MACHINE LEARNING

How does machine learning work?

1. ML algorithms aim at uncovering patterns in a set of training data (e.g., mapping inputs and outputs) 2. A model is created out of the uncovered patterns. 3. The model's performance is tested with test data. 4. If performance are satisfactory, the model can be useful to predict new inputs.

What is machine learning?

Field of study focusing on algorithm which enable the computers to learn from data, and even improve themselves, without being explicitly programmed.

Briefly explain what overfitting is in Machine Learning and why it is problematic when training an algorithm.

Overfitting happens when the model/function is fitting to closely the training data or a limited set of data. It is problematic because machine learning models must be able to generalize, i.e., adapt properly to new, previously unseen data, drawn from the same distribution as the one used to create the model. Real data always contains some noise and if the model is overfitting it is likely to be very sensitve.

A new machine learning facial recognition system is being built to identify pensioners who are eligible to use public transport for free in Auckland City. If the system recognises your face and you are a pensioner, then you get free use of public transport; otherwise, you have to pay. The dataset that is used to train the model has a large number of images of faces but 95% of those images are of people of ethnicity A, while 5% are of people with ethnicity B. However, in the population of eligible pensioners in Auckland, 70% are of ethnicity A, while 30% are of ethnicity B. Based on this information, is there potential for this system to be unfair? If so, please describe what kind of unfairness may occur, to whom, and why.

Yes, it will be unfair to ethnicity B, as 30% of the pensioners are from ethnicity B, which contrast to the dataset that only have 5% of them. So, only 5% from the 30% ethnicity could use the public transport for free while almost everyone from the ethnicity A will get free use of public transport, since 90% from only the 70% of their ethnicity had been recognised by the system.

misuse of machine learning

correlation does not imply causation