-
Notifications
You must be signed in to change notification settings - Fork 13
Engine overview
-
Neurons are dynamically created and removed based on experience patterns - this allows the network to dynamically reshape itself over time.
-
Neurons are labelled by function such as
novelty,stress, andreward
neural growth reflects the type of experience that produced them.
A "Bottom-Up" sensory system where raw environmental data is distilled into neural inputs, which are then filtered through the squid's Decision engine and personality to produce behaviour.
Built from scratch using NumPy.
- No TensorFlow.
- No PyTorch.
- Explicit neuron-level simulation
- Hebbian plasticity
- Structural growth (neurogenesis)
- Dual memory system (short-term and long-term)
- Headless training capability
- Plugin extensibility
- Optimised for interpretability not scale.
Treats neural networks not as static architectures, but as evolving structures.
-
Network grows via neurogenesis and self-trains via Hebbian learning
-
Built-in support for Spike‐Timing‐Dependent Plasticity (STDP)
-
Automatic pruning of redundant neurons and weights (can be turned off)
-
Experience buffer records and encodes learned experiences
-
decision_engine uses neural data to make decisions
- Beta (and optional) support for AI accelerators via ONNX Runtime
- Experimental and a work in progress
- Probably not the best way to do this! 😃
Read Next: Data flow Summary overview
- Neural Network: Technical Overview
- Brain Format (json)
- How Dosidicus differs from other systems
- Decision Engine
- Brain Widget
External links (Research and inspiration for this project)
- https://medium.com/@reutdayan1/hebbian-learning-biologically-plausible-alternative-to-backpropagation-6ee0a24deb00
- https://informatics.ed.ac.uk/sites/default/files/2024-03/Qiuye%20Zhang%20Lovelace%20Colloquium%20Poster.pdf
- https://en.wikipedia.org/wiki/Hebbian_theory
- https://www.youtube.com/watch?v=TvTQQO5yTa4
🦑 Raise digital squids whose brains grow & rewire themselves through Hebbian learning and Neurogenesis