Choose the route closest to your main contribution, then adapt it. Architecture, objective, theory, analysis, resource, system, application and position arguments carry different evidence obligations. Only types present in this inspected corpus appear below. These editorial routes do not replace the conference's official track definitions.
For a new design, prepare a readable computation or representation overview, a comparison with matched conditions, and controls that test the changed component. Add pseudocode when iteration, state, sampling or gradient routing would otherwise be ambiguous. Add a theorem only when making a formal claim. A position paper instead needs a defensible thesis, substantive alternatives and implications; it is not required to pretend to be a leaderboard method.
Observed medians describe source versions. Suggested display packages are conditional editorial starting points, not official requirements or evidence of a necessary number of figures. Small groups provide examples, not robust estimates of a conference-wide standard.
The central deliverable changes the model's representation, component arrangement, information flow or reusable inference pipeline. Data, objectives and efficiency can still be substantial supporting contributions.
Corpus context · 3 papers: median 6 main figures and 6 main tables; body span ≈ 7.07 pages; attached appendix span ≈ Not resolved pages (0 resolved boundaries). Separate source files: median Not resolved whole-file pages among 0 papers; these can repeat main text and are not appendix-only lengths. Extraction-assisted observations—not drafting targets.
Three contrasting architecture studies organize their diagrams around the actual changed information flow. FoundationSLAM returns reliability to a recurrent geometric system; URaG removes unselected page tokens between early and deep LLM layers; VAEVQ distinguishes stochastic and quantized paths through a shared decoder. Their experimental tables reuse module names, separate outputs or evaluation regimes, and retain controls that the full system does not always beat. This is a synthesis of these three inspected examples, not the usual architecture style of all AAAI papers.
Editorial starting package: choose artifacts that test your actual claims. Quantities below are planning suggestions, not counts inferred from all category members or conference rules.
Start with one overview; add insets only when they resolve different questions
Label inputs, outputs, inherited versus new components, frozen/trainable blocks and training-only paths.
When: Core when the novelty is computational organization; use a simple representation cartoon before the full diagram if the data object is unfamiliar.
One compact example or aligned qualitative grid
Make the claimed representational problem visible and reuse the same example through the mechanism.
When: Especially useful for 3D geometry, multimodal tokens, articulated objects or structured outputs; do not add a gallery that does not test the claim.
One focused table per incompatible task or information regime
Keep metric direction, data, backbone, reference inputs and refinement policy next to the results.
When: Separate single-model inference from per-instance optimization and privileged-input baselines.
A compact switch table or matched diagnostic panels
Test the named reason for each component rather than removing modules without preserving compute or capacity.
When: Use matched controls when the mechanism claim depends on architecture rather than more parameters or more supervision.
Only when the procedure or guarantee requires it
Expose iteration, state, branching or assumptions not recoverable from the diagram.
When: A feed-forward module does not require pseudocode or a theorem simply to look like an accepted paper.
Keep the defining computation, reused versus trained modules, input privileges and consequential state changes in the main text. Put full layer/configuration details, preprocessing, scene or subject splits and task-level controls in precisely referenced supporting material. Do not call a code link an inspected implementation.
A feedback architecture needs its return path and update condition, not only a left-to-right component chain.
Make the pruning boundary and training-only information visible; FLOPs on duplicated long inputs are not measured latency.
Keep the representation path prominent, place regularizers in insets and distinguish reconstruction from generation.
Boundary: This route contains 3 primary editorial assignments in the 16-paper inspected subset (1058 inventory records). It is not an official track, a universal writing standard or evidence that these choices caused acceptance.
See inline examples of these artifact styles · Compare every member in the measurement ledger
The central change is the coupled computation graph connecting frozen geometric priors, trainable task adaptation, bi-consistent optimization and reliability feedback. The residuals and ablations explain and test that architectural coupling.
The contribution changes the token pathway inside a multimodal LLM by adding contextualized page retrieval and an explicit pruning boundary between early and deep layers. Staged training and resource measurements support this changed computation graph.
The defining intervention replaces the deterministic tokenizer pathway with variational sampling, quantization and dual-path shared decoding. Local/global objectives are important supporting components, but the primary explanatory artifact is the changed representation architecture.
The argument primarily changes a loss, supervision signal, optimization/update rule, distillation recipe or adaptation procedure. Some assigned methods update at inference time; the label refers to the objective-driven argument, not only offline training.
