Introduction

1. Introduction

What is Machine learning?

• That gives computers the ability to lean without being explicitly programmed
• You don’t have a full control of program anymore – as it will produce diff outcome

When to use machine learning?

• Personalization- what is it that YOU are looking for?
• Field of language- translation, Speech to text and text to speech
• Online shopping – suggesting things you like, Netflix (confirms the prediction, if its wrong, the system learns)
  • Contest to improve the algorithm by 10%
• Self-driving – distinguishing pedestrians from the road

AI

• Supervised learning is mostly used to interpret larger amount of data
• Reinforcement learning – has to do with control. Once you have the understanding of the world around you behaviour will change. It won’t lead you to same conclusion. (finds max reward)
• When to write a normal program and when to write a ML program?
  • For normal programs, we have an algorithm, we implement it, and it outputs expected outcome.
  • ML becomes important when you don’t know what exactly you want, but you could provide examples
    • what music do you like? what exactly do you like about that genre?
    • Gather ‘features’, and use them to describe what you acc like

Supervised learning

• Regression- learning the pattern of the function

  • Predictions from the past, and deriving predictions for the future
  • A graph that has a solid trend – would match with the model when new data points are given

• Classification – shifting the burden of programming to throwing bunch of examples

  • e.g. distinguishing chair from a desk
  • Start by choosing features – stuff that differentiates the object
  • Then pass those features to algorithm classification

Unsupervised learning – we are still looking for categories, but the labels aren’t given

• Clustering – these points belong together however, human beings have to go through them and evaluate
• Reinforcement Learning – used to find the best possible path it should take in a specific situation. There is no answer, but the reinforcement agent decides what best to do to perform given test.

Machine learning process

1. Data collection
   • you need some data to work with, are also responsible for gathering data. you need variety of data. E.g. face classification – you need other ethnic groups too
   • aware of the fact that all classes should be represented
2. Feature selection
   • When you have bunch of raw data what are you gonna look at? – e.g. number of leg, surface, sizes etc
   • Deriving the distance btwn the ppl in the video – they recognize shaking hands, throwing stuff, etc depending on data, you need to be creative to see what it is that you r tryna look at
   • New way – deep learning! you really just give the neural network the image, and that will figure what’s important
3. Algorithm choice
4. Training
5. Evaluation- tbat to tell how good the model is