echo-state-network

Python package that provides predictive models for fault detection, soft sensing, and process condition monitoring.

A Python 3 framework for Reservoir Computing with a scikit-learn-compatible API.

Real-time detection and repair of LLM agent failures — a one-class behavioural monitor at ~200 µs/step, with 2,823 committed traces.

An easy-to-use independent machine learning library for .net. It offers MLP models (including deep RVFL aka ELM) for common ML tasks as well as Reservoir Computer for efficiently solving time series ML tasks.

Implementation of Echo State Networks (ESN) with experiments on MNIST and ECG5000. Includes comparison with Linear Regression and analysis of weight initialization methods for time-series and classification tasks.

Rust crate for chaotic semantic memory - echo state networks and hyperdimensional vectors.

Julia implementation of cross-scale reservoir computing for high-dimensional spatiotemporal forecasting. Validated on global sea-surface temperature and chaotic dynamics, achieving low error forecasts up to 8 Lyapunov times on Kuramoto-Sivashinsky. Neurocomputing (2026).

turbESN is an echo state network implementation, used in my PhD research as part of the DeepTurb project of the Carl-Zeiss Stiftung. See https://pypi.org/project/turbESN/

Reservoir computing with coupled genetic oscillators for arrhythmia classification

A general-purpose recursive self-improvement loop with a real fly brain: the larval Drosophila connectome as a fixed reservoir, wrapped in a closed, observable L1-L5 loop — search, curricula, gated online adaptation, and an inherited experience bank. The wiring never changes; the lab improves itself. Handwriting is just the first benchmark.

Does the fruit fly brain forecast Polish wind and solar power? The FlyWire connectome as an echo state network reservoir vs random reservoirs and standard baselines on PSE data: an honest negative result

Physics-Informed Reservoir Learning for Shallow-Water Flood Modelling

Recursive echo-state reservoir on a 3D Boolean hypercube: each vertex's tanh neuron is replaced by a full sub-reservoir of identical topology, nested to a configurable depth (1-5), with bidirectional parent/child coupling. C++23.

MUUFL Gulfport 高光谱地物分类：ESN 混合架构，验证准确率 87.28% → 90.11%

轴承健康状态监测与故障部位识别：STDP 脉冲神经网络 + 储备池计算，NASA/CWRU/Paderborn 三数据集

Echo State Network (ESN) on a Boolean hypercube reservoir: sparse, XOR-defined connectivity and no stored adjacency. A tunable per-neuron delay line scales memory capacity super-multiplicatively. C++23 and Python SDKs.

Between a reservoir and an RNN : start from a frozen random recurrence and learn only one chosen piece (rotation, gains, shear…). Finding : ease of learning beats target geometry, even when the exact solution lies in the family.

Reservoir computer on a Boolean hypercube graph. Sparse XOR-defined connectivity. C++23 and Python SDKs.

Echo state networks for learning variational level set image segmentation. Code behind the MSc dissertation, University of the Witwatersrand, 2018 to 2019.

The complete fruit fly connectome (MaleCNS v1.0, Google/Janelia 2026) as a token-in/token-out language model, benchmarked against a transformer at matched compute. Full scale, real data, fully replicable.