Discriminative vs Generative Models
To understand Generative AI, you need to know the two main families of AI models: discriminative and generative. They serve different purposes.
Discriminative Models (The Detectives)
Discriminative models learn the boundary between different classes. They answer: "Given this input, what class does it belong to?"
- Email spam filter: "Is this spam or not?"
- Image classifier: "Is this a dog or a cat?"
- Medical diagnosis: "Does this X‑ray show pneumonia?"
Generative Models (The Artists)
Generative models learn the underlying pattern of the data. They answer: "Given what I have learned, can I create a new example that looks real?"
- Create a new human face that doesn't exist.
- Write a poem in the style of Shakespeare.
- Generate a new piece of music.
Key Difference
| Aspect | Discriminative | Generative |
|---|---|---|
| Task | Classify / Predict | Create / Generate |
| Output | Label (e.g., "cat") | New data (e.g., an image) |
| Example Model | Logistic Regression, SVM | GAN, VAE, Diffusion |
Analogy
A discriminative model is like a security guard who can recognize if a person is allowed to enter. A generative model is like an artist who can draw a new person from imagination.
Two Minute Drill
- Discriminative models classify or predict labels.
- Generative models create new content.
- Examples of discriminative: spam filter, image classifier.
- Examples of generative: ChatGPT, DALL‑E, Midjourney.
Need more clarification?
Drop us an email at career@quipoinfotech.com