Ml Notes 01

Machine Learning Landscape

1. What is machine learning?

   • If you download a copy of all wikipedia articles, your computer has a lot more data. But it is not suddenly better at any task.

   • Field of study that gives computers the ability to learn without being explicitly programmed

   • a computer program is said to learn from experience E with respect to some task T and some performance measure E, if its performance on T as measured by P improves with experience E

2. Examples of Machine Learning System

   • Spam Filter

   • A ML program given examples of spam and regular emails flagged by users, can learn to flag spam.

   • Examples ML system uses to learn are called training set, each example is called training instance or sample

   • Part of the system that learns and makes predictions is called model

3. Why use Machine Learning?

   • Problems that evolves and complex to maintain

   • Problems too complex to solve with traditional approaches or have no know algorithm

   • Discover hidden patterns in data

4. Example Applications

   • Analyzing images of production line

   • Detecting tumors in scans

   • Classify news articles

   • Forecasting revenue next year

   • Responding to voice commands

   • Fraud detection

   • Customer segmentation

5. Types of Machine Learning Systems

   Classified in broad categories based on following criteria

   • How they are supervised during training

   • Can they learn on the fly

   • How they work. Compare with known data, or by building a predictive model.

6. Training Supervision

   • Supervised Learning
   • Unsupervised Learning
   • Semi supervised Learning
   • Reinforcement Learning
7. Supervised Learning
   • Data you feed to the algorithm includes desired solution called labels
   • Typical supervised learning tasks are classifications (spam filter)
   • Another typical task is to predict a target number. (price of a car)
8. Unsupervised Learning
   • Training data is un-labelled and the system tries to learn without supervision
   • Typical task is to cluster (customers into segments)
   • Another typical task is anomaly detection, novelty detection
   • Another typical task is association rule learning
9. Semi-supervised Learning
   • Training data is partially labelled
   • Clustering algorithm may be used to group similar instances and then label them with common label in the cluster
   • Once the whole dataset is labelled it is possible to use any supervised learning algorithm
10. Reinforcement Learning
    • Learning system observe an environment, select and perform an action and then get rewards in return
    • It must then learn what is the best strategy (policy) to get most reward over time
11. Batch Learning
    • System is trained with available data offline, and then launched into prod
    • Model performance decays over time (model drift) case the world evolve while model remains unchanged
    • Update the data and train a new version fo the system
    • Requires lot of computing resources
12. Online Learning
    • Train the system incrementally with small groups of data called mini batches
    • Systems that needs to adapt rapidly, how fast they adapt is called learning rate
13. Instance based
    • Learn by heart, remember training instances
    • Use a measure of similarity to compare them with learned examples to classify new instance
14. Model based Learning
    • Generalize from training set and build a model to make predictions
    • Example: Linear regression
      • study data
      • select a model
      • train with data
      • apply model to make predictions