Dontopedia

Universal Approximator

From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-06.)

Universal Approximator has 23 facts recorded in Dontopedia across 2 references, with 3 live disagreements.

23 facts·18 predicates·2 sources·3 in dispute

Mostly:has formula(2), initializes as(2), compatible with backend(2)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (2)

Other subjects in dontopedia point AT this entity as a value. These are inverse relationships — e.g. "X motherOf this subject" — and answer questions the forward facts can't. Grouped by predicate.

hasMemberHas Member(1)

introducesNewActivationIntroduces New Activation(1)

Other facts (22)

The long tail: predicates that appear too rarely to warrant their own section. Filter or scroll to find a specific one. Each row links to its source.

22 facts
PredicateValueRef
Has Formulaf(x) = silu(x) * gate + x * skip[1]
Has Formulaf(x) = silu(x) * gate + x * skip[2]
Initializes Asidentity-like (gate=1, skip=0)[1]
Initializes Asidentity-like[2]
Compatible With BackendCpu[1]
Compatible With BackendHelios Gpu[1]
Has Parametergate[2]
Has Parameterskip[2]
Broadcasts With[B*T, ffnDim] hidden activations[1]
Has Per Channel Gatingnull[1]
Is Universal Approximatornull[1]
Learns to Combinenonlinear + linear paths per channel[1]
Requires No New Kernelsnull[1]
Starts Identity Likenull[1]
Uses Learnable Paramsgate and skip are trainable [1, ffnDim] vectors[1]
Uses Param Shape[1, ffnDim][1]
Combines Silu and Linearnull[1]
Enables Per Channel Learningnull[1]
Rdf:typeActivation Function[2]
Has Featurelearnable per-channel gating[2]
Combines Pathsnonlinear + linear[2]
Has Learnable Coefficientstrue[2]

Timeline

Timeline axis is valid_time — when each source says the fact was true in the world, not when Dontopedia learned about it. Retracted rows are kept for provenance; coloured stripes indicate the context kind.

23 undated
—
broadcastsWithblah/training-and-evals/part-21
[B*T, ffnDim] hidden activations
—
hasFormulablah/training-and-evals/part-21
f(x) = silu(x) * gate + x * skip
—
hasPerChannelGatingblah/training-and-evals/part-21
null
—
initializesAsblah/training-and-evals/part-21
identity-like (gate=1, skip=0)
—
isUniversalApproximatorblah/training-and-evals/part-21
null
—
learnsToCombineblah/training-and-evals/part-21
nonlinear + linear paths per channel
—
requiresNoNewKernelsblah/training-and-evals/part-21
null
—
startsIdentityLikeblah/training-and-evals/part-21
null
—
usesLearnableParamsblah/training-and-evals/part-21
gate and skip are trainable [1, ffnDim] vectors
—
usesParamShapeblah/training-and-evals/part-21
[1, ffnDim]
—
combinesSiluAndLinearblah/training-and-evals/part-21
null
—
compatibleWithBackendblah/training-and-evals/part-21
ex:cpu
—
compatibleWithBackendblah/training-and-evals/part-21
ex:helios-gpu
—
enablesPerChannelLearningblah/training-and-evals/part-21
null
—
typeblah/training-and-evals/21
ex:ActivationFunction
—
labelblah/training-and-evals/21
Universal Approximator
—
hasFeatureblah/training-and-evals/21
learnable per-channel gating
—
hasFormulablah/training-and-evals/21
f(x) = silu(x) * gate + x * skip
—
hasParameterblah/training-and-evals/21
gate
—
hasParameterblah/training-and-evals/21
skip
—
initializesAsblah/training-and-evals/21
identity-like
—
combinesPathsblah/training-and-evals/21
nonlinear + linear
—
hasLearnableCoefficientsblah/training-and-evals/21
true

References (2)

2 references
  1. [1]Part 2114 facts
    ctx:discord/blah/training-and-evals/part-21
  2. [2]219 facts
    ctx:discord/blah/training-and-evals/21
    • full texttraining-and-evals-21
      text/plain2 KBdoc:agent/training-and-evals-21/9cfc0243-2772-4a86-8d1d-cdb625ab29f4
      Show excerpt
      [2026-02-25 11:49] ajaxdavis: https://alpha.omegaai.dev/runs/historic_chat_v2_20260225114638_5ke3 [2026-02-25 12:12] ajaxdavis: https://docs.google.com/document/d/1DTgZf5HC4xD1xntzZUIPAM5B4BrX9YW-sP34ekMpuHE/edit?tab=t.0' [2026-02-25 12:37]…

See also

Keep researching

Missing something or suspicious of what's here? Kick off a research session — a Claude agent will investigate, cite its sources, and file new facts into a dedicated context you can review before accepting into the shared view.

↑