Courses
13 / 1301020304050607080910111213
AI, machine learning and deep learning
Separate the task, the model and the way it learns.
Language models: probability, training and generation
From next-token probabilities to a complete answer.
Features, parameters and hyperparameters
Identify what comes from the input and what training changes.
Gradient descent and backpropagation
Trace an error back to each adjustable coefficient.
Generalization, overfitting and model evaluation
Understand what a training fit says about new data.
Images as tensors: representation and numerical transformations
Read spatial axes, channels and numerical encodings.
Convolution and directional edge responses
Calculate a directional edge response, one coefficient at a time.
From fixed features to trainable convolutional networks
Make feature extraction part of the optimization problem.
Nonlinearity, depth and residual connections
Explore what nonlinearities and shortcut paths contribute.
Tensor shapes, parameter counts and computational cost
Calculate output shapes, parameter counts and computation.
Compression, reconstruction and channel bottlenecks
Distinguish a smaller code from a cheaper operation.
Expansion, nonlinear transformation and projection
Build a nonlinear function through a wider intermediate layer.
Token embeddings and contextual representations
Connect discrete tokens to vectors and contextual computation.