The By Hand problem set series is designed to develop an understanding of the mathematics behind machine learning, artificial intelligence, cybersecurity, and related computational methods. The core problems are intentionally small enough to work through with paper, pencil, and a basic calculator before implementing the same ideas with software.
Problem sets are included with Patreon membership. Additional worked solutions and programming implementations may also be available for purchase from Patreon.
Course and Institutional Access: By Hand problem sets may be made available to students through a school's learning management system (LMS), such as Moodle, Brightspace, Canvas, or Blackboard, when access has been arranged for the course or institution. Individual access is also available to members through Patreon.
Educators, departments, and institutions interested in using By Hand problem sets in their courses may inquire about course or institutional licensing options.
Information Theory by Hand is a cybersecurity-focused problem set for learning information theory through hand calculations. Problems explore entropy, conditional entropy, Kullback-Leibler (KL) divergence, mutual information, information gain, cosine similarity, and related concepts using cybersecurity examples.
The goal is to understand what these information-theoretic measures actually calculate and why they are useful in cybersecurity and machine learning before relying on Python or a software library.
Included with Patreon membership.
Access the Problem SetAnomaly Detection by Hand explores the mathematics behind anomaly detection using Hopfield neural networks and cybersecurity examples. Problems develop concepts including bipolar representations, weight matrices, network updates, energy, pattern recall, reconstruction, and the identification of anomalous patterns.
The exercises show how a neural network can learn normal patterns and use deviations from those patterns to help identify unusual or potentially malicious activity.
Coming soon.
GPTs by Hand explores the mathematics behind Generative Pre-trained Transformers through calculations small enough to perform by hand. Problems examine tokens, embeddings, linear transformations, attention, softmax, cross-entropy loss, and the basic operations used during GPT inference.
The objective is to make the internal mathematics of a transformer understandable before moving to larger implementations using NumPy or PyTorch.
Coming soon.
Additional problem sets exploring machine learning, artificial intelligence, cybersecurity, dynamical systems, and other mathematical and computational topics are currently being developed.