Open Source Lua Neural Network Libraries

ResNeXt is a deep neural network architecture for image classification built on the idea of aggregated residual transformations. Instead of simply increasing depth or width, ResNeXt introduces a new dimension called cardinality, which refers to the number of parallel transformation paths (i.e. the number of “branches”) that are aggregated together. Each branch is a small transformation (e.g. bottleneck block) and their outputs are summed—this enables richer representation without excessive parameter blowup. The design is modular and homogeneous, making it relatively easy to scale (by tuning cardinality, width, depth) and adopt in existing residual frameworks. The official repository offers a Torch (Lua) implementation with code for training, evaluation, and pretrained models on ImageNet. In practice, ResNeXt models often outperform standard ResNet models of comparable complexity.

char-rnn is a classic codebase for training multi-layer recurrent neural networks on raw text to build character-level language models that learn to predict the next character in a sequence. It supports common recurrent architectures including vanilla RNNs as well as LSTM and GRU variants, letting users compare behavior and output quality across model types. It is straightforward: you provide a single text file, train the model to minimize next-character prediction loss, then sample from the trained network to generate new text one character at a time in the style of the dataset. The project is designed for experimentation, offering tunable settings for depth, hidden size, dropout, sequence length, and sampling temperature to control creativity and coherence. It is frequently used as a learning project for understanding sequence modeling, recurrent training dynamics, and the practical details of text generation.

A parallel-programming framework for concurrently running large numbers of small autonomous jobs, or microthreads, across multiple cores in a CPU or CPUs in a cluster. NeuraNEP emulates a distributed processing environment capable of handling millions of microthreads in parallel, for example running neural networks with millions of spiking cells. Microthreads are general processing elements that can also represent non-neural elements, such as cell populations, extracellular space, emulating sensory activity, etc. NeuraNEP handles microthread scheduling, synchronization, distribution and communication. This project is a fork of SpikeOS (sourceforge.net/projects/spikeos) and represents a major update to that code base, including a scripting interface and low-level rewrite of several components. SpikeOS was oriented towards computational modeling. NeuraNEP is oriented toward neural network research.