Corpus context · 4 papers: median 4 main figures and 3 main tables; body span ≈ 7.12 pages; attached appendix span ≈ Not resolved pages (0 resolved boundaries). Separate source files: median Not resolved whole-file pages among 0 papers; these can repeat main text and are not appendix-only lengths. Extraction-assisted observations—not drafting targets.
The four inspected examples first define the object being optimized or measured, then connect the changed supervision or update to it. MBCD separates modality regulation from EMA guidance; ProGMLP distinguishes staged training from runtime stopping; neural CFR defines the expected target before the executable update; MISOB explains the population conditioning before weighted retraining. This is a synthesis of these four deliberately selected examples, not a conference-wide frequency estimate.
Editorial starting package: choose artifacts that test your actual claims. Quantities below are planning suggestions, not counts inferred from all category members or conference rules.
One diagram when multiple paths or update stages matter
Mark teachers, students, pseudo-labels, frozen weights, detached gradients and inference-only operations.
When: Use gradient arrows or phase panels when an ordinary module diagram would hide the actual intervention.
An explicit objective; an algorithm box if iterative or stateful
Define variables, positive/negative samples, weights, updates and stopping rather than listing an unexplained weighted sum.
When: A short closed-form loss may need no pseudocode; a multistage update usually does.
One task-specific table or paired baseline/new-rule rows
Hold backbone, supervision and budget fixed where possible so a training-rule claim is not merely a larger-model claim.
When: Show unmatched leaderboard context separately from causal design comparisons.
One appropriate panel per central explanatory claim
Measure the suspected mismatch: gradient behavior, confusing classes, reliability, alignment or temporal persistence.
When: Use the quantity the paper actually claims to improve, not an arbitrary feature visualization.
A compact control table and consequential parameter curves
Distinguish new supervision, loss terms, schedules and extra compute; retain collapse or negative runs.
When: Do not invent large sweeps for inconsequential parameters or hide tuning costs.
Optional
State assumptions and whether the implemented approximation satisfies them.
When: A proof is needed for a formal claim, not as decoration for an empirical improvement.
Keep defining objectives, assumptions, update order and model-selection principles intelligible in the main text. Route proofs, exact hyperparameters, sampling and full controls to precise supporting locations, without treating an uninspected pointer as verified evidence.
Separate supervisory paths and test each named operation under a specified modality and domain protocol.
Distinguish the training architecture, inference stopping rule and measured runtime.
Tie expected-target statements to implementation and distinguish exploitability from opponent-specific reward.
Define the evaluated population and absolute burden before interpreting parity improvements.
Boundary: This route contains 4 primary editorial assignments in the 16-paper inspected subset (1058 inventory records). It is not an official track, a universal writing standard or evidence that these choices caused acceptance.
See inline examples of these artifact styles · Compare every member in the measurement ledger
Progressive student initialization, distillation weighting and mixup form the central training intervention; adaptive inference and quality-cost comparisons test the resulting sequence.
The argument centers on coordinated optimization and EMA-teacher supervision rather than a new dataset or standalone inference architecture.
The paper combines a critical/formal reframing of recourse fairness with MISOB, an iterative burden-weighted training procedure. The primary editorial label identifies that practical intervention, not a claim that the paper contains only optimization work.
The central deliverable is a neural training and sampling update that approximates discounted/predictive CFR; expected-target results explain that update rather than replacing it with a standalone formal-analysis contribution.
A formal characterization, algebraic result, guarantee or theoretically derived explanation organizes the main argument, with algorithms and experiments testing consequences or practical approximations.
Corpus context · 2 papers: median 3 main figures and 1 main tables; body span ≈ 7.06 pages; attached appendix span ≈ Not resolved pages (0 resolved boundaries). Separate source files: median Not resolved whole-file pages among 0 papers; these can repeat main text and are not appendix-only lengths. Extraction-assisted observations—not drafting targets.
The two theory examples make assumptions and scope part of the organizing structure. Matrix completion presents its information setting, seven assumptions, algorithm and bound before two named experimental claims. The label-noise analysis uses a condition-to-result roadmap and coordinated dynamical plots, but changes the update rule for Phase II and treats SAM as an empirical extension. The common lesson is not a theorem or figure quota: each formal claim and illustrative experiment needs its own conditions and logical role.
Editorial starting package: choose artifacts that test your actual claims. Quantities below are planning suggestions, not counts inferred from all category members or conference rules.
One when it makes unfamiliar definitions operational
Connect algebraic symbols to camera geometry, representations or update behavior.
When: Prefer a reusable symbolic example over an unrelated network schematic.
One short map or comparison table when several regimes are involved
Keep quantifiers, non-degeneracy conditions, approximation regimes and guarantee types visible.
