Sign inSign in
ConceptAI & ML

Linear probe

A technique where a linear projection is trained on top of a frozen representation space to evaluate the quality of learned representations.
as Sara Beery defines itMITThe wording closest to all 5 definitions
Watch it explained

Know first

Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.

Lecturers

5

Each in their own terms, from the easiest level.

  • Phillip Isola

    MIT · Undergraduate

    A linear operation trained on top of a frozen representation for a new prediction problem.
    • 31:37Representation Learning: Reconstruction-Based
  • Ilya Sutskever

    UC Berkeley · Graduate

    Evaluating learned representations by training a linear classifier on top of a frozen neural network layer.
    • 33:12An Observation on Generalization and Unsupervised Compression
  • William Gilpin

    UT Austin · Graduate

    A simple linear model trained on top of frozen representations to evaluate what features are encoded.
    • 41:51Model Scale Versus Domain Knowledge in Long-Term Forecasting of Chaos
  • Yasaman Bahri

    Harvard University · Research

    A method used to linearly decode variables from the internal representations of Large Language Models to check what concepts the model has learned.
    • 40:32On the Emergence of Representational Geometry from Natural Language Statistics

Connections

9

The lectures that define it or assume it, and the concepts next to it.

5 lecturers. Definitions are written by AI from each lecture’s transcript. A time with a play mark starts the lecture at the moment the lecturer says it.