This image was uploaded in the JPEG format even though it consists of non-photographic data. This information could be stored more efficiently or accurately in the PNG or SVG format. If possible, please upload a PNG or SVG version of this image without compression artifacts, derived from a non-JPEG source (or with existing artifacts removed). After doing so, please tag the JPEG version with {{Superseded|NewImage.ext}} and remove this tag. This tag should not be applied to photographs or scans. If this image is a diagram or other image suitable for vectorisation, please tag this image with {{Convert to SVG}} instead of {{BadJPEG}}. If not suitable for vectorisation, use {{Convert to PNG}}. For more information, see {{BadJPEG}}.
Summary
DescriptionGrokking modular addition.jpg
English: Training and test loss curves from multiple runs showing the sudden emergence of generalization ("grokking") in a neural network learning modular addition.
Created in the context of a paper replication walkthrough of "Progress measures for grokking via mechanistic interpretability".
This work is free and may be used by anyone for any purpose. If you wish to use this content, you do not need to request permission as long as you follow any licensing requirements mentioned on this page.
The Wikimedia Foundation has received an e-mail confirming that the copyright holder has approved publication under the terms mentioned on this page. This correspondence has been reviewed by a Volunteer Response Team (VRT) member and stored in our permission archive. The correspondence is available to trusted volunteers as ticket #2025012010000298.
to share – to copy, distribute and transmit the work
to remix – to adapt the work
Under the following conditions:
attribution – You must give appropriate credit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use.
share alike – If you remix, transform, or build upon the material, you must distribute your contributions under the same or compatible license as the original.
Training and test loss curves from multiple runs showing the sudden emergence of generalization ("grokking") in a neural network learning modular addition
inception<\/a>"}},"text\/plain":{"en":{"":"inception"}}},"{\"value\":{\"time\":\"+2023-03-12T00:00:00Z\",\"timezone\":0,\"before\":0,\"after\":0,\"precision\":11,\"calendarmodel\":\"http:\\\/\\\/www.wikidata.org\\\/entity\\\/Q1985727\"},\"type\":\"time\"}":{"text\/html":{"en":{"P571":"12 March 2023"}},"text\/plain":{"en":{"P571":"12 March 2023"}}}}" class="wbmi-entityview-statementsGroup wbmi-entityview-statementsGroup-P571 oo-ui-layout oo-ui-panelLayout oo-ui-panelLayout-framed">
Uploaded a work by Neel Nanda from https://www.neelnanda.io/mechanistic-interpretability (also the thumbnail of this video: https://www.youtube.com/watch?v=ob4vuiqG2Go) with UploadWizard
