Continuous Thought Machines
Continuous Thought Machines (CTM)¶
This folder contains a CTM-inspired "thought experiment" model: a recurrent system that computes over an internal time axis ("thought ticks") instead of solving inputs one-shot.
Why CTM fits this repository¶
- Dynamical systems lens: CTM treats inference as a trajectory, consistent with many simulations here (e.g. relaxation in Hopfield memory, CA evolution, active inference loops).
- Synchronization as representation: measuring pairwise temporal alignment echoes the repo's recurring theme that relations over time can be more informative than instantaneous state.
- Adaptive compute as self-regulation: variable numbers of internal ticks is a computational analogue of homeostasis / regulation under complexity.
Install¶
Base dependencies:
Optional (for CTM / PyTorch-based simulations):
Contents¶
ctm/: small reusable building blocks (starting with Neuron-Level Models / NLMs).