Skill Shot · AI literacy · working
Choose a Machine-Learning Approach by Its Learning Signal
By the end you can: choose the machine-learning training approach that matches whether a task provides labels, unlabelled data, or rewards and penalties.
About 7 minutes with AryaPrerequisite: Draft: awaiting human review
Arya · AI teacher
Arya is waiting.
Scene 1 of 7 · 0:00 to 0:53
Choosing by feedback prevents a mismatch between task and learning approach.
What should guide your choice when a team wants customer groups?
Key points
- Customer behaviour can be supplied without labelled input-output relationships.
- Pattern-finding differs from predicting labelled outcomes.
- The task's learning feedback guides the choice.
Illustrative case, written for this lesson
Customer grouping
- Customer behaviour and related variables are available.
- The task is to find patterns.
- Explicit input-output labels are not described.
TakeawayStart with the learning signal the task actually provides.
Arya says (captions)
Suppose a team wants to group customers using behaviour information. The team has customer behaviour and related variables, but no labels telling the system which customer belongs to which group. Someone asks you to choose how the system should learn before the team starts. That is the practical decision in this lesson. If you choose a method that expects labelled answers, you would be asking for information the case does not provide. If you choose a method that searches for patterns, the setup fits the stated task. This is an illustrative case, not a report about a real organisation. We will use the available feedback, or absence of it, to make one disciplined choice. The goal is not to select a particular algorithm or promise a business result.
Arya is an AI, and the voice is synthetic. This is a draft not yet reviewed by a person.
Ask Arya
Stuck, or want it said another way?
Arya is an AI. She answers only from this lesson, gives hints and not answers, and asks you to try it in your own words. Please do not share personal details.
Sign in and your questions are kept with your profile; as a visitor, nothing you ask is saved.
Check yourself
Three quick questions
1. A task provides labelled examples connecting inputs and outputs. Which approach matches that signal?
2. A system receives customer behaviour data without explicit labels and searches for patterns. Which approach matches?
Take it to work
This week, take one task discussed at work and write down its learning signal before recommending a machine-learning approach. Record the evidence and one approach that does not fit.
Try it, then ask Arya
Your practice note
Write a few sentences on how you would use this on a real problem from your work: what happened, what you did, and what you expect. Arya will read it against what this Skill Shot teaches. Please leave out names and personal details.
How useful was this Skill Shot?
One tap. Your rating tells us what to improve.
Remember
- Labels connecting inputs and outputs point to supervised learning.
- Unlabelled data used to find patterns points to unsupervised learning.
- Rewards and penalties during a task point to reinforcement learning.
- Choose from the learning signal, not from the task's topic.
Watch out for
- Choosing supervised learning without checking whether examples are labelled.
- Confusing pattern-finding without labels with learning from task rewards or penalties.
- Choosing based on a business topic rather than the learning signal described.
Limits: This check classifies a training approach; it does not select a specific algorithm or define a model.
Where this fits
- Skill set
- AI literacy
- Capability module
- Machine-learning approaches
- Learning path
- New joiners and working professionals
Strategy (part 2 of 12)
Next
- Compare machine-learning problem typesAfter identifying a training approach, learn how different problem types use inputs and outputs.
Full transcript and sources
1. Choosing by feedback prevents a mismatch between task and learning approach. · 53s
Suppose a team wants to group customers using behaviour information. The team has customer behaviour and related variables, but no labels telling the system which customer belongs to which group. Someone asks you to choose how the system should learn before the team starts. That is the practical decision in this lesson. If you choose a method that expects labelled answers, you would be asking for information the case does not provide. If you choose a method that searches for patterns, the setup fits the stated task. This is an illustrative case, not a report about a real organisation. We will use the available feedback, or absence of it, to make one disciplined choice. The goal is not to select a particular algorithm or promise a business result.
