Changelog

POYO+ (NeurIPS’23), POSSM (ICLR’25), NDT3, CEBRA (Nature’23)

ScKGBERT + Geneformer interfaces

37K param progressive SIG sharpening

AdaptiveThresholdMoE, QuantumInspiredLIF, DendriticNMDA, CochlearHairCell

CardiacPurkinjeFibre, MyelinatedAxon, SmoothMuscleCell

24x waveform, 750x spike, unified API

Gaussian Process Factor Analysis reconciled across all five backends (NumPy, Rust, Julia, Go, Mojo) with deterministic PCA initialisation, the exact Gaussian marginal log-likelihood, and the structured Cholesky estimator of Yu et al. (2009) — using the Woodbury identity and matrix-determinant lemma so it never forms the dense marginal covariance; cross-backend parity to float64 round-off.

Candidate-first RK4 integration across Python, Rust, Go, Julia, and Mojo for EIF, McKean, Wilson-HR, Pernarowski, and Terman-Wang models with bit-exact cross-backend parity; z3 0.12→0.20

Evaluator map and claim-traceability pass across README, docs, benchmark, and hardware evidence surfaces

Public Product Overview and Applications & Market pages, plus a notebooks README with recommended reading order and reproducibility rules

8 new neuromorphic hardware profiles (thermodynamic, p-bit, polariton, metamaterial emulators) and compiler extensions

Web IDE for equation editing, network design, training, compilation, and FPGA synthesis

Li et al. 2014 stochastic division circuit

ODE string to synthesisable SystemVerilog pipeline

Full compatibility with Python 3.14

9 AI-optimised neuron models with novel dynamics

End-to-end MNIST classification on FPGA hardware

Complete Brunel network model variants

Full 15+1 layer SCPN adapter coverage

Performance benchmarks and documentation updates