Minimum Description Length, MDL
Updated 2026-08-05
INTRODUCTION
English translation pending.
CORE DEFINITION
Introduced by Jorma Rissanen, the principle states that among competing hypotheses, the best explanation of a dataset is the one whose total encoding length, the model description plus the description of the data given that model, is shortest. A complex model can fit every point but costs many bits to describe, while a simple model is cheap to state but needs a long error description. Choosing the shortest total compresses the data most effectively, and because complexity is penalized by construction, the criterion naturally guards against overfitting.
SCAFFOLDING EFFECT
Reduce cognitive load
- Count both terms: add model length to error length instead of looking at fit alone. - Prefer the shortest: pick the model with the smallest total, not the best raw accuracy. - Test your summary: if you cannot compress a phenomenon briefly, you do not yet understand it.
Anchor fast decisions
A model is a code that compresses a dataset, so the whole enterprise is measured in bits. Adding parameters shrinks the error term but grows the model term, and beyond some point the second cost exceeds the first saving, so total length rises. The minimum of that sum is the best trade, and because extra complexity must pay for itself in reduced error, the criterion rejects models that merely memorize noise.
MINIMUM ACTION
In progress 0/1Practice this model in one real situation:
account_treeGenealogyexpand_more
menu_bookReferencesexpand_more
Source support: Explicit
- zh.wikipedia.orghttps://zh.wikipedia.org/wiki/%E6%9C%80%E5%B0%8F%E6%8F%8F%E8%BF%B0%E9%95%B7%E5%BA%A6verified
PRIVATE NOTES · Only visible to you
SAVED Q&A
ENTRY Q&A · Private saving available
Ask with a clear boundary
thinkingmodels answers from published entry context only.
Your question is sent to thinkingmodels. The answer uses public entry context only.
RELATED MODELS