Perceptron

Question 1

A perceptron computes its output as…

Accepted Answer

A step function on a weighted sum plus bias — It calculates w·x + b and fires (1) if that exceeds a threshold, else 0 — the simplest artificial neuron.

Question 2

The perceptron is a ___ classifier.

Accepted Answer

binary linear — It separates two classes with a linear hyperplane (w·x + b = 0); it cannot represent non-linearly-separable patterns by itself.

Question 3

The perceptron learning rule updates weights…

Accepted Answer

Only when an example is misclassified — On a misclassification it nudges weights toward the correct answer (w ← w + η·y·x); correctly classified points cause no update.

Question 4

The Perceptron Convergence Theorem guarantees the algorithm will converge IF the data is…

Accepted Answer

Linearly separable by a single decision hyperplane — If a separating hyperplane exists, the perceptron is guaranteed to find one in finite steps; if not, it never converges.

Question 5

The famous limitation exposed by Minsky and Papert is that a single perceptron cannot learn…

Accepted Answer

exclusive-or, whose decision boundary cannot be linear — XOR isn't linearly separable, so no single perceptron can solve it — a result that stalled neural-net research until multilayer networks.

Question 6

How is the XOR limitation overcome?

Accepted Answer

Stacking perceptrons into a multilayer network — Hidden layers with non-linear activations let multilayer perceptrons represent non-linear functions like XOR — the basis of deep learning.

Question 7

The bias term in a perceptron allows the decision boundary to…

Accepted Answer

Shift away from the origin — Without a bias, the hyperplane must pass through the origin; the bias lets it translate to fit the data.

Question 8

The perceptron's classic activation is the…

Accepted Answer

The Heaviside step function, outputting a hard 0 or 1 — The hard threshold outputs a binary decision. Its non-differentiability is why later neurons use smooth activations trainable by gradient descent.

Question 9

Why can't the classic perceptron be trained by gradient descent on its output?

Accepted Answer

The step activation has zero gradient — The step function's flat/undefined derivative gives no useful gradient; smooth activations (sigmoid, ReLU) enabled backprop-based training.

Question 10

A perceptron is the conceptual building block of…

Accepted Answer

Artificial neural networks and deep multilayer models — Each unit in a neural network is a perceptron-like neuron; stacking and connecting them forms deep networks.

Perceptron Visualizer
Output: 1

Input Layer

Weights & Bias

Decision Boundary

Current Calculation

Decision Rule

Weight Influence

Bias Effect

One perceptron update

Test your knowledge

A perceptron computes its output as…

Interview Questions

References & Further Reading

Love this visualization?

Perceptron

Perceptron VisualizerOutput: 1

Input Layer

Weights & Bias

Decision Boundary

Current Calculation

Decision Rule

Weight Influence

Bias Effect

One perceptron update

Test your knowledge

A perceptron computes its output as…

Interview Questions

References & Further Reading

Love this visualization?

Perceptron Visualizer
Output: 1