Dontopedia

KAN

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

KAN has 50 facts recorded in Dontopedia across 27 references, with 1 live disagreement.

50 facts·45 predicates·27 sources·1 in dispute

Mostly:full name(2), rdf:type(2), is standard attention with learned kernel(1)

Maturity scale raw canonical shape-checked rule-derived certified

Full NamefullName

  • Kolmogorov-Arnold Networks[11]all time · Part 172
  • Kolmogorov-Arnold Networks[26]sourceall time · 172

Inbound mentions (26)

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.

involvesInvolves(2)

areCompetitiveWithAre Competitive With(1)

canBeReplacedByCan Be Replaced by(1)

comparesAgainstCompares Against(1)

comparesModelsCompares Models(1)

competitiveQualityCompetitive Quality(1)

currentlyIncludesCurrently Includes(1)

doesNotMerelyApproximateDoes Not Merely Approximate(1)

evaluatedAgainstEvaluated Against(1)

includesIncludes(1)

includesModelIncludes Model(1)

includesVisualizationOfIncludes Visualization of(1)

isReorganizingIs Reorganizing(1)

isTrainableIs Trainable(1)

matchesPplWithMatches Ppl With(1)

narrowsPerformanceGapToNarrows Performance Gap to(1)

presupposesExistenceOfPresupposes Existence of(1)

presupposesRelationToPresupposes Relation to(1)

proposesReplacementProposes Replacement(1)

recommendsKeepingRecommends Keeping(1)

referencesKansReferences Kans(1)

regularizesRegularizes(1)

runsModelVariantRuns Model Variant(1)

suggestsAlternativeSuggests Alternative(1)

superiorEfficiencySuperior Efficiency(1)

Other facts (45)

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.

45 facts
PredicateValueRef
Rdf:typeArchitecture[22]
Rdf:typeConcept[27]
Is Standard Attention With Learned Kerneltrue[1]
Outperforms Others in SpeedAll Modes[1]
Has Best Ppltrue[1]
Has Fastest Inferencetrue[1]
Has Ppl Score91.4[1]
Has Relative Decode Speed1.0x[1]
Is Fastest Modetrue[1]
Supports Cached DecodeE081b1e[1]
Best in Dual MetricsPpl and Speed[1]
Extends AttentionLearned Kernel[1]
Provides Full Kv Cache Efficiencytrue[1]
Achieves GainsKan Like Gains[2]
Works WellFindings[2]
Champion inPairwise Quadratic[2]
Has RolePairwise Quadratic Champion[2]
Has Quadratic Quality BaselineNear Quadratic[3]
Has Ppl91.3[3]
Superior to Othersnull[4]
Provides Primary Signalnull[4]
Works Bestnull[4]
Benchmark ReferenceTrue[5]
Serves As Performance Baselinenull[6]
Is Beating All ComersCurrently[7]
Serves As Baselinetrue[8]
Achieves Best Ppl32.1[9]
Uses Lohe Synctrue[10]
References Kolmogorov Arnold Networksnull[12]
Needs to Learn Meanings ofSimilarities[12]
Tested WithKuramoto[12]
Requires MatchingAttention Mechanism[13]
Updates CouplingSpline[14]
Alternative Architecturetrue[15]
ShiftsByte Positions[16]
Normalizes Qkper-head via L2[17]
Replaces Fixed Nonlinearitiestrue[18]
Is InterestingXenonfun[18]
Shares Philosophical DnaOscillator Approach[18]
Philosophically Aligned WithResonance Architecture[18]
Referenced in ResearchLatest Kan Research[19]
Learns Modulation SchemeNot Dictionary[20]
Is Mauritian Creole Term forCamp[21]
Role ofPairwise Quadratic Champion[23]
Performance StatusWorks Well[23]

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.

