Content deleted Content added
mNo edit summary Tags: Visual edit Mobile edit Mobile web edit |
Added an image |
||
Line 52:
=== Interpretability ===
[[File:Grokking modular addition.jpg|thumb|upright=1.2|[[Grokking (machine learning)|Grokking]] is an example of phenomenon studied in interpretability. It involves a model that initially memorizes all the answers ([[overfitting]]), but later adopts an algorithm that generalizes to unseen data.<ref>{{Cite web |last=Ananthaswamy |first=Anil |date=2024-04-12 |title=How Do Machines ‘Grok’ Data? |url=https://www.quantamagazine.org/how-do-machines-grok-data-20240412/ |access-date=2025-01-21 |website=Quanta Magazine |language=en}}</ref>]]
Scholars sometimes use the term "mechanistic interpretability" to refer to the process of [[Reverse engineering|reverse-engineering]] [[artificial neural networks]] to understand their internal decision-making mechanisms and components, similar to how one might analyze a complex machine or computer program.<ref>{{Cite web |last=Olah |first=Chris |date=June 27, 2022 |title=Mechanistic Interpretability, Variables, and the Importance of Interpretable Bases |url=https://www.transformer-circuits.pub/2022/mech-interp-essay |access-date=2024-07-10 |website=www.transformer-circuits.pub}}</ref>
| |||