Graded Quiz: Checklist: Vision Transformers Using Keras :AI Capstone Project with Deep Learning (IBM AI Engineering Professional Certificate) Answers 2025
1. Did you load the pretrained CNN into cnn_model and print its summary?
✅ Yes
❌ No
Explanation:
Printing the backbone summary ensures the pretrained CNN is correctly loaded and layers are available for feature extraction.
2. Did you correctly set feature_layer_name to the layer used for extraction?
✅ Yes
❌ No
Explanation:
Selecting the correct feature layer ensures the hybrid model receives the right feature tensor for transformer processing.
3. Did you implement the custom positional-embedding layer inheriting from layers.Layer?
✅ Yes
❌ No
Explanation:
Custom positional embeddings encode spatial ordering so the transformer can understand patch positions.
4. Did you define a function to combine the CNN backbone and transformer layers?
✅ Yes
❌ No
Explanation:
A builder function (e.g., build_cnn_vit_hybrid) makes the hybrid-model assembly clean, reproducible, and modular.
5. Did you configure ImageDataGenerator / tf.data for training and validation?
❌ No
✅ Yes
Explanation:
Proper data loaders with image size and batch size ensure consistent preprocessing for model training.
6. Did you instantiate the hybrid ViT model using the builder function?
❌ No
✅ Yes
Explanation:
Instantiating the hybrid model ensures the CNN + ViT architecture is fully constructed and ready for training.
7. Did you compile hybrid_model with optimizer, loss, and metrics?
❌ No
✅ Yes
Explanation:
Compilation is required before training so TensorFlow can compute gradients, losses, and accuracy.
8. Did you pass a dummy tensor through the model to verify output shape?
❌ No
✅ Yes
Explanation:
A dummy forward pass catches shape mismatches early, preventing runtime training errors.
9. Did you specify hyperparameters and launch fit()?
✅ Yes
❌ No
Explanation:
Training requires epochs, steps per epoch, and callbacks to enable full learning and tracking.
10. Did you visualize training vs. validation curves?
✅ Yes
❌ No
Explanation:
Loss and accuracy curves reveal overfitting, convergence behavior, and generalization of the hybrid model.
11. Did you complete and download the notebook?
✅ Yes
❌ No
Explanation:
Downloading the completed notebook is required for submission and ensures your progress is saved.
🧾 Summary Table
| Q# | Correct Answer | Key Concept |
|---|---|---|
| 1 | Yes | CNN backbone loaded |
| 2 | Yes | Feature extraction layer |
| 3 | Yes | Positional embedding layer |
| 4 | Yes | Hybrid model builder |
| 5 | Yes | Data generator configuration |
| 6 | Yes | Hybrid model instantiation |
| 7 | Yes | Model compilation |
| 8 | Yes | Dummy shape verification |
| 9 | Yes | Training setup |
| 10 | Yes | Learning curve visualization |
| 11 | Yes | Notebook completion |