Difference between revisions of "Artificial Intelligence"
Line 1: | Line 1: | ||
− | + | =Types of AI= | |
==Symbolic== | ==Symbolic== | ||
Trying to capture intelligence in rules (Marvin Minsky, John McCarthy, Oliver Selfridge, Ray Solomonoff) | Trying to capture intelligence in rules (Marvin Minsky, John McCarthy, Oliver Selfridge, Ray Solomonoff) | ||
Line 14: | Line 14: | ||
===Reinforcement=== | ===Reinforcement=== | ||
Input and Critic. Constant feeding of new data, try to evaluate and have the result checked (by humans). | Input and Critic. Constant feeding of new data, try to evaluate and have the result checked (by humans). | ||
+ | |||
+ | =Conversational AI= | ||
+ | Important as natural language processing is increasingly important e.g. for digital assistants (Alexa, Siri). | ||
+ | ==Rule Based== | ||
+ | Based on rules find the answer to a question. Only predefined answers can be found and only answers can found for questions matching the rules. | ||
+ | |||
+ | ==Self Learning== | ||
+ | ===Retrieval based=== | ||
+ | The system is trained with questions and answers. It can analyze any question and finds the most relevant answer(s). | ||
+ | ===Generative=== | ||
+ | The system understands language and can generate answers itself. | ||
+ | |||
+ | |||
+ | |||
+ | =Classification= | ||
+ | * Features and labels | ||
+ | : The system is trained with labeled text is split in features. The features get a weight. | ||
+ | : By finding the features in text the system can determine what it is about. | ||
+ | |||
+ | * Extract entity and find intent | ||
+ | : Entity's can be features (see above) | ||
+ | |||
+ | * Word embeddings | ||
+ | : Words represent part of the context. Like features contexts get a value, this leads the system to the grasp of the question. |
Revision as of 22:30, 9 October 2018
Types of AI
Symbolic
Trying to capture intelligence in rules (Marvin Minsky, John McCarthy, Oliver Selfridge, Ray Solomonoff)
Subsymbolic
Learn to understand the world by trial and error. (Rosenblatt).
Neural Network, later called Machine Learning.
Supervised
Feed input/output pairs
Unsupervised
Input only. The system learns from finding commonalities in the data presented.
Reinforcement
Input and Critic. Constant feeding of new data, try to evaluate and have the result checked (by humans).
Conversational AI
Important as natural language processing is increasingly important e.g. for digital assistants (Alexa, Siri).
Rule Based
Based on rules find the answer to a question. Only predefined answers can be found and only answers can found for questions matching the rules.
Self Learning
Retrieval based
The system is trained with questions and answers. It can analyze any question and finds the most relevant answer(s).
Generative
The system understands language and can generate answers itself.
Classification
- Features and labels
- The system is trained with labeled text is split in features. The features get a weight.
- By finding the features in text the system can determine what it is about.
- Extract entity and find intent
- Entity's can be features (see above)
- Word embeddings
- Words represent part of the context. Like features contexts get a value, this leads the system to the grasp of the question.