Depending on what kind of adaptive program you have in mind, we can have three kind of learning histogram based on right and wrong user's answers.
- Positive Learning : Higher value shall only be given to the histogram if the user answers correctly. The histogram is updated only when success response happen. This type of learning is useful for preference or satisfaction level update. Behind the scene, we can explain the automatic marketing survey in simple term. Each item (i.e. book) has several attributes (e.g. mathematics and statistics, computer vision field, Image processing etc.). When the user selects certain item, the program will update the learning histogram of the user for that attributes with higher weight. The program is then search for more items with similar attributes that the user prefers. As the result, the program may offer more items that the users like most.
- Negative Learning : Higher value shall be given to the histogram only if the user gives wrong answers. For example in our Hiragana game, if the user makes many mistakes on certain syllables, those syllables will be asked more frequently rather than the syllables that users already know. This is a kind of punishment game without any reward.
- Reward and Punishment Learning: Higher value will be give to the histogram when the user response correctly and if the user give wrong answer, the histogram may get lower value. The weight of positive and negative update may not be the same. This kind of adaptive learning may be better response than positive or negative learning and extremely useful for many applications, from game programming, science and engineering to marketing and education. A care must be given so that the probability resulted from the histogram will not become negative or higher than one. See our of this learning
Preferable reference for this tutorial is
Teknomo, Kardi (2015) Learning Algorithm Tutorials. http:\\people.revoledu.com\kardi\ tutorial\Learning\