Neural Network Playground
Run forward passes and train a small network step by step — weights, activations and loss, visualised.
How to Use This Tool
Choose Input Data
Enter values or select from preset examples to feed into the network
Watch Forward Pass
See how data flows through layers, weights multiply inputs, and activation functions transform results
Observe Learning
Watch backpropagation adjust weights based on prediction errors to improve accuracy
Experiment with Settings
Adjust learning rate, activation functions, and network architecture to see how it affects training
Interactive Neural Network
Input Data
Quick Examples:
Network Settings
Inspired by the Brain
Neural networks are computing systems inspired by biological neural networks. Just like brain neurons connect and communicate, artificial neurons process and pass information through weighted connections.
Pattern Recognition
They excel at finding patterns in data by learning from examples. Each neuron receives inputs, applies weights, and produces outputs that help recognize complex patterns.
Learning Through Training
Networks improve their performance by adjusting connection weights based on training examples, gradually getting better at making accurate predictions.
Universal Problem Solvers
With enough neurons and training data, neural networks can solve diverse problems from image recognition to language translation and decision making.
1. Input Layer
Data enters the network through input neurons. Each input represents a feature or characteristic of the data (like pixel values in an image).
2. Weighted Connections
Each connection between neurons has a weight that determines how much influence one neuron has on another. These weights are the network's "memory."
3. Hidden Layers
Hidden neurons combine weighted inputs, add bias, and apply activation functions to transform data into increasingly complex representations.
4. Output Layer
Final neurons produce the network's prediction or decision, such as classifying an image or predicting a value.
5. Learning (Backpropagation)
When predictions are wrong, the network adjusts weights backwards through layers to reduce errors and improve future performance.
Neurons (Nodes)
Basic processing units that receive inputs, apply transformations, and produce outputs. Each neuron has an activation value representing its state.
output = activation(Σ(weight × input) + bias)
Weights
Numbers that determine the strength of connections between neurons. Positive weights excite, negative weights inhibit. Learning adjusts these values.
Bias
Additional parameter that shifts the activation function. Helps neurons activate even when inputs are zero, adding flexibility to learning.
Activation Functions
Mathematical functions that determine if and how strongly a neuron should activate based on its inputs. They introduce non-linearity.
Forward Propagation
Data flows forward through the network:
- Input values enter the first layer
- Each neuron calculates: (input × weight) + bias
- Activation function transforms the result
- Output becomes input for the next layer
- Final layer produces the prediction
Error Calculation
The network measures how wrong its prediction was:
- Compare prediction to actual answer
- Calculate loss using a loss function
- Larger errors mean more learning needed
Loss = (Prediction - Actual)²
Backpropagation
The network adjusts weights to reduce errors:
- Calculate how much each weight contributed to the error
- Adjust weights in proportion to their responsibility
- Work backwards from output to input layers
- Use gradient descent to find better weight values
Feedforward Networks
Simple networks where information flows in one direction. Great for basic classification and regression tasks like our playground example.
Recurrent Networks (RNN)
Networks with memory that can process sequences by maintaining information from previous inputs. Ideal for time-series and text data.
Convolutional Networks (CNN)
Specialized for processing grid-like data such as images. Use filters to detect features like edges, shapes, and patterns.
Transformer Networks
Advanced architecture using attention mechanisms to process sequences in parallel. Powers modern language models like GPT and BERT.
AI Assistants
ChatGPT, Claude, and other conversational AI systems use massive neural networks to understand and generate human-like responses.
Autonomous Vehicles
Self-driving cars use neural networks to process camera feeds, detect objects, and make driving decisions in real-time.
Medical Diagnosis
Neural networks analyze medical images, predict disease outcomes, and assist doctors in making more accurate diagnoses.
Content Creation
AI generates music, art, videos, and text by learning patterns from vast datasets of creative works.
Financial Trading
Trading algorithms use neural networks to analyze market patterns, predict price movements, and execute trades.
Search & Recommendations
Search engines and recommendation systems use neural networks to understand user intent and suggest relevant content.