4.24 Learning paradigms, and what a network is asked to do
You can distinguish supervised, reinforcement and unsupervised learning by the feedback each one receives, and name the learning tasks: pattern association, pattern recognition, function approximation, control, filtering and beamforming.
Before:01a. Linear Algebra01b. Calculus and Optimisation01c. Probability01d. Statistics and Inference02. Python — Basics to Advanced
The three paradigms differ in one thing only, and it is not the network: it is what comes back from the environment. A teacher supplies the desired answer, a critic supplies a single number, and unsupervised learning gets nothing at all and must find structure on its own. The tasks that follow are the shapes a problem takes, and beamforming is worth attention here because it is the point where an electronics background meets this material directly.
Work through these
Describe learning with a teacher, and the error surface it descends
The teacher supplies the desired response, the error drives the weights, and the mean square error over the weights is a surface the operating point moves down. It is a closed-loop system with feedback.
Describe reinforcement learning, and the role of the critic
No desired answer is given; a single number says how well things went, and the critic turns it into something more useful. It sits between the other two paradigms and learns through continued interaction.
Describe unsupervised learning, and what it can find without any feedback
No teacher and no critic, so the network can only find structure that is already in the data. Clustering is the familiar example, and it is descriptive rather than predictive.
Contrast the three paradigms on feedback, data and what they produce
Labelled examples and a predictive result on one side, hidden structure and a descriptive result on the other, and interaction with an environment between them. The contrast is about feedback, not architecture.
Name the learning tasks: pattern association, pattern recognition, function approximation
Storing and recalling pairs, assigning an input to one of several classes, and fitting an unknown mapping from examples. Most applied problems are one of these three wearing local clothes.
Describe control, filtering and beamforming as learning tasks
Driving a system towards a desired output, cleaning a signal, and steering sensitivity towards a direction of interest. Beamforming is where radar, sonar and the echolocating bat all meet this material.
Distinguish autoassociation from heteroassociation
Recalling a stored pattern from a noisy or partial version of itself, against recalling a different pattern that was paired with it. The first is a memory and the second is a mapping.
Sign in to keep your progress.
Free resources
Links last checked 29 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.