by Albert Major
This lecture is about how different types of artificial neurons—like McCulloch-Pitts neurons (fixed logic), perceptrons (learnable weights), and multi-layer perceptrons (deep learning)—model brain-inspired computation to form the basis of modern AI.
This slide introduces the biological basis of neural networks with a quote emphasizing that the brain is a vast network of interconnected neurons that gives rise to thought, memory, and emotion. It connects biology to AI by showing what inspired early artificial neurons—the idea that intelligence and behavior emerge from simple, interconnected units.
Neurons are the basic computational units of the brain.
This slide explains how biological neurons work and why they inspired artificial ones. It shows that signals travel from dendrites to the cell body, down the axon via an action potential, and end in the release of neurotransmitters across the synapse. The key takeaway is the all-or-none firing mechanism—either a neuron fires or it doesn’t—which became the basis for binary artificial neurons like McCulloch-Pitts models.
Structure:
This all-or-nothing behavior inspired the creation of artificial neurons.
This slide presents the foundational assumptions from McCulloch and Pitts’ 1943 paper, which treated neural activity as **logical computation**. Because neurons follow an **all-or-none firing** rule, the authors proposed that neural events could be modeled using **propositional logic**. They assumed that a neuron fires only if a fixed number of **excitatory synapses** are active within a brief time window (latent addition), independent of input position or prior activity. The only significant delay comes from **synaptic transmission**, not electrical signaling. Crucially, they also stated that **inhibitory inputs** entirely prevent a neuron from firing, and that the **network’s structure is static**—there’s no learning or rewiring over time. These simplifications allowed them to model neurons as **linear threshold gates**, laying the groundwork for viewing neural networks as **universal computers**, though highly abstracted from biological reality.
This slide lists the core assumptions of the McCulloch-Pitts neuron, an early model of artificial computation inspired by biology. It simplifies neurons as binary threshold units that fire in an all-or-none fashion, based on whether enough inputs (synapses) are active. It ignores biological complexity—assuming fixed structure, no learning, only synaptic delay, and hardwired logic. These rules laid the foundation for treating neurons as logic gates and linking neural activity to propositional logic.
This slide shows the **formal equation** for a McCulloch-Pitts neuron: it outputs 1 if the weighted sum of binary inputs exceeds a **threshold** (t), and 0 otherwise. This is called a **linear threshold gate (LTG)** and functions as the neuron's **activation function**. It mathematically captures the **all-or-none firing** idea and lays the foundation for building logic gates using neurons.
McCulloch-Pitts neurons are simple artificial models that output 1 if the weighted sum of binary inputs reaches a threshold, unlike real biological neurons which use complex, continuous signals, adapt over time, and support learning and memory. (We dont use these really anymore)
This diagram shows how simple artificial neurons (McCulloch-Pitts neurons) can do basic logic. Each neuron adds up its inputs, and if the total is 0 or more, it "fires" (outputs 1). For AND, both x and y must be 1 to reach the required total. For OR, just one of them being 1 is enough. For NOT, the input x is flipped using a negative weight, so the neuron only fires when x is 0. The little numbers are weights (positive = excitatory, negative = inhibitory), and the black circle is a built-in always-on input to help set the threshold.