Single Layer Perceptron Explanation

This Perceptron Simulator is a visual and interactive representation of a simplified Single Layer Perceptron. A Single Layer Perceptron is one of the oldest and most fundamental neural network models, originally proposed by Frank Rosenblatt in 1958. It's primarily used for binary classification tasks, effectively computing logical gates like AND, OR, and NOR.

Core Functionality of a Single Layer Perceptron (as Simulated)

• Inputs (Cells): The 4x4 grid of cells represents the input layer. Each cell's "on" or "off" state corresponds to a binary input (1 or 0).
• Weights (Biases): The adjustable bias values associated with each cell are the "weights" of the perceptron. These weights determine the influence of each input on the output.
• Weighted Sum: The updatePotentiometer() function calculates the weighted sum of the inputs. When a cell is "on," its bias (weight) is added to the sum. When it is off, the bias is subtracted. This is a form of weighted summation.
• Activation Function (Simplified): In this simulation, the potentiometer's needle position acts as a simplified, linear representation of the activation function's output. The clamping of the needle position, acts as a very rough simulation of a non linear activation function. The on/off status of the leds, acts as a step activation function.
• Output (Potentiometer): The potentiometer's position visually displays the output of the perceptron, which is the result of the weighted sum (with a simplified activation).
• Training (Auto-Training Buttons): The train buttons simulate a basic type of training. By adjusting the weights (biases) based on the desired output direction (positive or negative), a simplified form of weight adjustment is performed, similar to perceptron learning.

Key Concepts Demonstrated by This Simulation

• Linear Separability: This simulation demonstrates how different combinations of inputs and weights can produce varying outputs.
• Weight Adjustment: The bias adjustments show how changing the weights alters the perceptron's output.
• Simplified Model: It's important to note that this simulation is a simplified model. A real-world perceptron would typically involve more complex activation functions and training algorithms.

Relation to MNIST Dataset

While this simulation uses a 4x4 grid, the MNIST dataset involves 28x28 pixel images. A single-layer perceptron could be used to classify MNIST digits, but it would require a much larger input layer (784 inputs) and more sophisticated training. However, single layer perceptrons are limited to linearly seperable problems, and therefore have very limited use cases with complex datasets like MNIST.

In summary, this simulator provides a valuable visual and interactive way to understand the fundamental building blocks of a single-layer perceptron. It effectively demonstrates the concepts of inputs, weights, weighted sums, and simplified activation.

This project was developed by Priyangsu Banerjee.
Inspiration: Welch Labs
Source Code: View on GitHub
Connect with me: LinkedIn | Website
Last updated: 06.03.2025