# Linear probe

Source: https://onco.cc/terms/linear-probe/  
OnCo record `linear-probe` (Term). Data CC BY-NC 4.0, attribute "Data from OnCo (onco.cc)"; commercial use needs a licence.

## TL;DR

A linear probe is a simple linear model (logistic or ridge regression) trained on a frozen model's embeddings to test how much useful information those embeddings hold.

## Summary

Logistic regression models the log-odds of an outcome as a linear combination of inputs (Wikipedia); trained on frozen embeddings it becomes a probe of the representation, since whatever it can predict must be linearly readable from the features. Pathology foundation-model papers report linear-probe accuracy across many tasks as their headline comparison, and frozen embedding, then PCA to a few hundred dimensions, then a linear head is the standard extraction recipe for transcriptome models too.

## Fields

- Kind: Term
- Last checked: 2026-09-24
- Also known as: linear probe; linear probing; linear evaluation; linear head on frozen features; frozen embedding to linear head
- Tags: cansim-terms

## Notes

- Listed in the CanSim terms map 1.0.0 (docs/onco/terms.json, generated 2026-09-24), CC BY 4.0, attribution: CanSim project, an open, public-data-first cancer foundation-model programme; CanSim page path /terms/linear-probe.

## Sources

- Wikipedia: https://en.wikipedia.org/wiki/Logistic_regression
- Wikipedia: https://en.wikipedia.org/wiki/Logistic_regression

## Connected records

- terms: [Cancer AI vocabulary (CanSim terms map)](https://onco.cc/terms/cancer-ai-vocabulary/), [Embedding (learned representation)](https://onco.cc/terms/embedding/), [Fine-tuning versus frozen features, and LoRA](https://onco.cc/terms/fine-tuning-vs-frozen/), [Logistic regression and nearest-centroid classifiers](https://onco.cc/terms/logistic-regression-term/), [Ridge regression and the multilayer perceptron](https://onco.cc/terms/ridge-regression/)

---
JSON: https://onco.cc/api/v1/entities/linear-probe.json