PM2: Machine Learning

Machine Learning with AI: What is Machine Learning? How do we use AI to build things? How can we start in robotics without actually knowing how to program? How can we use AI tools to modify and be creative with what tools we have? 

Dr. Milan Dahal, Tufts CEEO-Room 2

Vocabulary for Supervised  Machine Learning:

Supervised Learning
A type of machine learning where the computer is given examples with both input and correct output, and it learns to predict the output for new inputs.

Training Data
A set of examples used to teach the machine. Each example has input features and the correct answer (label).

Label
The correct answer or category for a piece of training data. The model tries to predict this.

Feature
An individual piece of information used to describe the data. For example, height and weight could be features used to predict a person’s shoe size.

Model
The result of the learning process. It’s a program that takes inputs and gives predictions based on what it has learned.

Prediction
The output the model gives when it’s asked to guess the label for new, unseen input data.

Classification
A type of supervised learning where the output is a category, like “cat” or “dog” or “spam” or “not spam”.

Regression
Another type of supervised learning where the output is a number, like predicting someone’s age or the price of a house.

Overfitting
When the model learns the training data too well—including the noise or random details—and doesn’t perform well on new data.

Accuracy
A measure of how often the model’s predictions are correct. It’s used to evaluate how well the model has learned.

Video 2: The 7 Steps of Machine Learning

Create a model to determine if a drink is wine or beer.

     2 features: Color, alcohol%

Step 1: Gather data samples- quality and amount of data will determine accuracy of model

Step 2: Data Preparation-

·      put gathered data into a table,

·      randomize the order,

·      clean up data,

·      equal amounts of samples, normalize( all data on same scale)

·      split data into training data set and test set (80/20)

Step 3: Choosing a model-

There are various models depending on the application

We have a 2-feature system- small layer model

Step 4: Training the model

Model: Y= m*X +b

where x is data, m is slope, and b is y intercept(bias)- training the model consists of finding optimal values for m and b

w= weights matrix (slope)

b= bias matrix

we start with random values for m and b, then adjust them depending upon the resulting errors

Step 5: Evaluation

Model is tested on new data(test set)

Step 6: Adjust Hyper parameters, Learning Rate( amount parameters are adjusted on each step

Step 7: Make predictions using new data