A self-operating machine

An implementation of a [thinking machine][rf1].

Virie, P. (2015). (Yet) Another Theoretical Model of Thinking. arXiv preprint arXiv:1511.02455. [rf1]: https://arxiv.org/abs/1511.02455

Behavior

• A generative model: the machine will learn to execute and produce most likely data sequences conditioned on a fixed sequence of past data.
• Learning while executing: like LSTMs, the machine can keep information up to an arbitary step into the past; while like condition models, it only needs to be trained stepwise.
• The machine accepts the very first data sequences as the inputs and learns from them.

Graph

Prerequisites

1. Python 2.7.x
2. Cuda Toolkit 8.0 (https://developer.nvidia.com/cuda-toolkit)
3. TensorFlow r1.0 (https://www.tensorflow.org/install/)
4. array2gif https://github.com/tanyaschlusser/array2gif

License

MIT