Canonical taxonomy updated: This is a historical research view. The current four-level classification separates AI categories, learning paradigms, methods, and model/system architectures; retired labels are traceable without being counted as distinct active types. Open current taxonomy (v1.2) · Matching CSV.

Research & Knowledge / Relational Awareness / Living Atlas

Archived study-combination matrix: 1,000 proposals

A working register of 40 model architectures, algorithms, methods and system patterns, combined into 1,000 proposed study configurations. The base family list and the scenario table are separately classified.

Correction · 14 November 2026: The former spreadsheet conflated model architectures, model families, algorithms, agent systems and workflow patterns. This audit identifies 40 base families or methods and 1,000 unimplemented proposed combinations (40 × 5 × 5). They are not 1,000 verified or technically independent model architectures. Context links do not establish validation.
1,000proposed study configurations
40classified base families / methods
8categories
5shared framework candidates

01 / Base families and methods

This is the distinct architecture-and-method index, rather than 1,000 invented architecture names. Each entry has a classification and 25 unvalidated study scenarios. Labels distinguish model designs from algorithms, workflow patterns and agent systems.

#Family or methodClassificationCategoryResearch stage
01CNN feature hierarchymodel architectureComputer vision25 proposed combinations · unvalidated
02Vision Transformermodel architectureComputer vision25 proposed combinations · unvalidated
03Object detection networkmodel architecture classComputer vision25 proposed combinations · unvalidated
04Semantic segmentation networkmodel architecture classComputer vision25 proposed combinations · unvalidated
05Instance segmentation networkmodel architecture classComputer vision25 proposed combinations · unvalidated
06Encoder Transformermodel architectureLanguage and knowledge25 proposed combinations · unvalidated
07Autoregressive Transformermodel architectureLanguage and knowledge25 proposed combinations · unvalidated
08Retrieval-augmented generationsystem integration patternLanguage and knowledge25 proposed combinations · unvalidated
09Sequence-to-sequence Transformermodel architectureLanguage and knowledge25 proposed combinations · unvalidated
10Knowledge-graph question answeringapplication architectureLanguage and knowledge25 proposed combinations · unvalidated
11Gradient-boosted treesmachine-learning algorithm familyPredictive and tabular25 proposed combinations · unvalidated
12Generalised additive modelstatistical model familyPredictive and tabular25 proposed combinations · unvalidated
13Temporal forecasting Transformermodel architecturePredictive and tabular25 proposed combinations · unvalidated
14State-space sequence modelmodel architecturePredictive and tabular25 proposed combinations · unvalidated
15Survival and hazard modelstatistical model familyPredictive and tabular25 proposed combinations · unvalidated
16Graph convolutional networkmodel architectureGraphs and relational data25 proposed combinations · unvalidated
17Graph attention networkmodel architectureGraphs and relational data25 proposed combinations · unvalidated
18Relational graph neural networkmodel architectureGraphs and relational data25 proposed combinations · unvalidated
19Temporal graph neural networkmodel architectureGraphs and relational data25 proposed combinations · unvalidated
20Neural relational database modelresearch model/system conceptGraphs and relational data25 proposed combinations · unvalidated
21Denoising diffusion modelgenerative model architectureGenerative and multimodal25 proposed combinations · unvalidated
22Latent diffusion modelgenerative model architectureGenerative and multimodal25 proposed combinations · unvalidated
23Generative adversarial networkgenerative model architectureGenerative and multimodal25 proposed combinations · unvalidated
24Variational autoencodergenerative model architectureGenerative and multimodal25 proposed combinations · unvalidated
25Vision-language modelmultimodal model architecture classGenerative and multimodal25 proposed combinations · unvalidated
26Tool-using LLM agentagent system architectureAgents and automation25 proposed combinations · unvalidated
27Plan-and-execute agentagent orchestration patternAgents and automation25 proposed combinations · unvalidated
28Multi-agent orchestrationmulti-agent system patternAgents and automation25 proposed combinations · unvalidated
29AI-enabled BPM workflowbusiness workflow architectureAgents and automation25 proposed combinations · unvalidated
30Event-driven decision agentagent system architectureAgents and automation25 proposed combinations · unvalidated
31Vision-based robotic policyrobotics policy architectureRobotics and control25 proposed combinations · unvalidated
32Model-predictive controllercontrol methodRobotics and control25 proposed combinations · unvalidated
33Reinforcement learning policylearning-policy methodRobotics and control25 proposed combinations · unvalidated
34Imitation learning policylearning-policy methodRobotics and control25 proposed combinations · unvalidated
35Spiking neural controllerneural control architectureRobotics and control25 proposed combinations · unvalidated
36Rule-based expert systemsymbolic AI systemLogic, search and hybrid decision25 proposed combinations · unvalidated
37Constraint-satisfaction solverconstraint-solving algorithmLogic, search and hybrid decision25 proposed combinations · unvalidated
38Bayesian networkprobabilistic graphical modelLogic, search and hybrid decision25 proposed combinations · unvalidated
39Neuro-symbolic reasoning systemhybrid reasoning architectureLogic, search and hybrid decision25 proposed combinations · unvalidated
40Case-based reasoning systemreasoning methodLogic, search and hybrid decision25 proposed combinations · unvalidated

