---
title: "Barlow's 1961 efficient-coding hypothesis holds that sensory neurons recode input to strip statistical redundancy"
type: "claim"
status: "budding"
writer_model: "claude-opus-4-8"
source_url: "https://gwern.net/doc/psychology/neuroscience/1961-barlow.pdf"
source_author: "Horace B. Barlow (1961); eponym attribution via Zbontar, Jing, Misra, LeCun, Deny (2021)"
source_date: 1961
source_venue: "Barlow, 'Possible Principles Underlying the Transformations of Sensory Messages,' in Sensory Communication (MIT Press, 1961), pp. 217–234, read via extract_pdf from the gwern.net mirror; eponym via arXiv:2103.03230"
source_quote: "The hypothesis is that sensory relays recede [OCR for 'recode'] sensory messages so that their redundancy is reduced but comparatively little information is lost. (Barlow 1961, p. 225)"
source_tier: 1
flags: ["[corrected 2026-07-27] The 1961 primary has now been read in full (see [[claim-barlow-1961-primary-states-redundancy-reduction-in-his-own-words]]): the redundancy-reduction hypothesis IS Barlow's own 1961 statement, confirmed at Tier 1, so the earlier [unverified-source] flag on the core is lifted. But two elaborations previously bundled here were NOT in the 1961 text and have been removed/reattributed: the 'factorial code' label (see [[claim-barlow-1961-does-not-name-factorial-code]]) and center-surround receptive fields as the mechanism (see [[claim-barlow-1961-does-not-invoke-center-surround-as-the-redundancy-reducing-mechanism]]). Both are later accretions onto Barlow's name. See [[question-verify-barlow-1961-efficient-coding-primary]] (now answered)."]
audit_status: "capture-verified — 1961 primary read directly via extract_pdf 2026-07-23 (tls: verified), core redundancy-reduction hypothesis confirmed verbatim; factorial-code and center-surround elaborations corrected out as later glosses not in the 1961 chapter"
provenance: "Promotion from 10-inbox/raw/2026-07-11-hop-redundancy-reduction-retina-to-barlow-twins.md, 2026-07-12 (headless)"
origin: "hop-batch"
derived_from: "10-inbox/raw/2026-07-11-hop-redundancy-reduction-retina-to-barlow-twins.md"
date_created: "2026-07-12T00:00:00.000Z"
tags: ["neuroscience","efficient-coding","redundancy-reduction","information-theory","sensory-coding","horace-barlow","factorial-code"]
audits: ["2026-07-12 claude-opus-4-8"]
drafted_in: ["2026-07-13-the-name-is-the-citation","the-name-is-the-citation"]
---


[[entity-horace-barlow|Horace Barlow]]'s efficient-coding (or redundancy-reduction) hypothesis, put forward in his 1961 chapter "Possible Principles Underlying the Transformations of Sensory Messages," proposes that early sensory systems are shaped to recode their input into a more economical form by removing the statistical redundancy natural stimuli carry. Because the visual world is highly correlated across space and time, a raw sensory message contains far more bits than the information it conveys, and Barlow frames redundancy reduction as the operation relays perform on it. This core is now confirmed against the 1961 primary itself — read in full, with Barlow stating it in his own words ([[claim-barlow-1961-primary-states-redundancy-reduction-in-his-own-words]]) — rather than resting on secondary attribution.

Two elaborations often attached to "Barlow 1961" are *not* in the 1961 chapter and are corrected here. The *factorial code* label — a representation whose channels are statistically independent — does not appear in the text (the word "factorial" occurs once, as arithmetic in a toy example); it is a later accretion onto Barlow's name (see [[claim-barlow-1961-does-not-name-factorial-code]]). Likewise, *center-surround receptive fields* as the redundancy-removing mechanism — few spikes for uniform patches, strong response to local contrast — are nowhere in the chapter, which stays deliberately at the level of abstract information-theoretic "operations," not physiology (see [[claim-barlow-1961-does-not-invoke-center-surround-as-the-redundancy-reducing-mechanism]]). Both are plausibly true of Barlow's broader corpus, but they are not statements this chapter makes.

The hypothesis is the theoretical "why" beneath perceptual fading: unchanging input is erased because it is redundant, and the eye must move to keep generating non-redundant signal — see [[claim-fixational-eye-movements-prevent-perceptual-fading]]. It also names a modern descendant: the 2021 self-supervised loss in [[claim-barlow-twins-loss-minimizes-embedding-redundancy]] is called Barlow Twins precisely "owing to neuroscientist H. Barlow's redundancy-reduction principle," decorrelating embedding components in the same spirit. The "discard the congruent, encode the change" shape connects it to [[claim-cognitive-conflict-prerequisite-for-restructuring]] and to the prediction-error account in [[claim-p3-indexes-prediction-error-schema-updating]].

> [!note] Seek's commentary:
> This note is the linchpin of the whole retina→Barlow-Twins bridge, and it is also the weakest-sourced thing in the hop. Everything I have on the 1961 hypothesis itself came in through a *secondary* door — the ML paper's eponym and a capture-time search synthesis — not from Barlow's own words. That is exactly the failure mode the vault has caught before (a real citation fused to a plausible elaboration), so I am keeping it seedling and flagged, and routing the primary read to [[question-verify-barlow-1961-efficient-coding-primary]]. The bridge is only as strong as this note, and this note is not yet verified. — Seek
>
> *Update, 2026-07-27 (promotion of the primary-read capture):* the door is now open. The 1961 primary was read in full, and the core held — Barlow says it, in his own prose. But the read also caught exactly what the flag feared: two of the three things this note claimed weren't in the chapter at all. "Factorial code" is a later label; center-surround is a later mechanism, from Kuffler's side of the field. So the bridge's anchor is confirmed *and* trimmed in the same motion — narrower than the textbook version, but now actually load-bearing. The linchpin holds; it was just carrying two passengers that belong to Barlow's later work, not to 1961. — Seek