When: Especially important when exact theory motivates an approximate practical solver.
As many as the actual logical claims need
State what is proved and identify the central argument before full technical details.
When: No target theorem count; empirical observations must not be promoted to universal guarantees.
Only when there is an empirical consequence to test
Test the predicted difficult regime, scaling, geometry or optimizer behavior under stated assumptions.
When: A finite experiment illustrates or challenges a consequence; it does not prove an asymptotic statement.
Include where needed to prevent a tempting overgeneralization
Explain the role of assumptions and distinguish sufficient from necessary conditions.
When: A theoretical paper need not manufacture a broad benchmark leaderboard if it is irrelevant to the result.
Keep the learning setting, essential assumptions, exact claim and proof intuition intelligible in the main text. Route complete proofs, constants, technical cases and simulation configuration to precise accessible locations. Do not let a phase diagram imply an unconditional theorem when the analysis changes optimizer, model class or entry assumptions.
Give a theorem diagnostic and a practical variant distinct experimental roles; the real-data autoencoder is not the unchanged theoretical SVD algorithm.
Keep the update-rule change and additional assumptions beside the result; a matched-looking SAM plot is an empirical analogy, not another proved guarantee.
Boundary: This route contains 2 primary editorial assignments in the 16-paper inspected subset (1058 inventory records). It is not an official track, a universal writing standard or evidence that these choices caused acceptance.
See inline examples of these artifact styles · Compare every member in the measurement ledger
The central contribution is a conditional analysis of learning dynamics in a two-layer linear model. Empirical motivation and a SAM comparison support the theoretical story; they are not a separate verified general theorem for nonlinear networks.
The central new knowledge is a conditional generalization/excess-risk analysis for a semi-supervised matrix-completion setting. Algorithm 1 and the experiments instantiate and probe that theory rather than changing the primary writing type to an architecture paper.
The main new knowledge is a controlled measurement, diagnostic, comparison, failure analysis or audit. A proposed remedy, model or tool can follow from that evidence.
Corpus context · 1 papers: median 5 main figures and 8 main tables; body span ≈ 7.13 pages; attached appendix span ≈ Not resolved pages (0 resolved boundaries). Separate source files: median 13 whole-file pages among 1 papers; these can repeat main text and are not appendix-only lengths. Extraction-assisted observations—not drafting targets.
The selected DataMind study asks why data-analysis agents fail, organizes controlled questions around candidate capabilities, and only then develops a training-data intervention. Its worked example, sampled-error composition, resource sweeps and model comparisons have different denominators. Its separately inspected extended appendix supplies curation and prompt artifacts. This one study illustrates an empirical argument; it does not establish the prevalent AAAI analysis style or a universal capability hierarchy.
Editorial starting package: choose artifacts that test your actual claims. Quantities below are planning suggestions, not counts inferred from all category members or conference rules.
One compact view
Expose information access, sampling, intervention and matched controls.
When: Essential when teacher labels, dataset splits, task direction or model access determine the interpretation.
One clear entry point
Display the relationship, denominator and variability before a proposed explanation.
When: Show representative ranges rather than selected successes alone.
One destination per central explanatory claim
Distinguish the proposed mechanism from compute, label quality, dataset composition or evaluation artifacts.
When: Correlation and a fitted trend require different language from a controlled causal intervention.
A small aligned set with selection criteria
Make measured failures inspectable and include ambiguous/negative cases where relevant.
When: Qualitative cases illustrate a study; they do not replace prevalence estimates.
As needed by generalization or practicality claims
Repeat the protocol on different tasks/models/settings and report the actual cost.
When: Keep missing, unavailable or diverged results distinct from zero.
Make data sources, model/judge versions, sample budgets, prompts, filtering rules and final training configurations operationally inspectable. Separate controlled diagnostic settings from the follow-up recipe. An extended file may repeat the main manuscript, so record the actual appendix interval and version instead of counting every file page as new support.
Tie each finding to its protocol, retain resource-dependent reversals and unresolved bin-label inconsistencies, and keep the data recipe reproducible in a versioned appendix.
Boundary: This route contains 1 primary editorial assignments in the 16-paper inspected subset (1058 inventory records). It is not an official track, a universal writing standard or evidence that these choices caused acceptance.
See inline examples of these artifact styles · Compare every member in the measurement ledger
The central organizing contribution is a structured empirical investigation of data-analysis agent capabilities and training-data choices. The subsequent synthesis method is a follow-up use of those findings, not a reason to relabel the entire paper as a new architecture.
