Machine Learning — 1

Machine learning can be divided into different types based on the information they use:

1. Supervised learning

• Data contains desired output/result.
• {(x, y)} : we have correct label for the training data.

2. Unsupervised learning

• Data does not contain desired output/result.
• training data is of {(x)} form. we do not know what is correct output for the data.

3. Semi-Supervised learning

• some data includes desired output/result.
• we have some data in the form {(x, y)} and {(x)}
• It allows us to do better with both labelled and unlabeled data.

4. Reinforcement learning

• Data includes some indirect measure.
• {(x, evaluation of output g(x))}, that is it gives us estimation of how good the result is, we still don’t know what the right answer is.
• We do not know if the result is acceptable or not.
• The evaluation can be used for comparison and pick the best.

Supervised Learning vs Neural Network

• Supervised learning in ANNs involves a mechanism of providing the desired outputs with the corresponding inputs.
• In neural network algorithms, the supervised learning process is improved by constantly measuring the resulting outputs of the model and fine-tuning the system to get closer to its target accuracy.
• The level of accuracy obtainable depends on two things: the available labeled data and the algorithm that is used.

Machine Learning algorithms based on types:

Machine learning Algorithms have three main elements:

• Representation: what the model looks like; how knowledge is represented. It’s the characterization of space of all possible results it can give.
• Representations can be: set of Instances, Decision trees, Set of rules, Graphical models, Neural networks, Support vector machines, Models of Ensembles, these are the representation of what the learned functions are.
• Evaluation: how good models are differentiated; how programs are evaluated.
• Accuracy/error, Precision/recall, Likelihood, Cost/Utility, Margin, Entropy, Similarity .,
• Optimization: the process for finding good models; how programs are generated. Finding out how to improve the performance.
• Different optimization approaches are:
• 1. Combinatorial Optimization: Find global best solution.
• 2. Local optimization: Locally optimize, Gradient Ascent/descent
• 3. Local improvement: Locally improve the solution.
• 4. Constrained Optimization: Optimize within given conditions.

Reference:

1. Lecture notes by Prof. Manfred Huber