Symbolic Computation Using Cellular Automata-Based Hyperdimensional Computing.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Complex Systems

Background:

• Reservoir computing (RC) traditionally relies on recurrent neural networks.
• Existing RC methods like echo state networks (ESNs) can be computationally intensive.
• Integrating symbolic computation with connectionist approaches remains a challenge.

Purpose of the Study:

• Introduce a novel reservoir computing framework utilizing cellular automata.
• Demonstrate the framework's capability for both connectionist machine intelligence and symbolic computation.
• Propose a computationally efficient alternative to traditional RC methods.

Main Methods:

• Employed a cellular automaton as the dynamical system reservoir.
• Randomly projected input data onto the initial conditions of automaton cells.
• Utilized the automaton's rule-based evolution to create a space-time state-space volume as the reservoir.
• Combined binary reservoir feature vectors using Boolean operations, inspired by hyperdimensional computing.

Main Results:

• The proposed framework exhibits long-term memory capabilities.
• Achieved significant computational savings (orders of magnitude less) compared to echo state networks.
• Demonstrated potential for concept building and symbolic processing through Boolean operations on reservoir features.
• Successfully performed analogies on image data, exemplified by the query 'What is the automobile of air?'

Conclusions:

• Cellular automata offer a powerful and efficient reservoir for computing.
• The framework bridges connectionist and symbolic AI, enabling novel applications.
• This approach holds promise for advancing machine intelligence with reduced computational overhead.