nucleation

LIT-1624: An Overview of Early Vision in InceptionV1

2 min read

Properties5
StatusActive — read 2026-10-27 (NOTE-1206) in full, from the Distill article; worth reading as the only place the Circuits thread gives a denominator: it sorts every one of the 1,056 units of InceptionV1's first five layers into 77 named groups, with each group's share of its layer (THEORY-1042). Read it knowing that the groups are, in the authors' words, "ad-hoc, human defined", that 391 of the 1,056 units (37%) sit in groups the authors' own captions call catch-all, broad and not well defined, low-confidence, unstudied or not understood, and that being sorted into a group is not the same as being understood in Zoom In's sense. The authors' wider claim, that neuron families are real and recur across models, is THEORY-1040.
Version2
Tagsrepresentation-learning, anthology-candidate
Cited as “paper” byNOTE-1206 — Olah et al. — An overview of early vision in InceptionV1
Cited as “source” byTHEORY-1040 — The units of the early layers of trained vision networks fall largely into neuron families, sets of variants of one feature across orientation, colour or scale, which reflect structure in the network rather than a convenience of the person sorting them, and the same families form across models, as Olah, Cammarata, Schubert, Goh, Petrov and Carter propose, THEORY-1042 — In the Overview of Early Vision's taxonomy of InceptionV1, every one of the 1,056 units of conv2d0 to mixed3b is assigned by the authors to exactly one of 77 groups, and 391 of them (37%), from 7% of conv2d2 to 48% of conv2d1 and 45% of mixed3b, are in groups whose own captions call them catch-all, broad and not well defined, low-confidence, unstudied or not understood; the mixed3b curve family is 10 of that layer's 480 units

Graph View

Olah, C., Cammarata, N., Schubert, L., Goh, G., Petrov, M. and Carter, S. (2020). An overview of early vision in InceptionV1. Distill, 1 April 2020. https://distill.pub/2020/circuits/early-vision/ (the page prints doi:10.23915/distill.00024.002, which did not resolve on 2026-10-27).

Read in full on 2026-10-27 (NOTE-1206). Two accounts were drawn from it: THEORY-1042 (Active) and THEORY-1040 (Proposed).

Why this is here

Registered and read at the owner’s request of 2026-10-27 (“file the next readings”), as one of the works the circuits readings named and did not register. THEORY-1013, Zoom In’s claim that units of a trained vision network are typically meaningful features, is Proposed because its evidence is chosen examples with no denominator; its promote_when asks for “the fraction of units that meet” a stated criterion over whole layers. Curve Detectors (LIT-1575, NOTE-1198) gave one family, ten of mixed3b’s 480 units. This article is the thread’s only census: it puts every unit of five layers somewhere. The record’s circuits line held the programme statement and two single-family articles, and nothing that said how much of a layer the families cover. Read against THEORY-1013, it supplies a denominator but not the criterion: being sorted into a group is not being understood, and the authors say most groups were given “a few minutes of cursory investigation”.

Backlinks