- LPU MCA Semester 1
- CAB107
- Unit 4
CAB107 · Unit 4: Machine learning fundamentals
Supervised, unsupervised and reinforcement learning
10 practice questions · in the ETE syllabus
Key points
- Supervised learning needs labels
- Unsupervised learning finds structure without labels
- Reinforcement learning learns from rewards
- Classification is discrete, regression is continuous
Machine learning has three main paradigms.
| Paradigm | Data | Goal | Example |
|---|---|---|---|
| Supervised | Features plus known labels | Predict the label | Spam detection, house prices |
| Unsupervised | Features only | Find structure | Customer segmentation |
| Reinforcement | Agent interacts, receives rewards | Maximise cumulative reward | A program learning chess by self-play |
Supervised learning splits into classification, which predicts a discrete category, and regression, which predicts a continuous number. A weather model that outputs a probability of 0.7 for rain is still classification, because the target is the category rain or no rain and the probability is only a score.
Unsupervised learning includes clustering and dimensionality reduction; no target exists.
In reinforcement learning there is no labelled correct action at each step. The agent learns by acting and receiving rewards, for example +1 for a win and -1 for a loss, so a chess agent playing millions of self-games is reinforcement learning, not supervised learning.
Semi-supervised learning mixes a small labelled set with a large unlabelled one.
Exam trap: a numeric output does not make a task regression. Look at the target: continuous value means regression, category means classification.
Try two questions
How does reinforcement learning differ from supervised learning?
- An agent is told the correct action for every single state by a human teacher at each step of play
- The agent learns from a fixed table of labelled images that is prepared in advance by human annotators
- An agent learns by acting and receiving rewards, with no labelled correct action for each step
- The agent groups the data into clusters without any feedback, reward or goal being involved
Show the answer
C. Reinforcement learning optimises long-term reward through trial and error.
Why not A: Per-step correct answers describe supervised learning.
Why not B: A fixed labelled table describes supervised learning.
Why not D: Grouping without feedback describes unsupervised learning.
A retailer wants to split 10,000 customers into behaviour groups, but no group labels exist. Which approach fits?
- Supervised classification such as logistic regression
- Unsupervised clustering such as K-Means
- Linear regression on the group number
- Reinforcement learning with a reward for each purchase
Show the answer
B. Without labels, grouping by similarity is a clustering task.
Why not A: Classification needs labelled groups, which do not exist.
Why not C: There is no numeric target to regress on.
Why not D: Nothing here is an agent acting in an environment.