2. A feedback check connects the task to its learning approach. · 53s
By the end, choose the machine-learning training approach that matches a task's feedback: labelled examples, unlabelled data, or rewards and penalties. You will make that choice by asking what the system receives while learning, not by guessing from the business label or the sophistication of the tool. For the customer grouping case, success means identifying that no input-output labels are provided and selecting the approach that finds patterns without explicit labels. You should also be able to reject two tempting mismatches. This is a small decision aid, not a complete model-design process. Use it when someone is defining how a machine-learning task will be trained. The check is observable: state the evidence in the task, name the matching approach, and explain why the other two do not fit.
3. The learning signal separates three machine-learning approaches. · 53s
Think of three ways to sort a pile of examples. In one, each example comes with an answer key. In another, there is no answer key, so you look for groupings. In the third, you try actions and receive feedback about the result. That is the plain-language distinction. The proper terms are supervised learning, unsupervised learning, and reinforcement learning. Supervised learning uses labelled data so a system can learn patterns for predictions. Unsupervised learning receives data without explicit input-output labels and finds patterns, such as customer segmentation. Reinforcement learning rewards correct results and penalises bad results while the system performs a task. This analogy helps you ask what feedback is present; it does not claim these approaches are interchangeable. The key is to classify the learning signal before choosing.
4. Checking labels and task feedback makes the choice reviewable. · 54s
Use a short decision sequence. First, name the task in practical terms: what should the system learn to do? Second, check whether examples have explicit labels connecting inputs with outputs. If so, the notes describe supervised learning. Third, if no such labels are supplied, ask whether the task is to detect patterns in data; that points to unsupervised learning. Fourth, check whether the system performs a task and receives reward for correct results and punishment for bad results. That points to reinforcement learning. If none of these signals is clear, pause and clarify the setup rather than forcing a choice. These questions are not a ranking. They separate the approaches by how learning is guided. State your evidence in one sentence, then name the approach. That makes the decision reviewable.
5. Customer pattern-finding without labels points to unsupervised learning. · 54s
Return to the customer grouping case. The task is to find patterns in customer behaviour and related variables. The case does not give labelled input-output relationships. It also does not describe a task where actions receive rewards or punishment. Apply the sequence: identify the task, check for labels, then check for task feedback. The evidence points to unsupervised learning because the algorithm finds patterns without explicit labels. This is the same customer-segmentation example described in the reference, presented here as an illustrative workplace case. A useful non-example is spam detection when emails are labelled as spam or not spam. That setup uses supervised learning, because labels guide prediction. Do not choose based only on the word customer or the word AI. Choose based on the learning signal the case actually supplies.
6. Rewards and penalties during chess learning point to reinforcement learning. · 69s
Now make one decision in a new problem. A system is learning to play chess. The rules of the game are known, and the system gets rewarded for correct results and punished for bad results. Which learning approach matches that feedback? Choose one option after the pause. Take at least six seconds to think. Option A is supervised learning; it would fit labelled examples, but those are not the signal described here. Option B is unsupervised learning; it finds patterns without explicit labels, but the case describes feedback on task results. Option C is reinforcement learning; it matches rewards and penalties while performing a task. The best choice is C. The reason is the learning signal, not the fact that chess is a game. If your choice differed, return to the evidence and classify the feedback first.
7. Naming the learning signal makes an approach choice easier to defend. · 56s
Keep the decision rule simple: labelled examples point to supervised learning; unlabelled data used to find patterns points to unsupervised learning; task results shaped by rewards and penalties point to reinforcement learning. Before recommending an approach, say what evidence the task provides. Then name the matching approach and explain one mismatch. Remember the limit: this check classifies a training approach only. It does not select a specific algorithm, define a model, or guarantee performance. The reference also notes that people set up training data, algorithms, and models for machine learning, while the machine learns from more data. For this week, use the check on one real task discussion and record the evidence before the choice. Arya is an AI; this voice is synthetic, and this lesson draft has not yet been reviewed by a person.
Sources
- S1 · Test Admin — coach's own words from an approved lesson.
- S2 · Illustrative customer-grouping case written for this lesson, based on the customer-segmentation example in S1.
Attribution: Test Admin. Version 0.1.0.