Blog banner

SaddlePoint

Academic and technical blog — statistics, machine learning, and adjacent topics.

Maths, Systems and Research in the Age of AI

research
mathematics
systems
ai
pedagogy

If AI can write code and solve Olympiad-level mathematics, what remains for us to solve—and what skills must we develop? An essay about framing research problems, seeing systems, and using mathematics to recognize shared structure.

Sep 6, 2026
30 min

Writing Books for Interdisciplinary Fields in the Age of AI

ai
pedagogy
publishing
reproducibility
quarto

Interdisciplinary knowledge is siloed, slow to publish, and quick to fragment into disconnected chapters. Software workflows and generative AI collapse the first two costs — leaving the author as editor-in-chief, whose real job is the spine that carries one problem through.

Jul 7, 2026
14 min

Towards a Testable Theory of Emergence

ai
machine-learning
transformers
emergence
theory

A proposal for testing emergence as heldout but reachable coherence under recursive retrieval.

Jun 30, 2026
33 min

Science Moves. Engineering Compounds.

ai
machine-learning
engineering
moats

Scientific advantage diffuses faster every year; what endures is engineering muscle memory — operational capability earned through execution, not read from papers.

Jun 13, 2026
5 min

PageRank and KNN: an equivalence

graphs
ml
notebook

A tutorial-style notebook showing how PageRank on a weighted similarity graph relates to KNN-style neighborhood scores.

Dec 15, 2020
5 min

ML App Interoperability with Deep Learning, Part 2

machine-learning
ml-apps
interoperability
deep-learning

Deep learning frameworks provide a practical default for model interchange: readable specifications, scalable execution, and broad commercial adoption.

Jan 10, 2018
4 min

ML App Interoperability with PMML, Part 1

machine-learning
ml-apps
interoperability

PMML is a serious model-exchange effort, but weak importer support, inconsistent model interfaces, and XML-heavy design limit its role as a default.

Jan 9, 2018
3 min

A Small Step Toward Data Science Automation

machine-learning
automation
data-science

Model selection can be treated as a decision problem: learn rules from data characteristics, then use those rules to guide downstream modeling.

Nov 8, 2017
3 min

DataFrame as the ML Data Interface

machine-learning
dataframe
data-engineering

Even when data originates in flexible NoSQL systems, ML workflows need structured responses. The DataFrame is a practical default container.

Oct 27, 2017
3 min

A USB Interface for ML Apps

machine-learning
ml-apps
architecture

A USB metaphor for machine learning apps: separate data, code, config, and authentication so components can connect cleanly across environments.

Oct 26, 2017
3 min

Axioms for Reliable ML Apps, Part 2

machine-learning
ml-apps
architecture

A concrete architecture for ML apps: versioned data, code, config, micro apps, DAG orchestration, and DataFrames as the default data interface.

Oct 24, 2017
3 min

Reliable ML Apps: Goals and Guarantees

machine-learning
ml-apps
architecture

Machine learning apps need explicit guarantees: repeatability, auditability, recoverability, scalability, testability, security, responsiveness, and representation.

Oct 22, 2017
3 min
No matching items