Supervised versus Unsupervised Learning
- Roughly, supervised learning is a modeling process for some response variable that has already been observed
- On the other hand, unsupervised learning is a modeling process for some variable that has not been observed
- In other words, supervised learning refers to examples (or training data) that have already been labeled by someone who knew what they were doing
- And, unsupervised learning refers to discovering examples that have not been observed in the data yet
- Supervised learning allows us to verify the performance on the labeled training data, whereas unsupervised learning does not allow us to do this
Types of Learning Problems
| Known Classes? | Class Labels | Type of Learning Problem |
|---|---|---|
| Yes | Given for training data | Classification; supervised learning |
| Yes | Given for some but not all training data | Semi-supervised learning |
| Yes | Hints/feedback | Feedback or reinforcement learning |
| No | None | Clustering; unsupervised learning |
Description of Learning Problems
- The type of learning problem is based on having known categories and labeled examples of the categories
- Supervised learning refers to the process of categorizing data if there are known categories and labeled examples
- In this case, we could verify our classes using test data
- Semi-supervised learning refers to the process of categorizing data if there are partially known categories and labeled examples
- In this case, we could use methods involving imputation to fill in this data or remove the missing data entries
- Reinforcement learning refers to the process of categorizing data if there are known categories but no labeled examples
- In this case, we could do some kind of query, feedback, or reinforcement method by verifying our guesses about category membership (i.e. Rocchio's algorithm)
- Unsupervised learning refers to the process of categorizing data if there are no known categories and no labeled examples
- In this case, we could try discover categories which are implicit in the data themselves
Examples of Learning Problems
-
Supervised Learning
- Image recognition with completely distinguishable images
- Weather forecasting with absolutely known days of weather
- Stock forecasting with no missing days of share prices
-
Semi-supervised Learning
- Image recognition with some indistinguishable images
- Weather forecasting with some overlapping, unmarked days of weather
- Stock forecasting with missing days of share prices
-
Reinforcement Learning
- Image recognition with a pre-determined set of possible image types
- Weather forecasting with a pre-determined set of possible weather types
- Stock forecasting with a pre-determined set of possible stock directions (i.e. up or down)
-
Unsupervised Learning
- Image recognition of subgroups of images that haven't been labeled yet
- Weather forecasting of subgroups of weather types that haven't been labeled yet
References
Next