flâneur — a map of the web's best reading

Circuits in Superposition 2: Now with Less Wrong Math — LessWrong

lesswrong.com · 115,882 words · saved by 1 readers

This post is a continuation and clarification of Circuits in Superposition: Compressing many small neural networks into one. That post presented a sketch of a general mathematical framework for compressing different circuits into a network in superposition. On closer inspection, some of it turned out to be wrong, though. The error propagation calculations for networks with multiple layers were incorrect. With the framework used in that post, the errors blow up too much over multiple layers. This post presents a slightly changed construction that fixes those problems, and improves on the original in some other ways as well.[1] By computation in superposition we mean that a network represents features in superposition and performs more computations with them than it has neurons, across multiple layers. Having better models of this is important for understanding how and even if networks use superposition, which in turn is important for mech-interp in general. Performing computation in sup

x Circuits in Superposition 2: Now with Less Wrong Math — LessWrong window.__lwSsrGql.inject("query PostsPageWrapper($documentId: String, $sequenceId: String) {\n post(input: {selector: {documentId: $documentId}}, allowNull: true) {\n result {\n ...PostsWithNavigation\n }\n }\n}\n\nfragment PostsMinimumInfo on Post {\n _id\n slug\n title\n draft\n shortform\n hideCommentKarma\n af\n userId\n coauthorUserIds\n rejected\n collabEditorDialogue\n}\n\nfragment PostsBase on Post {\n ...PostsMinimumInfo\n url\n postedAt\n sticky\n metaSticky\n stickyPriority\n status\n frontpageDate\n meta\n deletedD

Explore this link on the map →

related reading