basic concepts and tools for image understanding using Deep Learning.
implement a linear image classifier while getting familiar with numpy
and the benefits of vectorized operations in Python.
This exercise has 2 parts:
- Implementing loss functions, calculating gradients and implementing gradient descent.
- Linear perceptron
- Training and evaluating several classifiers.
- Visualize my results.
implement a binary classifier with one linear layer on a small portion of CIFAR-10. In this exercise, you will implement a three layer multi-class neural network.
- Implement a loss function (Cross-entropy)
- Implement a L2 Regularization
- Implement a Fully Connected Layer: Forward Pass, Backward Pass.
- Implement a Fully ReLU: Forward Pass, Backward Pass.
- Building the Network.
- Train a convolutional network using PyTorch and evaluate it.
- Visualize a neural network layer and interpet their meaning.
- Train a Localization as Regression network using a pre-trained model.
- Build and train a recurrent neural network that will generate text.
The goal of image captioning is to describe a given image using natural language. Using neural networks, we can partition the problem into two separate challenges. First, we need to extract meaningful features regarding the image that would help us describe it. Second, we need to generate a sequence of words that best fit those features.