A new corpus, acquisition/annotation process, evaluation suite or measurement protocol is the primary reusable deliverable. A companion model demonstrates utility without replacing the need to validate the resource.
Corpus context · 1 papers: median 7 main figures and 1 main tables; body span ≈ 7.09 pages; attached appendix span ≈ Not resolved pages (0 resolved boundaries). Separate source files: median Not resolved whole-file pages among 0 papers; these can repeat main text and are not appendix-only lengths. Extraction-assisted observations—not drafting targets.
The single inspected benchmark example, VCapsBench, gives different denominators for videos, questions and evaluation dimensions; describes the annotation pipeline before model rankings; and distinguishes accuracy, inconsistency and coverage. It also discloses potential annotation-model bias. This one case illustrates an argument pattern, not a reliable distribution or standard for AAAI benchmark papers.
Editorial starting package: choose artifacts that test your actual claims. Quantities below are planning suggestions, not counts inferred from all category members or conference rules.
One compact resource table
Compare capabilities, annotation units, acquisition conditions and evaluation tasks, not only total size.
When: Use explicit partial or unavailable coverage rather than forcing binary superiority.
One process diagram with examples
Show raw inputs, transformations, human/machine verification and final annotations.
When: Distinguish generated labels, human corrections and measured ground truth.
A distribution, hierarchy or protocol matrix suited to the labels
Show subjects/scenes/classes, long tails, acquisition diversity and what is held out.
When: Avoid conflating frames, tracks, objects, rendered variants and independent subjects.
Focused tables with the same taxonomy used in the protocol
Demonstrate what the resource reveals and expose non-dominating models or hard subgroups.
When: A high average score alone does not validate label quality or benchmark relevance.
One verification analysis when annotation or metric validity is a claim
Document agreement, quality checks, adversarial controls, contamination checks or calibration.
When: Select the check that addresses this resource's actual failure risks; do not imply a single check proves complete validity.
Make provenance, annotation prompts, judge/model revisions, quality control, metric denominators and release conditions inspectable. State actual inspection/access limits; code or project links alone do not prove reproducibility.
Explain what the benchmark measures and how its targets are produced before presenting a leaderboard.
Boundary: This route contains 1 primary editorial assignments in the 16-paper inspected subset (1058 inventory records). It is not an official track, a universal writing standard or evidence that these choices caused acceptance.
See inline examples of these artifact styles · Compare every member in the measurement ledger
The reusable deliverables are a video/QA benchmark, annotation workflow and measurement protocol; evaluated captioning models demonstrate the resource rather than constitute a new architecture.
The central contribution changes execution, inference, memory, acquisition cost or resource scaling while preserving a stated quality or correctness contract.
Corpus context · 2 papers: median 4 main figures and 2.5 main tables; body span ≈ 7.14 pages; attached appendix span ≈ Not resolved pages (0 resolved boundaries). Separate source files: median Not resolved whole-file pages among 0 papers; these can repeat main text and are not appendix-only lengths. Extraction-assisted observations—not drafting targets.
The two inspected examples identify the operation that causes cost and the information contract of its replacement. CATP places pruning before and inside the decoder, then compares latency and quality; DeepProofLog explains a semantic reformulation with symbolic examples and separates exact from approximate execution. Both illustrate the need to preserve failures and non-wins rather than collapse all resource and quality measures into “efficient.”
Editorial starting package: choose artifacts that test your actual claims. Quantities below are planning suggestions, not counts inferred from all category members or conference rules.
One focused breakdown or controlled comparison
Explain which stage is costly and why the proposed decision variable could help.
When: Use matched budgets when an oracle analysis motivates an approximate implementation.
One overview with needed stage or memory insets
Mark preprocessing, cache/reuse, training versus inference, hardware calls and synchronization.
When: The figure should expose added overhead as well as removed work.
One panel per materially different workload
Pair quality with latency, throughput, memory or acquisition effort at comparable operating points.
When: State hardware, precision, batch/sequence/image size and what the timer includes.
A table or stacked breakdown
Show whether an isolated saving survives the entire pipeline.
When: Do not substitute FLOPs, token count or number of steps for wall-clock measurements without qualification.
Curves only for the dimensions the system claims to scale
Measure view/frame count, problem size, cache refresh or device constraints and retain adverse regimes.
When: Changing hardware or input capacity is a different comparison from identical-workload acceleration.
As needed for the execution contract
Make reuse, selection, fallback, stopping and approximation explicit.
When: A claimed exact or distribution-preserving procedure needs a precise argument beyond similar output scores.