02 / Proposed research configurations

Browse the audited source directly on this page. Search and filter all 1,000 proposals; 25 rows are shown per page to keep mobile and desktop scrolling responsive. Rows are research scenarios, not independently demonstrated model types.

Loading audited catalogue…
Page 1
Audited proposed research configurations. Use the page controls to browse and swipe horizontally for additional columns.
IDBase family or methodFamily classificationProposed configurationCategoryRelationship studiedEvidence conditionReasoning strategySpecialist componentsShared framework IDsResearch status
Loading searchable 1,000-entry catalogue… If scripting is unavailable, use the audited CSV export above.

Family context links are informational; they do not validate the proposed variants. The original v0.1 CSV remains archived for provenance.

02 / Common relational reasoning frameworks

A shared framework should provide interoperability where meaningful, not impose one representation or model on every task. It must not be conflated with the solution being evaluated. Candidates are listed by their likely scope and evidence limitations.

Native observationsArchitecture-specific extractionTyped relation contractContext and uncertainty analysisIndependent QA

FR-01 · Typed relation and context graph

Across most families: subject, relation, object, polarity, domain, context and conditions. A graph records claims but does not prove them.

FR-02 · Temporal event ledger

Especially forecasting, robotics and automation: event order, valid times, versions, state changes and corrections. Sequence does not imply causality.

FR-03 · Statistical and causal-hypothesis layer

Especially predictive analytics and intervention research: variables, populations, confounders, uncertainty and identifying assumptions. Causal claims require stronger evidence.

FR-04 · Constraints, permissions and obligations

Especially intelligent agents and workflow systems: permission checks, exceptions, actor roles, rule precedence and jurisdictional applicability.

FR-05 · Provenance, contradictions and revision

Across families: evidence identifiers, derivations, competing findings, responsible actors, model and dataset versions, recorded corrections.

Adoption criterion: A common framework must preserve domain-specific meaning and demonstrate useful improvements against independently specified tests. Over-generalisation is an evaluated failure, not a benefit.

03 / Specialised components for each architectural application

Each of the 40 base families has its own linked Wikipedia context, native relationship type and specialised component set. These are implementation candidates rather than verified deployments.

