architecture-zoo

Name: architecture-zoo
Author: Aperivue

208stars53forks

How to install architecture-zoo?

aperivue/medsci-skills/architecture-zoo

$npx -y skills add Aperivue/medsci-skills --skill architecture-zoo

Installs into the current project.

Summary

Choose a model architecture for a medical-imaging research question before scaffolding. Maps the task (classification, segmentation, detection, transfer), modality and dimensionality, labelled-data scale, and class imbalance to a shortlist of architectures, each grounded in its s

Files · 1

View on GitHub

SKILL.md

1	# Architecture-Zoo Skill
2
3	## Purpose
4
5	This skill turns a medical-imaging research question into a paper-grounded architecture choice —
6	so the build starts from the right archetype (and a known validation setup) rather than from whatever is
7	fashionable, and the choice carries its source citation into the Methods. It is the front end of the
8	model-engineering lane: `architecture-zoo (choose)` → `/model-scaffold (build)` → `/model-validation
9	(validate)`.
10
11	It is advisory (Layer D): it writes a short decision note, never code or weights. The actual repo is
12	`/model-scaffold`. It describes archetypes and the task → family → constraint logic, not a live SOTA
13	leaderboard (SOTA churns; the logic does not).
14
15	## When to use
16	- You need to pick an architecture/backbone for a classification, segmentation, detection, or
17	transfer-learning question and want it grounded in the literature with a sensible default.
18
19	## When NOT to use
20	- Generating the runnable repo → `/model-scaffold`.
21	- Auditing a trained model's validation design → `/model-validation`.
22	- Metrics / calibration → `/model-evaluation` + `/analyze-stats`.
23	- General study/validity design → `/design-study`; AI-vs-expert benchmark → `/design-ai-benchmarking`.
24	- LLM / MLLM → `/mllm-eval`.
25
26	## Workflow
27
28	### Phase 1 — Frame the question
29	State the task (classification / segmentation / detection / transfer), the **modality +
30	dimensionality (2-D vs 3-D volume), the labelled-data scale** (events / structures, not just
31	images), label availability (lots / few / unlabelled pool), and constraints (class imbalance,
32	small structures, interpretability, deployment compute).
33
34	### Phase 2 — Walk the decision tree
35	Open `${CLAUDE_SKILL_DIR}/references/index.md` and follow task → constraints → default pick. It routes to
36	a family card.
37
38	### Phase 3 — Read the family card
39	- `${CLAUDE_SKILL_DIR}/references/classification.md` — ResNet / DenseNet / EfficientNet / Inception /
40	ViT / Swin / DeiT.
41	- `${CLAUDE_SKILL_DIR}/references/segmentation.md` — U-Net / 3-D U-Net / V-Net / Attention & Residual
42	U-Net / nnU-Net / SegResNet / Swin-UNETR / Mask R-CNN.
43	- `${CLAUDE_SKILL_DIR}/references/detection.md` — R-CNN family / Faster R-CNN + FPN / Mask R-CNN /
44	RetinaNet / YOLO / DETR.
45	- `${CLAUDE_SKILL_DIR}/references/synthesis.md` — Pix2Pix / CycleGAN / SPADE / diffusion (DDPM, latent) /
46	VAE / fastMRI reconstruction.
47	- `${CLAUDE_SKILL_DIR}/references/foundation_models.md` — SAM / MedSAM / MedSAM2 / TotalSegmentator /
48	SegVol / BiomedCLIP / DINO / MAE / SimCLR / MoCo.
49	- `${CLAUDE_SKILL_DIR}/references/graph.md` — GCN / GraphSAGE / GAT / GIN / BrainGNN for brain
50	connectomes & population graphs (integrate PyTorch Geometric / DGL; not scaffolded by model-scaffold).
51	Each card gives the paper, core idea, when-to-use, medical-imaging use, reference implementation, and the
52	typical validation/experiment setup for that architecture class.
53
54	### Phase 4 — Write the decision note
55	Record `decisions/architecture_choice.md`: the task, the chosen architecture, its **source
56	paper, the reason against the constraints, the runner-up + why not**, and the matching
57	`/model-scaffold` template. Naming the source paper is mandatory; cite, never invent, any benchmark
58	number.
59
60	### Phase 5 — Hand off
61	Carry the decision note to `/model-scaffold` (instantiate the template), then `/model-validation`
62	(split / validation design), `/model-evaluation` + `/analyze-stats` (metrics), and `/write-paper`
63	(the Methods cite the architecture's source paper).
64
65	## Anti-Hallucination
66
67	- *Never recommend an architecture without naming its source paper.