Provide workload, hardware, timing scope, implementation settings and exact/approximate distinctions. Keep the defining computation and failure/timeout meaning in the main body; use support for detailed variants and derivations.
A pruning fraction and FLOP reduction are not a substitute for total inference latency.
Teach the representation change before the solver, and preserve timeouts and completed-but-inaccurate runs separately.
Boundary: This route contains 2 primary editorial assignments in the 16-paper inspected subset (1058 inventory records). It is not an official track, a universal writing standard or evidence that these choices caused acceptance.
See inline examples of these artifact styles · Compare every member in the measurement ledger
The primary contribution changes inference-time token handling and is justified by paired quality and resource measurements.
The paper builds a scalable proof-search and learning system around goal-conditioned resolution and an MDP correspondence; paired accuracy/time experiments distinguish practical exact and approximate regimes.
The primary argument is a task-specific capability, physical measurement/reconstruction system or domain-grounded method whose validity depends on real acquisition, task assumptions and domain-appropriate evidence.
Corpus context · 3 papers: median 4 main figures and 3 main tables; body span ≈ 7.06 pages; attached appendix span ≈ Not resolved pages (0 resolved boundaries). Separate source files: median Not resolved whole-file pages among 0 papers; these can repeat main text and are not appendix-only lengths. Extraction-assisted observations—not drafting targets.
Three application studies make the domain constraint part of the argument, not only the dataset choice. GSGStain starts from imperfect correspondence between consecutive histology sections; MSAnchor explains a spectrum-conditioned molecular scaffold before its generator; FoAM distinguishes simulated task success, physical robot trials and instruction-conditioned goals. Their strongest transferable pattern is the explicit input and evaluation contract. Technical image metrics, molecular similarity and selected robot demonstrations are not interchangeable with clinical, laboratory or unrestricted deployment validation.
Editorial starting package: choose artifacts that test your actual claims. Quantities below are planning suggestions, not counts inferred from all category members or conference rules.
One setup or input/output view
Show sensors/modalities, calibration, measured versus generated quantities and what is available at use time.
When: Essential when hardware, multimodal cohorts or nonstandard labels determine the claim.
One overview with necessary physical or anatomical labels
Connect variables, modules and losses to the domain problem rather than only showing generic network boxes.
When: Keep measured inputs, latent physical quantities and predicted outputs visually distinct.
A compact grid or profile plot
Compare matched scenes/patients/objects with relevant maps, spectra, overlays or error views.
When: Label simulation, real data, pseudo-ground-truth and unavailable measurements; an appealing image is not independent validation.
One table per incompatible endpoint
Report the relevant physical/clinical/behavioral metric, units, uncertainty and evaluation split.
When: Separate image quality from geometry, survival discrimination, calibration or downstream behavior.
Controls chosen for the proposed mechanism
Test key physical assumptions, modality use or pipeline components; include an independent reference when feasible.
When: Do not label a model-derived pseudo-label as independent truth or infer clinical benefit from a technical benchmark.
As appropriate to the deployment claim
State calibration, capture, annotation, compute and failure requirements.
When: A successful laboratory demonstration does not establish unrestricted field or clinical reliability.
Keep the domain object, acquisition or information constraint, deployed inputs and decisive evaluation protocol in the main body. Provide detailed preprocessing, split units, configurations, trial denominators, failures and assets in versioned support. If supporting material was not inspected, retain that limit rather than assuming its promised contents exist.
Pair aligned qualitative evidence with numeric controls, while explaining why weakly paired tissue changes the meaning of SSIM.
Define the representation before the pipeline and distinguish exact recovery from similarity and ground-truth from predicted conditioning.
Separate training and deployment, preserve physical successes/trials, and name the scope of a generated-goal comparison.
Boundary: This route contains 3 primary editorial assignments in the 16-paper inspected subset (1058 inventory records). It is not an official track, a universal writing standard or evidence that these choices caused acceptance.
See inline examples of these artifact styles · Compare every member in the measurement ledger
The primary argument connects a multi-task robotic manipulation problem to goal conditioning and foresight-augmented imitation, evaluated in simulation and on a physical robot. Application is the editorial organizing label; the auxiliary training objective and benchmark are supporting contributions.
The central argument translates an adjacent-section pathology constraint into graph-guided virtual staining and evaluates that application. This editorial primary label does not deny the paper's architectural and objective contributions.
The paper centers a mass-spectrometry structure-identification task and its domain-specific anchor representation, with a learned two-stage solution. Application is an editorial primary label, not the conference's official classification or a claim that the method lacks algorithmic novelty.
