A hands-on project inspired by the Continuous Thought Machines (CTM) paper by Darlow et al.
This interactive simulation demonstrates the behaviour of biologically-inspired neurons that:
- Process inputs over time
- Store recent memory
- Fire only when synchronized to a global rhythm
It’s a visual, intuitive way to explore concepts like memory, temporal weighting, and neural phase alignment — all from a browser.
- 🧠 Simulates CTM-style neurons with short-term memory and weighted history
- 🔄 Synchronizes neurons via configurable phase cycles
- ⚙️ Lets you control number of neurons, memory size, input stream, and step-by-step simulation
- 🎨 Visualizes outputs over time using a dark-themed chart — great for demos or concept teaching
- ✅ Step-by-step neuron activation with animation-style controls
- ✅ Dynamic charts reflecting real-time neuron behavior
- ✅ Clean Streamlit interface for showcasing machine learning concepts
- ✅ Great starting point for further explorations into neural dynamics, adaptive compute, or temporal AI models
Inspired by the 2024 CTM paper, this project bridges the gap between neuroscience concepts and visual, code-based experimentation.
It’s built for:
- Curious ML engineers
- Students of neural computation
- Anyone who wants to see what "thinking over time" looks like
git clone https://github.com/calvinx16/ctm-concept.git
cd ctm-concept
pip install -r requirements.txt
streamlit run app/main_app.pyPython · Streamlit · NumPy · Matplotlib
Fully modular — core neuron logic lives in utils/neuron.py
No GPU needed, runs entirely in the browser or locally
Built with neurons, caffeine, and curiosity
by Calvin Paperwala
This project was inspired by the paper:
Darlow, L., Regan, C., Risi, S., Seely, J., & Jones, L. (2025).
Continuous Thought Machines
arXiv:2505.05522
Special thanks to ChatGPT for co-developing, debugging, and hyping every neuron of this project.