Cellular Automata Can Reduce Memory Requirements of Collective-State Computing

Area of Science:

• Distributed Information Processing
• Computational Neuroscience
• Artificial Intelligence

Background:

• Nonclassical computing approaches like neural networks, reservoir computing (RC), and vector symbolic architectures (VSAs) utilize collective-state computing.
• Collective-state computing represents computational variables as a single high-dimensional state vector, requiring storage of fixed random patterns.
• Efficiently managing these random patterns is crucial for the performance and scalability of collective-state computing models.

Purpose of the Study:

• To investigate the potential of elementary cellular automaton rule 90 (CA90) for optimizing collective-state computing models.
• To demonstrate how CA90 can enable a space-time tradeoff, reducing memory requirements by generating patterns dynamically.
• To evaluate the performance of CA90-based collective-state computing against traditional methods using pseudorandom number generators.

Main Methods:

• Utilized elementary cellular automaton rule 90 (CA90) for on-the-fly expansion of representations from short seed patterns.
• Investigated CA90's randomization behavior, including randomization period length, grid size dependency, and similarity preservation under noise.
• Applied and tested the CA90 expansion method in concrete scenarios within reservoir computing (RC) and vector symbolic architectures (VSAs).

Main Results:

• CA90 enables a space-time tradeoff for collective-state computing models using random dense binary representations.
• CA90 effectively preserves similarity in the presence of initialization noise, crucial for robust computation.
• Collective-state computing models employing CA90 expansion demonstrated performance comparable to traditional models that store pre-generated random patterns.

Conclusions:

• Elementary cellular automaton rule 90 (CA90) provides an efficient mechanism for dynamic pattern generation in collective-state computing.
• CA90-based expansion optimizes memory usage by eliminating the need to store large sets of random patterns.
• This approach offers a viable alternative for building scalable and memory-efficient distributed information processing systems.