Project: Build, Train, and Benchmark a CNN for CIFAR-10

Your goal is to develop a Convolutional Neural Network (CNN) that achieves the lowest possible validation loss on the CIFAR-10 dataset.

Performance Tiers (Based on val_loss)

Validation Loss Range	Points
< 0.63	10 pts
0.63 – 0.70	9 pts
0.70 – 0.80	8 pts
0.80 – 0.90	7 pts
0.90 – 1.30	6 pts
> 1.30	Fail

---

Key Concepts to Master

Expect a deep dive into these topics during the review. Use notes or code comments if needed, but ensure you actually understand the logic:

• Data Prep: What is One-Hot Encoding and why do we use it for labels?

• Optimization: * Batch Size: What value did you pick? What are the pros/cons of very small vs. very large batches?

• Learning Rate: How does this affect convergence?

• Callbacks: Which ones did you trigger (e.g., EarlyStopping, ModelCheckpoint)? Explain their parameters.

• Architecture Details:

• Layers: Be ready to explain only the layers you actually used (e.g., Conv2D, BatchNormalization, MaxPooling2D, Dropout, Dense).

• Activation Functions: Which ones (ReLU, Softmax, etc.) and why?

• Weights: How were they initialized? How does the program decide which weights are the "final" best version?

• Metrics & Analysis:

• Loss vs. Accuracy: What is the fundamental difference between val_loss and accuracy?

• Confusion Matrix: * What do the X and Y axes represent?

• What does the diagonal tell you?

• What do the off-diagonal numbers signify regarding model errors?

---

Would you like me to help you draft a high-performance CNN architecture in Python that targets that < 0.63 loss threshold?