Psych 189 Lecture 5 (Artificial components and networks big picture map)

Neuron vs computation neuron

A biological neuron is a real cell that processes electrical/chemical signals. A computational neuron is a simplified math model inspired by neurons, designed to mimic signal integration and firing in a network for tasks like pattern recognition or learning.

• Biological neurons require consolidation to maintain connections; computational neurons store weights digitally and don’t naturally “forget” unless programmed to.

1. Computational neuron

Structure:

• Dendrites → receive inputs

• Soma (cell body) → sums inputs

• Axon → sends output

---

Computation:

• Add up incoming signals:

Σ(inputs)\Sigma (\text{inputs})Σ(inputs)

• Compare sum to a threshold

• Output is binary:

  • Fire (1)

  • No fire (0)

---

Fundamentals

• Neurons sum their inputs

• Inputs may be active or inactive

• Inputs may be excitatory or inhibitory

• Output depends on:

  • Input values

  • Input weights

  • Threshold

Computational Neuron: Perceptron!

What are they

• specific type of computational neuron

---

Structure

• same as computational neuron

---

Computation

• same as computational neuron

---

Fundamentals

• Make decisions based on inputs

---

Example: Should I go to a concert?

Inputs:

• Friends are going (+)

• I know the artist (+)

• Ticket is expensive (–)

• Weighted sum ≥ threshold → Go (1)

• Weighted sum < threshold → Stay home (0)

Synaptic Integration and spatial summation

• Neurons sum all incoming EPSPs and IPSPs

• If total charge exceeds threshold → action potential fires

Botox

Many pharmacological drugs influence synaptic transmission

• Botox (botulism toxin) works by blocking the release of a neurotransmitter (acetylcholine) that is required for muscle contraction

IAC models

What is the IAC model?

• Interactive Activation and Competition (IAC) network

---

Structure

• A type of connectionist model

• Nodes represent features, concepts, or people

• Nodes interact via excitatory and inhibitory connections

---

Computation

• Art is represented as a pattern of activation, not a single node

• Nodes representing Art’s features (e.g., occupation, hobbies, traits) are:

  • Excitatorily connected to the “Art” node

  • Inhibitorily connected to competing person nodes

• When the “Art” node is activated:

  • Activation spreads to connected feature nodes

  • Competing nodes are suppressed

Memory = stable activation pattern across the network, not storage in one place.

• IAC networks store knowledge in connections between units.

• Partial input can activate the full memory via excitation + inhibition dynamics.

• This is emergent — the network doesn’t have a “lookup table”; the remembered pattern arises from the interaction of nodes.

Artificial Neural Network

a layered network/system of computational neurons that learn by changing connection weights