Embedded Files
Machine learning (ML) and neural networks rely on fundamental mathematical ideas of sums, series, and weighted polynomials. These math concepts ultimately help calculate probabilities and guide decision-making in machine learning, particularly in image recognition. This activity allows students to better visualize the process which is undertaken within an ML perceptron, and take part in a human-based image recognition algorithm.
This Group Perceptron Activity will help you:
This Group Perceptron Activity will help you:
Visualize the process undergone during a simple probabilistic machine learning model.
Understand the necessity of varying and evaluating weights of components within a machine learning model.
Explain the mathematical model followed in basic machine learning models.
Understand the importance of spatial understanding in machine learning.
Understand the importance of intensity/color knowledge in machine learning.
Required Materials/Resources:
Required Materials/Resources:
Access to a computer and large screen (if you want share with others)
A Google account for access and ability to use Google Sheets
Keywords:
Keywords:
Perceptron - A fundamental machine learning concept which is generally used in binary classification applications. In general, a perceptron takes an input layer and maps it to an output decision.
Weight - The influence assigned to each input feature that relates to how important the feature is on the output decision. Weights are multiplied with the numerical training and test data assigned to input features to be added into a weighted sum for all inputs. The weight of a feature is adjusted through a model’s training to determine the optimal weighting structure.
Activation Function - A mathematical function that applies a decision boundary to the weighted sum obtained from the input features and their respective weights. The activation function outputs the end classification of the perceptron.
Background:
Background:
In a probability-based machine learning model, the computer analyzes various numerical inputs (x1,x2,x3…), and determines weights for each of these inputs (w1,w2,w3…). The machine then takes a sum of each of these inputs, products of the inputs and their respective weights. Finally, the machine applies an activation function which assigns a meaningful value to the calculated sum, such as the probability of a specific outcome.
In this activity, you will act as a node in a basic neural network. Many machine learning algorithms work the same way, but with far more neurons and many repeated iterations to adjust the weights of their inputs. Sums and series are the mathematical tools these algorithms use to “think,” just like the simple sum you used in the following exercise.
Activity Layout & Instructions
Activity Layout & Instructions
- Preparation
Read through the Group Perceptron Activity - Handout and ensure you understand the layout of the activity.
Make a copy of the Group Perceptron Activity - Activity Spreadsheet for your group to use for this activity.
Designate Participant Numbers to each member of your group.
- Activity
As an individual, view your 'pixel', decide on the value and weight of your pixel, and input these values into their respective cells in the Google spreadsheet.
Once everyone has input their initial guesses, the summation page will be updated and the weighted sum will be displayed.
Review the ending summation and discuss the end prediction implied by this number.
You can iterate through this activity as many times as you would like. You will notice that there is an improved and more definitive prediction for each iteration.
Downloadable Content & Links
Downloadable Content & Links
Further Reading and Resources:
Further Reading and Resources:
Learn more about machine learning and how perceptrons work.
Page updated
Report abuse