isStandardAttentionWithLearnedKernelblah/watt-activation/part-68
true
outperformsOthersInSpeedblah/watt-activation/part-68
ex:all-modes
hasBestPplblah/watt-activation/part-68
true
hasFastestInferenceblah/watt-activation/part-68
true
hasPplScoreblah/watt-activation/part-68
91.4
hasRelativeDecodeSpeedblah/watt-activation/part-68
1.0x
isFastestModeblah/watt-activation/part-68
true
supportsCachedDecodeblah/watt-activation/part-68
ex:e081b1e
bestInDualMetricsblah/watt-activation/part-68
ex:ppl-and-speed
extendsAttentionblah/watt-activation/part-68
ex:learned-kernel
providesFullKvCacheEfficiencyblah/watt-activation/part-68
true
achievesGainsblah/watt-activation/part-70
ex:kan-like-gains
worksWellblah/watt-activation/part-70
ex:findings
championInblah/watt-activation/part-70
ex:pairwise-quadratic
hasRoleblah/watt-activation/part-70
ex:pairwise-quadratic-champion
hasQuadraticQualityBaselineblah/watt-activation/part-71
ex:near-quadratic
hasPPLblah/watt-activation/part-71
91.3
superiorToOthersblah/watt-activation/part-69
null
providesPrimarySignalblah/watt-activation/part-69
null
worksBestblah/watt-activation/part-69
null
benchmarkReferenceblah/watt-activation/part-74
ex:true
servesAsPerformanceBaselineblah/watt-activation/part-76
null
isBeatingAllComersblah/watt-activation/part-64
ex:currently
servesAsBaselineblah/watt-activation/part-75
true
achievesBestPPLblah/watt-activation/part-72
32.1
usesLoheSyncblah/watt-activation/part-118
true
fullNameblah/watt-activation/part-172
Kolmogorov-Arnold Networks
referencesKolmogorovArnoldNetworksblah/watt-activation/part-177
null
needsToLearnMeaningsOfblah/watt-activation/part-177
ex:similarities
testedWithblah/watt-activation/part-177
ex:kuramoto
requiresMatchingblah/watt-activation/part-195
ex:attention-mechanism
updatesCouplingblah/watt-activation/part-220
ex:spline
alternativeArchitectureblah/watt-activation/part-390
true
shiftsblah/watt-activation/part-418
ex:byte-positions
normalizesQKblah/watt-activation/part-647
per-head via L2
replacesFixedNonlinearitiesblah/watt-activation/part-174
true
isInterestingblah/watt-activation/part-174
ex:xenonfun
sharesPhilosophicalDnablah/watt-activation/part-174
ex:oscillator-approach
philosophicallyAlignedWithblah/watt-activation/part-174
ex:resonance-architecture
referencedInResearchblah/watt-activation/part-304
ex:latest-kan-research
learnsModulationSchemeblah/watt-activation/part-416
ex:not-dictionary
isMauritianCreoleTermForval-mauritius/wf7-16-truth-and-justice-commission-mauritius-final-report-vol
ex:camp
typeblah/watt-activation/63
ex:Architecture
labelblah/watt-activation/70
KAN
roleOfblah/watt-activation/70
ex:pairwise-quadratic-champion
performanceStatusblah/watt-activation/70
ex:works-well
labelblah/watt-activation/74
KAN
labelblah/watt-activation/118
KAN
fullNameblah/watt-activation/172
Kolmogorov-Arnold Networks
typeblah/watt-activation/490
ex:Concept

References (27)

