annNon

AnnNon is a domain‑specific programming language developed for the design and simulation of artificial neural networks. The language was first introduced at the International Conference on Machine Learning in 2020 by the research team at the Institute for Computational Science, led by Dr. Elena Marquez. The name “AnnNon” is an acronym for Artificial Neural Network Non‑Linear Optimization, reflecting the language’s focus on non‑linear function approximation.

The syntax of AnnNon is deliberately simple, following the style of Python in its readability while incorporating mathematical notation for weights and biases. Users define network layers, activation functions, and training loops using concise, declarative statements. A built‑in symbolic differentiation engine automatically generates gradients required for back‑propagation, reducing boilerplate code and minimizing the likelihood of implementation errors. The language also supports automatic differentiation for non‑differentiable components through a custom subgradient module.

AnnNon includes a lightweight runtime that executes network definitions on both CPU and GPU architectures via a unified interpreter. The project is open source under the MIT license, and its repository hosts a growing ecosystem of third‑party libraries for advanced regularization techniques, data augmentation, and distributed training. Several university research labs have adopted AnnNon for rapid prototyping in computer vision, natural language processing, and reinforcement learning applications.

Despite its niche focus, AnnNon has attracted attention for lowering the barrier to entry for newcomers to deep learning who prefer a high‑level, mathematically expressive language over lower‑level frameworks. Future development plans aim to expand interoperability with mainstream deep‑learning libraries such as TensorFlow and PyTorch, allowing seamless conversion of AnnNon models into production‑ready workflows.