Cyber Analytics Machine Learning Studio

A separate doctoral learning studio

Machine Learning, from estimand to governed action.

This section follows the research path: define what must be estimated, construct cutoff-valid evidence, evaluate probabilities, compare modern models, replay decisions, and refuse unsafe automation.

Start the textbook Download the lecture deck
13
chapters
144
lecture frames
22
Mermaid diagrams
GO
classroom deck audit
Signal Quest map connecting evaluation metrics with temporal validation
Every score inherits the label, population, clock, and decision policy that produced it.

Anchor concept

What is an estimand?

An estimand is the exact quantity a study seeks to estimate. In Signal Quest, it is the conditional probability of a declared settlement outcome, for an eligible decision time, using only observations permitted at that cutoff.

P(Yt = 1 | Xt = x)

The hypothesis asks whether a relationship exists. The estimand names the quantity. The model is only the instrument used to estimate it.

Learning sequence

Four movements, one scientific contract

The curriculum moves from definitions to deployment without allowing a metric or model to outrun its evidence.

01

Measurement

Define the label, cutoff, eligible population, clocks, and estimand before choosing a model.

02

Inference

Read confusion matrices, threshold curves, proper scores, calibration, and uncertainty together.

03

Representation

Compare boosted trees, self-supervised encoders, and causal transformers under the same evidence contract.

04

Governed action

Replay costs and latency, define abstention, and keep prediction separate from execution authority.

Teaching package

Read, teach, draw, and run

Each artifact has a distinct role. The textbook explains, the deck stages the argument, and the notebook runs labeled teaching experiments.

Interactive textbook

Signal Quest: Build an Honest Machine

Thirteen chapters with technical figures, worked examples, and Vantablack or white reading modes.

Read online Printable textbook

Signal Quest PDF

The 131-page textbook by Dr. Mallarapu, prepared for classroom reading and annotation.

Open PDF Lecture deck

Doctoral ML Masterclass

A 144-frame PowerPoint with 103 concepts, staged calculations, plots, images, and progressive reveals.

Download PowerPoint Slide deck · story edition

Masterclass Deck, Told as One Journey

A 156-slide web viewer for the story-arc rewrite: reward, problem, naive statistics, ML, metrics, model comparison, implementation, resolution, and the closing defense.

Open slide viewer Lecture deck · story edition

Masterclass Deck, Story Edition (PowerPoint)

The same 156-slide deck as a downloadable PowerPoint file, one slide per frame.

Download PowerPoint Live diagramming

Rendered Mermaid Library

Twenty-two rendered diagrams with copyable Mermaid source for explaining data contracts, metrics, models, replay, and safety gates.

Open diagrams and source Mermaid source

Mermaid Excalidraw Source

The canonical editable source file for copying diagrams into Excalidraw or lecture notes.

Open source Executed notebook

Technical Masterclass — HTML

A fully executed, browser-readable notebook with code, outputs, plots, and validation cells.

Open executed notebook Executable lab

LLM-Guided Technical Notebook

A Jupyter notebook that generates code and runs labeled synthetic teaching checks, experiments, plots, and validation.

Download notebook Executed capstone

LLM-Guided Capstone — HTML

A fully executed, browser-readable capstone with generated code, teaching outputs, and experiment results.

Open executed capstone Code-generation capstone

LLM-Guided Capstone Notebook

The original on-demand code-generation lab, preserved at its established URL for students and prior course links.

Download capstone