Representations, Algorithms, and Neural Implementations in Cognitive Systems

Base 10 and Base 2 Representations in Maximum-Finding Algorithms

• Base 10 (Decimal) Numerical Representation:

  • When evaluating numbers such as $75$, $15$, and $25$, values are represented using positional place values, specifically a tens place and a ones place.

  • An algorithm defines a systematic set of rules used to determine the maximum number among a set of candidate options.

  • The base 10 maximum-finding algorithm evaluates the highest place value (tens place) first, scanning sequentially for the highest digit starting from $9$, then $8$, and then $7$.

  • Identifying a $7$ in the tens place for $75$ uniquely establishes it as the largest value among $75$, $15$, and $25$ without requiring evaluation of the ones place.

  • Positional significance dictates that placing a $7$ in front of a $5$ makes the $7$ higher in importance for determining the overall magnitude of the outcome.

• Base 2 (Binary) Numerical Representation:

  • Digital computing systems achieve the exact same computational goal of finding a maximum value using binary (base 2) representations.

  • Binary positional values increment by powers of 2, featuring a ones place ($2^0$), twos place ($2^1$), fours place ($2^2$), eights place ($2^3$), and sixty-fours place ($2^6$), utilizing only the digits $0$ and $1$.

  • The decimal number $75$ is represented in binary format as a $1$ in the 64s place, a $1$ in the 8s place, a $1$ in the 2s place, and a $1$ in the 1s place, corresponding to the sum $64 + 8 + 2 + 1 = 75$.

  • The binary algorithm for determining the maximum value starts at the highest place value and scans rightward across digits until only one candidate remains with a $1$ while competitors have 0$.\n * If candidate numbers tie at higher place values, the algorithm continues evaluating successive lower place values sequentially until the tie is broken.\n\n# Analog Intensity Representations and Competitive Dynamics\n\n* **Non-Digital Analog Signals:**\n * An alternative representation to discrete digital or numerical codes is an analog signal format based on continuous activation intensities, analogous to sound volume levels.\n * In an analog representation, candidate options directly drive individual processing units with proportional signal intensities rather than encoding explicit numerical digits.\n\n* **Competitive Lateral Inhibition Algorithm:**\n * Finding the maximum value in an analog intensity system is performed through mutual competitive inhibition, where each active unit attempts to deactivate all competing units.\n * Negative lateral connections mediate inhibitory interactions between competing units.\n * The strength of inhibition exerted by a unit is directly proportional to its driving intensity; highly driven units exert strong negative inhibition on rival units.\n * When options drive units simultaneously, a unit receiving a higher input (such as an intensity corresponding to 75) exerts dominant negative inhibition, suppressing the activity of weaker units over time.\n * Iterating this dynamical process forward across consecutive time steps leads to the complete deactivation of competing options, leaving the maximally driven unit as the sole remaining active option.\n\n# Classical Computational vs. Connectionist Systems in Cognitive Science\n\n* **Intuitive Cognitive Processing:**\n * Human intuitive processing rarely relies on explicit, step-by-step symbolic manipulation across tabular digit representations.\n * *Intuitive Physics*: Human observers excel at making rapid physical judgments, such as determining whether stacked block towers are stable or unstable within approximately 0.5\,\text{seconds}$$.

  • Intuitive physical assessments are poorly modeled by sequential rule-based calculations. They are better modeled as continuous parallel processing where competing potential answers receive positive or negative evidence simultaneously.

• Categorization of Cognitive Architectures:

  • Classical Computational Systems: Theoretical architectures that utilize discrete symbolic representations and operate through centralized, step-by-step sequential rule execution.

  • Connectionist Systems (Neural Network Systems): Theoretical architectures, presented in lectures by Professor Marcin, that utilize distributed networks of interconnected units interacting simultaneously in parallel without centralized or synchronized rule sequences.

Biological Implementation of Cognitive Algorithms in Neural Circuits

• Neural Intensity Encoding and Synapses:

  • Biological neural networks naturally implement competitive analog intensity algorithms.

  • Signal intensity is encoded in biological systems by a neuron's firing rate, measured as the frequency of generated action potentials (voltage spikes).

  • Under high-power microscopic visualization utilizing fluorescent dyes, biological neuron cell bodies glow bright green.

  • Neurons connect via specialized junctions called synapses, divided functionally into excitatory and inhibitory types.

  • Using targeted fluorescent staining techniques, excitatory connections are visualizable in green, while inhibitory connections are visualizable in red.

• Winner-Take-All Neural Circuits:

  • A biological winner-take-all circuit computes the maximum value across inputs using excitatory driving signals paired with reciprocal inhibitory synaptic connections between competing neurons.

  • When input choices excite competing neurons, the neuron receiving the highest level of excitatory drive fires at the highest rate, exerting strong synaptic inhibition that turns off adjacent neurons.

  • This biological circuit resolves competition over time until only the maximally driven neuron retains active firing.

• Implementation Knowledge Gaps:

  • While reciprocal inhibitory neural circuits provide a straightforward biological implementation for analog winner-take-all algorithms, the biological hardware implementation of discrete symbolic concepts (such as base 10 digit representations) is significantly less understood.

  • Symbolic processing in the brain likely relies on complex, distributed populations of neurons representing abstract numerical concepts.

Tri-Level Framework and Methodological Principles in Cognitive Science

• Tri-Level Analytical Framework:

  • Computational Level: Defines the overall functional goal, logic, and operational objectives of a cognitive system.

  • Representational and Algorithmic Level: Defines the specific information tracked by the system, the format used to represent that information (e.g., base 10, binary, or analog intensity), and the step-by-step procedures or dynamic rules executed to process it.

  • Physical/Hardware Implementation Level: Defines how the representations and algorithms are physically realized in biological structures, such as cellular firing rates and synaptic network wiring in the brain.

• Methodological Implications for Cognitive Research:

  • Understanding the mind requires integrating theoretical models across all three analytical levels; neurobiological investigation alone is insufficient to fully explain cognitive function.

  • Understanding physical brain hardware provides critical insights into algorithmic constraints, revealing direct overlap between implementation and algorithm selection.

  • Attempting to decipher cognitive function purely through empirical measurement of neural firing without top-level theoretical models is ineffective beyond basic physiological processing.

  • Top-down theoretical hypotheses regarding potential representations and algorithms are necessary to interpret observed neural firing patterns and test how computational goals are executed by biological brain tissue.