Machine Learning.
Generalisation, leakage and the difference between a model and a description.
Articles
0
Quick Takes
0
Videos
0Soon
Podcasts
0Soon
Clarity Index
—Soon
Followers
—Soon
A model that fits the past perfectly has usually memorised it. Machine learning coverage lives or dies on one distinction: performance on data the model has seen, versus data it has not.
What this field actually measures
- Train/test split
Holding out data the model never trains on.
ReferenceThe minimum standard; cross-validation is stronger.
- Overfitting gap
Difference between training and validation performance.
ReferenceA large gap means the model learned the sample, not the phenomenon.
- Feature leakage
Information in the inputs that would not exist at prediction time.
ReferenceThe most common cause of spectacular results that fail in production.
Evidence & metrics
The published work behind the numbers above, and the public datasets you can open to check us. If a claim can’t be traced here, we don’t print it.
Citations
Breiman — Machine Learning · 2001
Established ensemble averaging as a variance-reduction method and formalised out-of-bag error estimation.
Hidden technical debt in machine learning systems
Sculley et al. — NeurIPS · 2015
Model code is a small fraction of a deployed ML system; most failure comes from data and pipeline coupling.
Datasets
UCI Machine Learning Repository
University of California, Irvine · free
Hundreds of benchmark datasets with documented provenance.
OpenML community · free
Datasets plus reproducible run results for thousands of tasks.
Still open
- How much sports-modelling accuracy is leakage rather than insight?
- Which model classes are worth their interpretability cost in decision settings?
- What should a public accuracy claim be required to disclose?
What we won’t say
We will not quote an accuracy figure without knowing the base rate it beat.
- VO-1Evidence before wit
- VO-2Punch at claims, never at people
- VO-3Say the uncertainty out loud
Everything published on this topic.
No articles have been published for Machine Learning yet. New long-form pieces will land here.
Everything published on Machine Learning.
The permanent record of every piece we publish under this topic — grouped by year, oldest kept as history.
Short reactions, polls, and open questions.
Faster than the long form. Same standard.
No quick takes match this filter yet.
Evidence, counterpoints, patterns, and the standard.
Every entry links back to its source article. Nothing duplicated.
Framework entries appear here as articles publish.
Open questions and measurable predictions.
We log the question, the confidence, and the review date. When the evidence moves, so does our position.
No active predictions are currently being tracked for this topic.
Coverage on screen.
Long-form, shorts, and playlists connected back to the source articles.
Video coverage is coming soon.
Coming SoonWe haven't published video on Machine Learning yet. Playlists, chapters, and shorts will surface here as they roll out.
Long-form conversations.
No podcast episodes yet.
Coming SoonWhen we publish podcast episodes on Machine Learning, each will link back to its source article.
Connected hubs.
Where this topic overlaps with the rest of the 2CS knowledge graph.
Explore Machine Learning
Every surface connected to this topic.
One permanent home per topic — the whole 2CS library on Machine Learning, indexed for readers and search engines alike.
- All Machine Learning articlesLong-form analysis and film breakdowns tagged Machine Learning.
- Machine Learning quick takesShort reactions, polls, and open questions on the record.
- Machine Learning in the 2CS Analysis FrameworkEvidence, counterpoints, patterns, and the standard.
- Machine Learning Future WatchPredictions we're tracking with confidence and review dates.
- Video coverageLong-form, shorts, and playlists — filter for Machine Learning coverage.
- Podcast episodesLong-form conversations that reference Machine Learning.
- Community investigationsReader-led inquiries connected to Machine Learning.
- Claims & evidence lockerEvery claim we've logged, with the evidence behind it.
- Machine Learning in the Knowledge GraphHow this topic connects to people, teams, and events.
Related topics
Where Machine Learning overlaps.
Full index
Browse every topic.
Every permanent home in one place — grouped by league, discipline, and idea.
All topicsThe conversation, on the record.
Community discussion is coming soon.
Coming SoonA dedicated space for readers to discuss this topic will live here — moderated to the same standard as the editorial.
Follow Machine Learning.
One weekly dispatch. Signal over noise. No spam.
No spam. Unsubscribe anytime.