toy-models

Ensembler is a python package that provides fast and easy access to 1D and 2D model system simulations. It can be used for method development or to deepen understanding of a broad spectrum of modeling methods, from basic sampling techniques to enhanced sampling and free energy calculations. It is easy to install, fast, increases shareability, co…

A replication of "Toy Models of Superposition," a groundbreaking machine learning research paper published by authors affiliated with Anthropic and Harvard in 2022.

Towards an interpretable understanding of MLP behaviour in applied machine learning

Extraterrestrial Pets Simulation: 2D toy models, agent-based models, and cellular automata on triangle, square, and hexagon grids, built with Java and JavaFX

JAX-accelerated finite volume solver for 2D ideal magnetohydrodynamics with HLL Riemann solver and Dedner divergence cleaning. Includes comprehensive conservation, turbulence, and information-theoretic diagnostics with CF-compliant NetCDF output.

Toy-model experiments on how training configurations affect feature-entanglement proxies in neural representations.

T-REX (Toy Reactive Exchange) Monorepo: the toy models, EVM contracts and super app (for superfluid protocol).

Reproducing toy models of superposition: how a network stores more features than it has dimensions.

A very-insecure toy model, rather a baby model of SSH.

A short demonstration of GANs learning a probability distribution

A Numba-accelerated Python library for simulating 2D Lagrangian particle transport in idealized semi-enclosed ocean basins using Bell's incompressible flow field. This is an idealized 2D model intended for process studies and educational purposes, not operational forecasting.

Creating a toy-model approach for applying Quantum Principal Component Analysis for Educational Data

Auditable 2D source-to-lapse toy model with reproducible numerical baselines and source-geometry comparisons.

Minimal control-theoretic toy models for studying stability and failure modes under partial observability.

Toy-model audit of superposition capacity and the Sarkar & Deka 2026 minimum-width distillation floor: an equilibrium-corrected predictor that fixes Eq. 2 near/above d*.

Вычислительная археология науки и поиск скрытых математических паттернов в архивных физических моделях.

Toy models of computation in superposition (100 sparse ReLUs on 50 neurons): the superposed code grows continuously out of the dedicated-neuron solution rather than appearing at a jump. This repo is the preliminary, single-seed first pass; the 256-job CPU sweep and the 40-page write-up are not public yet.

Toy Model implemention to train a Reinforcement Learning Agent for Within Visual Range (WVR) aerial combat.