27 references
  1. [1]Part 6811 facts
    ctx:discord/blah/watt-activation/part-68
  2. [2]Part 704 facts
    ctx:discord/blah/watt-activation/part-70
  3. [3]Part 712 facts
    ctx:discord/blah/watt-activation/part-71
  4. [4]Part 693 facts
    ctx:discord/blah/watt-activation/part-69
  5. [5]Part 741 fact
    ctx:discord/blah/watt-activation/part-74
  6. [6]Part 761 fact
    ctx:discord/blah/watt-activation/part-76
  7. [7]Part 641 fact
    ctx:discord/blah/watt-activation/part-64
  8. [8]Part 751 fact
    ctx:discord/blah/watt-activation/part-75
  9. [9]Part 721 fact
    ctx:discord/blah/watt-activation/part-72
  10. [10]Part 1181 fact
    ctx:discord/blah/watt-activation/part-118
  11. [11]Part 1721 fact
    ctx:discord/blah/watt-activation/part-172
  12. [12]Part 1773 facts
    ctx:discord/blah/watt-activation/part-177
  13. [13]Part 1951 fact
    ctx:discord/blah/watt-activation/part-195
  14. [14]Part 2201 fact
    ctx:discord/blah/watt-activation/part-220
  15. [15]Part 3901 fact
    ctx:discord/blah/watt-activation/part-390
  16. [16]Part 4181 fact
    ctx:discord/blah/watt-activation/part-418
  17. [17]Part 6471 fact
    ctx:discord/blah/watt-activation/part-647
  18. [18]Part 1744 facts
    ctx:discord/blah/watt-activation/part-174
  19. [19]Part 3041 fact
    ctx:discord/blah/watt-activation/part-304
  20. [20]Part 4161 fact
    ctx:discord/blah/watt-activation/part-416
  21. ctx:genes/val-mauritius/wf7-16-truth-and-justice-commission-mauritius-final-report-vol
  22. [22]631 fact
    ctx:discord/blah/watt-activation/63
    • full textwatt-activation-63
      text/plain2 KBdoc:agent/watt-activation-63/1bd53136-248b-4353-b53c-b8c81b9d26f4
      Show excerpt
      [2026-03-07 15:30] xenonfun: ### no GPU contention run ``` Clean results at 5K iters: ┌───────────────────┬──────┬───────┬──────────┬───────────┬──────────┬───────┬─────────┐ │ Config │ it/s │ tok/s │ Avg Loss │ Final PPL
  23. [23]703 facts
    ctx:discord/blah/watt-activation/70
    • full textwatt-activation-70
      text/plain2 KBdoc:agent/watt-activation-70/7e42cc6f-7fd7-4300-aa25-d8cf3a3c0db3
      Show excerpt
      [2026-03-07 17:06] xenonfun: 7. What I’d do next I’d narrow the agenda hard: keep KAN as the pairwise quadratic champion keep skip-dynamics as a “feature map without rollout” probe stop spending much time on vanilla Kuramoto at this siz
  24. [24]741 fact
    ctx:discord/blah/watt-activation/74
    • full textwatt-activation-74
      text/plain3 KBdoc:agent/watt-activation-74/f0e790ce-8e1d-4951-819d-93d9164a6692
      Show excerpt
      [2026-03-07 18:16] xenonfun: The loop formulation exactly matches the quadratic reference A_ij = Σ_a w_ia * w_ja. Now the user has sent a second message with an even better approach — the streaming prefix-state formulation using two eins
  25. [25]1181 fact
    ctx:discord/blah/watt-activation/118
    • full textwatt-activation-118
      text/plain3 KBdoc:agent/watt-activation-118/ed79098d-1144-44f5-9941-e6b2b9c1caa7
      Show excerpt
      [2026-03-08 23:43] xenonfun: Code Changes (3 important patterns) 1. Fused QKV projection in SpectralAttention - Separate q_proj, k_proj, v_proj → single qkv_proj = Linear(d_model, 3 * d_model). One matmul instead of three. We should po
  26. [26]1721 fact
    ctx:discord/blah/watt-activation/172
    • full textwatt-activation-172
      text/plain3 KBdoc:agent/watt-activation-172/00a8d0a2-d9fd-4586-8848-72d39c310385
      Show excerpt
      [2026-03-09 22:02] xenonfun: We run out of Claude at this rate on `Tuesday Mar 10, 4:13 PM` [2026-03-09 22:03] lisamegawatts: "This is a React/TypeScript application built with Vite that provides an interactive 3D visualization interface fo
  27. [27]4901 fact
    ctx:discord/blah/watt-activation/490
    • full textwatt-activation-490
      text/plain2 KBdoc:agent/watt-activation-490/0995e13f-78e2-4e6d-97c3-788b633f495b
      Show excerpt
      [2026-03-22 06:02] xenonfun: ✳ Creating GeometricCodec… (3m 52s · ↓ 3.0k tokens) ⎿  ◼ Phase 2: GeometricCodec — KAN byte↔sphere codec ◼ Phase 2: ThomsonHelmholtz dynamics block ◻ Phase 3: Backprop-free LoheOptimizer ◻ Phase

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.