It earns a place when
- Use diffusion when high-quality conditional generation and controllable sampling are valuable.
- You can evaluate it against a credible baseline.
- Its output fits the decision and data constraints.
Machine learning models · Phase 5 · Lesson 44
Diffusion models
Understand diffusion as learning to reverse a gradual noising process.
Lesson goal
Understand diffusion as learning to reverse a gradual noising process.
The core idea
Training adds noise to real data. The model learns how to remove it. Generation starts from noise and repeatedly denoises toward a sample.
Mental model
Picture it this way. Training adds noise to real data. The model learns how to remove it. Generation starts from noise and repeatedly denoises toward a sample. The important question is what assumption this picture makes, and whether that assumption fits the data.
Mathematical core
A forward process adds Gaussian noise over timesteps. A denoising network predicts noise or a related target, and sampling follows a learned reverse process.
Worked example
Text-to-image diffusion starts from a noisy latent, uses text conditioning to guide denoising and produces an image over many steps.
When to use it
Do not make it the default when
- Do not treat a generated image as evidence that the model understands the prompt or has not memorised training data.
- A simpler model has not been tested.
- The data or target definition is still unclear.
Failure modes
Watch for this. Sampling cost, prompt adherence, bias, copyright concerns and small-object errors matter. Latent and pixel-space choices change quality and cost.
When a result looks surprisingly good, inspect the split, target timing, error slices and data-generating process before celebrating.
Practice
Use this as a small experiment rather than a recipe to copy blindly. Change one thing, record the result and explain the change.
Do this. Run a small diffusion tutorial or inspect denoising checkpoints. Vary the number of sampling steps and compare quality against latency.
Retrieval check
Answer from memory first. The buttons reveal feedback, but the durable step is explaining why.
1. What does diffusion training teach?
2. Where does generation begin?
3. What does more sampling steps usually trade?
Transfer prompt. Describe one real problem where this model or idea would be a sensible candidate. Name the target, the main risk and the metric you would inspect.
Primary source
Denoising Diffusion Probabilistic Models. Use the source for the deeper treatment after you can explain the lesson's core idea without looking.