CategoryWikipedia-linked familyNative relationshipSpecialised components
Computer visionCNN feature hierarchylocal spatial receptive fieldsconvolution backbone; spatial feature pyramid; image calibration
Computer visionVision Transformerpatch-token global attentionpatch encoder; attention maps; positional geometry
Computer visionObject detection networkobject localisation and identitydetector head; box matching; non-max suppression
Computer visionSemantic segmentation networkdense pixel-level category assignmentmask decoder; pixel alignment; boundary checks
Computer visionInstance segmentation networkseparate masks for object instancesinstance tracking; mask matching; overlap verification
Language and knowledgeEncoder Transformercontextual bidirectional language representationtokeniser; encoder; entailment head
Language and knowledgeAutoregressive Transformernext-token generation with long-context dependenciesdecoder; context window management; factuality evaluator
Language and knowledgeRetrieval-augmented generationretrieval-to-answer grounding relationshipsretriever; reranker; citation entailment checker
Language and knowledgeSequence-to-sequence Transformerinput-output sequence transformationsencoder-decoder; alignment tracer; translation evaluator
Language and knowledgeKnowledge-graph question answeringentity-link and graph-path evidenceentity linker; SPARQL query builder; path verifier
Predictive and tabularGradient-boosted treesnonlinear feature-outcome interactionstabular preprocessing; tree ensemble; feature attribution
Predictive and tabularGeneralised additive modeladditive, interpretable predictorsspline feature functions; calibration plots; residual checks
Predictive and tabularTemporal forecasting Transformertime-varying series dependencieswindow sampler; horizon evaluator; temporal attention
Predictive and tabularState-space sequence modellatent evolving state relationshipsstate update kernel; filter; state diagnostics
Predictive and tabularSurvival and hazard modelevent-time and censoring relationshipscensoring handler; hazard estimation; calibration
Graphs and relational dataGraph convolutional networkneighbourhood topology aggregationgraph encoder; adjacency manager; graph split evaluator
Graphs and relational dataGraph attention networkattention-weighted neighbourhood relationshipsattention layers; edge sampler; neighbourhood audit
Graphs and relational dataRelational graph neural networktyped edges and heterogeneous relationsrelation-aware message passing; edge type validator
Graphs and relational dataTemporal graph neural networktime-stamped dynamic graph changesevent buffer; temporal neighbour sampler; leakage checks
Graphs and relational dataNeural relational database modelforeign-key and multi-table entity relationsschema parser; table join graph; relational encoder
Generative and multimodalDenoising diffusion modelnoise-to-sample denoising trajectoriesnoise scheduler; denoiser; output consistency checks
Generative and multimodalLatent diffusion modelcompressed latent-to-output relationshipslatent encoder; denoiser; decoder; reconstruction checks
Generative and multimodalGenerative adversarial networkgenerator-discriminator training dynamicsgenerator; discriminator; mode-collapse evaluator
Generative and multimodalVariational autoencoderlatent distribution and reconstruction relationsprobabilistic encoder; latent sampler; decoder
Generative and multimodalVision-language modelcross-modal image-text correspondencesvisual encoder; text encoder; alignment verifier
Agents and automationTool-using LLM agentmodel-to-tool requests and permissionstool registry; approval gate; audit trail
Agents and automationPlan-and-execute agentgoal decomposition and action dependenciesplanner; executor; plan validity checker
Agents and automationMulti-agent orchestrationdelegation and message provenanceagent router; role policy; inter-agent trace
Agents and automationAI-enabled BPM workflowtasks, transitions and business rulesBPMN engine; rule adapter; process conformance checker
Agents and automationEvent-driven decision agentevent conditions and triggered actionsevent bus; decision table; idempotency and rollback
Robotics and controlVision-based robotic policyvisual input to motor actioncamera perception; policy controller; safety envelope
Robotics and controlModel-predictive controllerfuture-state planning under constraintsdynamics model; trajectory optimiser; constraint monitor
Robotics and controlReinforcement learning policystate-action-reward dynamicspolicy/value networks; simulator; reward audit
Robotics and controlImitation learning policydemonstration-to-action generalisationtrajectory dataset; imitation learner; shift monitor
Robotics and controlSpiking neural controllerevent-driven neural timing dynamicsspiking units; temporal encoder; event simulator
Logic, search and hybrid decisionRule-based expert systemif-then inferences and exception rulesrule engine; conflict resolver; justification trace
Logic, search and hybrid decisionConstraint-satisfaction solvervariables, constraints and feasible assignmentsconstraint model; solver; proof-of-infeasibility reporter
Logic, search and hybrid decisionBayesian networkconditional probabilistic dependenceDAG model; inference engine; posterior calibration
Logic, search and hybrid decisionNeuro-symbolic reasoning systemlearned representations with explicit symbolic rulesneural encoder; logic layer; contradiction detector
Logic, search and hybrid decisionCase-based reasoning systemanalogy and difference across past casescase memory; similarity measure; adaptation validator

04 / Further columns for a research-grade register

The first table prioritises legibility. A second-stage expansion should add: input modality; training objective; supervision mode; parameterisation; inference mechanism; memory/state representation; output modality; deployment context; edge/cloud constraints; data provenance; intended use; limitations; model/dataset versions; evaluation dataset; result metrics; uncertainty; safety controls; legal applicability; system owner; evidence links; reviewer and reassessment trigger. Do not fill fields with guesses or treat missing research as completed.

Keep numerical performance and compliance claims separate from architecture categories. A research candidate without a prototype must have an explicit not tested status.

05 / Research basis and limitations

Wikipedia links provide reader context only; research validity and architecture naming require independent technical literature. Proposed hybrids are not automatically distinct established subfields, nor is business process automation itself necessarily a standalone neural-network architecture.