You work as a Deep Learning engineer in a self$-$driving car company and you$$re given a labelled dataset of $100,000$ images of cars belonging to $1,000$ different categories. You must use the dataset to train a model that will achieve at least $90\%$ accuracy in recognizing the correct car model. Moreover, your model must be as lightweight as possible meaning that it should contain as few parameters as possible. Your advisor suggests that you can use the VGG$-$Net architecture where you can specify the number of blocks N used. As N increases you can get models that can potentially achieve higher accuracy but also contain more parameters. Describe your strategy $($i$.$e$.$ training process$)$ for finding the optimal N that meets the given requirements.