals, expiry, fallback and resource contention around a 60 Hz execution loop. The mathematics behind the field Results that belong to other people, animated because the derivations are worth seeing move. The papers are cited on each page.
08 Gradient Pinball How machines learn: the learning-rate cliff, the √κ momentum speedup, and the saddle-point surprise. After Polyak, Nesterov and Dauphin et al. 09 Block Race Why "6 confirmations" is a probability, not a promise: Bitcoin's §11 double-spend math, animated. After Nakamoto. 10 Redundancy Reactor Superlinear safety, until correlation installs a floor that destroyed Ariane 5. Classical TMR and the Ariane 5 inquiry. Open directions, not yet published Questions I am working on rather than results I have proved. Labelled so no one mistakes them for reviewed work. Feedback and collaborators welcome.
11 ML & Blockchain Oracles By what mechanism can a deterministic chain accept a claim about a model it cannot re-execute? 12 Universal Jira Board Can prediction markets price and settle engineering work without a central planner? 