Mohammad Zbeeb
Notes on machine learning, large language models, and interpretability.
I am a computer and communication engineering student at the American University of Beirut, working on large language models, reasoning transfer, model merging, and Arabic NLP. This is where I write up what I am learning. Find me on GitHub or read a little about me.
-
Probability Concepts for Machine Learning Engineers
Academic tutorial notes on the probability concepts machine learning engineers use in practice: random variables, distributions, expectation, Bayes' rule, likelihood, cross-entropy, calibration, distribution shift, and decision theory.
-
Deep Learning Specialization Notes
Annotated PDFs for neural networks, improving deep neural networks, structuring machine learning projects, convolutional neural networks, and sequence models. Each file opens directly in the browser.
-
Do Language Models Share Unsafe Directions in Activation Space?
Testing whether a single safety axis emerges across models once their activations are aligned into a shared space. Per-model safety vectors are extracted, aligned, and combined into a dominant “universal” safety direction that separates safe from unsafe activations.
-
What we borrowed in ML from info theory?
A walk through entropy, cross-entropy, KL divergence, and mutual information — the information-theoretic ideas that quietly underpin modern machine learning, with cozy figures for each one.