What is now the number of parameters of the model?

How do you interpret the results that you get from the two architectures? Is it worth adding another layer? Look at the early stopping behavior in the Best Model Info card - what does that tell us about our more complex model?

Add another hidden layer to the architecture - after the hidden layer with 64 output features, add another one with 32 output features. Retrain your model (keep the same training settings including early stopping).

What is now your test set accuracy?

Experiment with the architecture of the model to get a better result. Re-train the model with the new architecture (keep the same settings). Check how many parameters your model has now. What are most of the parameters in the model?

How many parameters does the model have that you have just trained with preset 1xLinear?

Copy & Paste the architecture that you used here. Explain why you defined it in that way. Include the decision boundary plot of that model and comment on it.

Select the 2D Blobs dataset with the following settings (seed=2025):

• Samples: 500
• Noise: 0.3
• Threshold: 1.3
• Offset: 0.25, 0.25

Choose the Toy Classification (Binary) 1xLinear preset and train the model for 50 epochs (leave learning rate and momentum at 0.01 and 0.9 respectively, set training seed to 2025, do not use early stopping). Attach the plots for the decision boundary and the loss curve here.

How do you interpret the decision boundary? How does this model perform?

Explain the parameters of the model - what kind of values are they (check out the card 'Parameter Values')?

Attach the plot of the fitted model using GD here. Interpret the model that was fitted - why does the curve look like the way it does?