Learn how AI actually works.

iiNNoo is a free set of hands-on, interactive lessons that teach how neural networks learn. Every node shows its value and every connection shows its weight, so you can follow each number through the network and see the math, all of it elementary. No experience required.

Unit 04 · 21 lessons

Examples

Examples are where the three units come together, in networks built to do a real job. These are complete neural networks, from spam filters to traffic lights, to run and explore freely, with no task to finish.

Open the Examples

Section 1 · 7 lessons

Small Networks

  1. The Tree Growth Predictor 01

    This network predicts how tall a tree will be next year. The input value is the tree's height today. The output value is its height next year. The weight is the growth rate.

    Open lesson

  2. The Lemonade Stand Predictor

    This network predicts how many cups a lemonade stand sells. The input value is the temperature in degrees. The output value is the cups sold. The weight is how much each degree adds. The bias is what sells even at freezing. Edit either one and watch the output values change.

    Open lesson

  3. The Video Views Predictor

    This network predicts how widely a video has been watched. The input value is the days since it was posted. The output value is the views. The bias is the views it starts with, and the weight is what each day adds. Step through all 4 input values.

    Open lesson

  4. The Flood Predictor

    This network predicts how far a river will spill over its bank. The input value is the water level upstream, in cm. The output value is the cm it spills over. The bias of -40 and the activation function make a threshold together, so below 40 it predicts no flooding at all. It may help to zoom out (-) to see all input / output pairs.

    Open lesson

  5. The Honey Harvest Predictor

    This network predicts how much honey a hive makes. The input values are the flowers growing nearby and the days in the month. The output value is the kilograms of honey. One weight is what each flower is worth, and the other is what each day of the month adds.

    Open lesson

  6. The Tree Growth Predictor 02

    This network predicts how tall a tree will be next year. The input value is the tree's height today. The output value is its height next year. This time the network learns the weight, the growth rate, by itself. Step through the training data and watch the weight go from 1 to 1.27, then try the testing data.

    Open lesson

  7. The Crop Yield Predictor

    This network predicts a field's yield. The input values are the rain in cm and the bags of fertilizer. The output value is the yield, in tons. Step through the training data and watch the 2 weights. It works out that fertilizer counts for more than rain. Then switch to testing data to see how close the errors are to zero.

    Open lesson

Section 2 · 7 lessons

Medium Networks

  1. The Cafe Line Predictor

    This network predicts how long you will wait in line at a cafe. The input values are the coffees, teas and meals ordered ahead of you. The output value is the wait in minutes. 3 connections arrive at 1 node, and each weight is a number of minutes.

    Open lesson

  2. The Festival Turnout Predictor

    This network predicts what a festival needs. The input value is the number of bands. The output values are the visitors, meals and liters of water. 1 input value travels down 3 connections and becomes 3 output values. Each weight is that total's share per band.

    Open lesson

  3. The Overheating Detector

    This network spots an engine about to overheat. The input values are its heat, its load and how hard the fan is working, each scored from 0 to 10. The output value is the alarm. The fan's weight is negative, so it pushes back. Below the threshold the activation function reports nothing at all.

    Open lesson

  4. The Car Trip Predictor

    This network predicts what a car trip takes. The input values are the distance in km and the traffic, scored from 0 to 5. The output values are the hours and the liters of fuel. There are four weights, one for each connection, so each output value depends on both the distance and the traffic. Traffic costs time more than fuel.

    Open lesson

  5. The Spam Filter

    This network tells spam from a real message. The input values are the links, the shouted words and the polite words in it. The output values are the spam and safe ratings. Links and shouting push one way, politeness the other. It may help to zoom out (-) to see all input / output pairs.

    Open lesson

  6. The Recycling Sorter

    This network tells glass from plastic. The input values are how heavy and how shiny a piece is. The output values are the glass and plastic ratings. Step through the training data and watch the two output values separate. The network picks whichever output value is larger.

    Open lesson

  7. The Instrument Recognizer

    This network recognizes a flute, a violin and a drum. The input values are the pitch, the brightness and how sharply a note starts. The output values are flute, violin and drum. Step through the training data and watch each instrument come to own the measurement it is loudest on.

    Open lesson

Section 3 · 7 lessons

Large Networks

  1. The Seed Sprouting Predictor

    This network predicts how fast a seed sprouts. The input value is the temperature. The output value is the growth in mm a day. Seeds grow best at 25 degrees and not at all below 10 or above 40. Each middle node switches on at one of those temperatures. Step through all 5 temperatures.

    Open lesson

  2. The Traffic Light Decider

    This network decides how long each of 3 roads gets a green light. The input values are the cars waiting on the north, east and south roads. The output values are the 3 green light times. Each road has its own path through a middle node. Click on a connection to see its weight.

    Open lesson

  3. The Smoothie Nutrition Predictor

    This network predicts what is in a smoothie. The input values are the bananas and cups of berries. The output values are the calories and fructose. Observe the errors shrink as you step through all 32 smoothies. Then switch to testing data to try 4 new recipes.

    Open lesson

  4. The Tide Predictor

    This network predicts the tide. The input values are the moon's distance and the wind, each scored from 0 to 5. The output values are the tide's height in meters and the hours until it turns. This network has a middle layer, so back propagation runs here. Step through all 24 readings and watch the weights and biases settle.

    Open lesson

  5. The Clothing Advisor

    This network suggests what to wear. The input values are how cold, rainy and sunny it is. The output values are a coat, an umbrella and sunglasses. Observe it learn as you step through all 24 days, then switch to testing data. Look for the day it gets wrong.

    Open lesson

  6. The Air Quality Monitor

    This network turns 4 roadside readings into 4 pollutant levels. The input values are the traffic, wind, temperature and humidity, each scored from 0 to 5. The output values are the dust, nitrogen, ozone and carbon. With 255 steps a round, try swift mode with this one. Train it, then switch to testing data to confirm it learned correctly.

    Open lesson

  7. The Next Word Predictor

    This network predicts the next word in a sentence, like a tiny version of the AI that writes text. The input value is a word, and the output values show which word comes next. It learns from 7 short sentences, such as “the dog is cute” and “the sky is blue.” Train it in swift mode, then switch to testing data. Which words does it expect after “is”?

    Open lesson

Ready to learn how AI learns?

Start Learning How AI Learns