Learning

Learning is the sub field of AI and intersects with cognitive science, information theory, and probability theory, among others. Machine learning is the study of how to build computer system that adapt and improve with experience. It represents inductive reasoning. Learning is one of the most important characteristics an agent must posses in order to claim intelligent.
Definition by Simon (1983)
“changes in the system that are adaptive in the sense that they enable the system to do the same task or tasks drawn from the same population more efficiently and more effectively the next time.”
A general model of learning agent
A learning agent can be divided into four conceptual components as shown in fig below:
1) Critic:

• The critic tells the learning element how the agent is doing with respect to a fixed performance standard
• Critic is necessary because the percepts themselves provide no indication of the agent success e.g. a chess program could receive a precept indicating that it has checkmated its opponents but it needs a performance standard to know that this is good thing; the percept itself does not say so.

2) Learning Element:

• Responsible for making improvements
• Design of learning element depends very much on the design of the performance element.

3) Performance Element:

• Performance element is responsible for selecting external actions (It takes in percepts and decide on actions).

4) Problem Generator:

• Responsible for suggesting actions that will lead to new and informative experiences.
• It helps the agent to explore a little so that it might discover better actions for certain situations. The problem generator suggests such a explanatory actions. This is what scientists do when they carry out experiments. Galileo did not think that dropping rocks from the top of tower of Pisa  was valuable itself. He was not trying to break the rocks, nor to modify the brains of unfortunate passer-by. His aim was to modify his own brain, by identifying a better theory of the motion of objects.

Types of Learning
1) Rote learning:

• method of study based on learning facts etc by heart without considering their meaning.
• most trivial form of learning
• simple storing of computed information (Everything is memorized ) e.g. computer simply stores a piece of data in knowledge base.
• Many computer program e.g. database systems can be said to learn. The act of storage allows the program to perform better in future.
• There may be situations where computation is more expensive than recalling the previously done job. Thus remembering the previous would certainly help performance better. This form of learning is called rote learning.

Capabilities of rote learning
Though rote learning is very simple and does not require sophisticated problem solving capabilities, it should have following capabilities

• Organize storage of information: there must be a mechanism to access the appropriate information very quickly
• Generalization: to keep the number of stored information down to a manageable number, the number of distinct objects must be generalized and stored.

2. Learning by taking advice

• A computer can do very little without a program for it to run. When a programmer writes a series of instructions into a computer, a rudimentary kind of learning is taking place.
• In this scenario, the programmer is a kind of teacher and the computer is a learner, Now if the program is written in high level language, it needs to be interpreted and compiled before it can be executed. Thus it is necessary for the system to operationalize the knowledge before it can use it.
• This form of learning where the teacher gives instruction into high level language and the system converts it to machine understandable form and uses in problem solving is known as “learning by taking advice”.

3) Inductive Bias Learning (Learning by Example)

• The real world problem domain tends to be very large. Hence, a learner usually only examines a fraction of all possible examples. From this limited experience, the teacher must generalize correctly to unseen instance of the domain. This is the problem of induction.

Inductive Bias

• Refers ti any criteria a learner uses to constrain the concept space or to select concepts within that space
• Example- For a certain football , we have properties like round, big, and heavy an for certain ball like ping pong bal, we have properties like round, small, light bouncy.
• Now from the  above information, the knowledge base should extract certain properties which uniquely identify the ping pong ball. The learner can use a bias that does not consider the color of the object and that consider only the shape and bouncing properties of the object. The properties that must be stored are : ” Round and Bounce”

4) Winston’s learning Program

• structural concept of learning program
• This program operated in a simple block world domain. Its goal was to construct representations of the definitions of concepts in the block domain.

Learning Algorithm
1) Supervised Learning

• a teacher or oracle is available which provides a desired action corresponding to a perception.
• provides training set

2) Unsupervised Learning

• no teacher is available
• learner only discovers persistent patterns in the data consisting of a collection of perception
• also called exploratory learning

3) Active Learning

• Here not only a teacher is available, the learner has the freedom to ask the teacher for suitable perception-action example pairs which will help the learner to improve its performance.

4) Reinforcement Learning

• a teacher is available, but the teacher instead of directly providing the desired action corresponding to a perception, return reward and punishment to the learner for its action corresponding to a perception.