People who liked this also viewed

6 picks

Type

architecture-zoo

208stars53forks

View on GitHub

How to install architecture-zoo?

aperivue/medsci-skills/architecture-zoo

$npx -y skills add Aperivue/medsci-skills --skill architecture-zoo

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Architecture-Zoo Skill
2
3	## Purpose
4
5	This skill turns a medical-imaging research question into a paper-grounded architecture choice —
6	so the build starts from the right archetype (and a known validation setup) rather than from whatever is
7	fashionable, and the choice carries its source citation into the Methods. It is the front end of the
8	model-engineering lane: `architecture-zoo (choose)` → `/model-scaffold (build)` → `/model-validation
9	(validate)`.
10
11	It is advisory (Layer D): it writes a short decision note, never code or weights. The actual repo is
12	`/model-scaffold`. It describes archetypes and the task → family → constraint logic, not a live SOTA
13	leaderboard (SOTA churns; the logic does not).
14
15	## When to use
16	- You need to pick an architecture/backbone for a classification, segmentation, detection, or
17	transfer-learning question and want it grounded in the literature with a sensible default.
18
19	## When NOT to use
20	- Generating the runnable repo → `/model-scaffold`.
21	- Auditing a trained model's validation design → `/model-validation`.
22	- Metrics / calibration → `/model-evaluation` + `/analyze-stats`.
23	- General study/validity design → `/design-study`; AI-vs-expert benchmark → `/design-ai-benchmarking`.
24	- LLM / MLLM → `/mllm-eval`.
25
26	## Workflow
27
28	### Phase 1 — Frame the question
29	State the task (classification / segmentation / detection / transfer), the **modality +
30	dimensionality (2-D vs 3-D volume), the labelled-data scale** (events / structures, not just
31	images), label availability (lots / few / unlabelled pool), and constraints (class imbalance,
32	small structures, interpretability, deployment compute).
33
34	### Phase 2 — Walk the decision tree
35	Open `${CLAUDE_SKILL_DIR}/references/index.md` and follow task → constraints → default pick. It routes to
36	a family card.
37
38	### Phase 3 — Read the family card
39	- `${CLAUDE_SKILL_DIR}/references/classification.md` — ResNet / DenseNet / EfficientNet / Inception /
40	ViT / Swin / DeiT.
41	- `${CLAUDE_SKILL_DIR}/references/segmentation.md` — U-Net / 3-D U-Net / V-Net / Attention & Residual
42	U-Net / nnU-Net / SegResNet / Swin-UNETR / Mask R-CNN.
43	- `${CLAUDE_SKILL_DIR}/references/detection.md` — R-CNN family / Faster R-CNN + FPN / Mask R-CNN /
44	RetinaNet / YOLO / DETR.
45	- `${CLAUDE_SKILL_DIR}/references/synthesis.md` — Pix2Pix / CycleGAN / SPADE / diffusion (DDPM, latent) /
46	VAE / fastMRI reconstruction.
47	- `${CLAUDE_SKILL_DIR}/references/foundation_models.md` — SAM / MedSAM / MedSAM2 / TotalSegmentator /
48	SegVol / BiomedCLIP / DINO / MAE / SimCLR / MoCo.
49	- `${CLAUDE_SKILL_DIR}/references/graph.md` — GCN / GraphSAGE / GAT / GIN / BrainGNN for brain
50	connectomes & population graphs (integrate PyTorch Geometric / DGL; not scaffolded by model-scaffold).
51	Each card gives the paper, core idea, when-to-use, medical-imaging use, reference implementation, and the
52	typical validation/experiment setup for that architecture class.
53
54	### Phase 4 — Write the decision note
55	Record `decisions/architecture_choice.md`: the task, the chosen architecture, its **source
56	paper, the reason against the constraints, the runner-up + why not**, and the matching
57	`/model-scaffold` template. Naming the source paper is mandatory; cite, never invent, any benchmark
58	number.
59
60	### Phase 5 — Hand off
61	Carry the decision note to `/model-scaffold` (instantiate the template), then `/model-validation`
62	(split / validation design), `/model-evaluation` + `/analyze-stats` (metrics), and `/write-paper`
63	(the Methods cite the architecture's source paper).
64
65	## Anti-Hallucination
66
67	- *Never recommend an architecture without naming its source paper.

People who liked this also viewed

6 picks

Type