Stage 1. Generating Networks
Construct the directed networks to be studied before any degeneration begins.
Generative models, parameter sweeps, random seeds, and any imported connectome variants.
Network Degeneration Simulation Pipeline
Technical note on deterministic node and edge degeneration schedules for directed networks, with emphasis on experiment structure before downstream spiking simulation.
Question
How should we model progressive degeneration in a way that is both biologically interpretable and structurally reproducible? This note focuses on the scheduling logic: which nodes or edges disappear, in what order, and what that means for the resulting network instances before simulation begins.
Pipeline Stages
This outline separates the pipeline into clear stages, so the note can grow step by step without mixing setup, degeneration logic, structure readout, and simulation handoff.
Stage 2. Base Network
Specify or load the directed network that will be subjected to degeneration.
Network source, graph family, size, density, and any biological constraints.
Stage 3. Degeneration Policy
Choose the node- or edge-deletion rule that defines the order of structural loss.
The ten strategies, their ordering rules, and the rationale for comparing them.
Stage 4. Staged Extraction
Generate intermediate snapshots at controlled deletion levels rather than only the final damaged graph.
Stage fractions, snapshot naming, storage format, and reproducibility notes.
Stage 5. Structure Readout
Measure what each staged graph retains structurally before any dynamics are simulated.
Degree summaries, matching-derived quantities, components, and path-related observables.
Stage 6. Dynamics Handoff
Pass the staged graphs onward to spiking simulation or other downstream analyses.
Simulator inputs, parameter inheritance, and links back to postsimulation figures.
Edge Deletion Strategies
The gallery below uses one shared directed base graph and compares ten degeneration schedules against it: five edge-deletion schemes and five node-deletion schemes. The slider advances the deletion stage, while each SVG card shows what survives structurally under that policy.
- Which schedules preserve hubs, paths, and matching-support edges for longer
- How node-first and edge-first degeneration differ even at the same deletion stage
- How the same base graph can produce very different failure profiles under different orders
Base nodes: 6Base edges: 10Ordered matching edges: 4
All five edge strategies act on the same directed base graph. Ordered-pruning and resilient-pruning mark matching-priority links in green, including the removed ones as dashed traces.
Out-pruning
Remove outgoing edges node by node.
Remaining nodes6/6
Remaining edges6/10
In-pruning
Remove incoming edges node by node.
Remaining nodes6/6
Remaining edges6/10
Random-pruning
Fixed random edge schedule.
Remaining nodes6/6
Remaining edges6/10
Ordered-pruning
Remove matching-critical edges first.
Remaining nodes6/6
Remaining edges6/10
Resilient-pruning
Reverse the ordered schedule.
Remaining nodes6/6
Remaining edges6/10
Node Deletion Strategies
The node strategies delete vertices in different orders, then inherit the corresponding edge loss. This makes the hub-targeting policies easy to compare against weak-first and random schedules.
Increasing out-degree
Delete weak broadcasters first.
Remaining nodes4/6
Remaining edges5/10
Increasing degree
Delete low-degree nodes before hubs.
Remaining nodes4/6
Remaining edges4/10
Random node deletion
Fixed random node schedule.
Remaining nodes4/6
Remaining edges5/10
Decreasing degree
Delete hubs first.
Remaining nodes4/6
Remaining edges3/10
Decreasing out-degree
Delete strong broadcasters first.
Remaining nodes4/6
Remaining edges4/10
Summary
The project overview explains the full computational workflow, but the degeneration rules deserve their own space. This note is where we can compare the ten policies side by side, make their ordering assumptions visually explicit, and then connect those schedules to structure summaries and downstream dynamical outcomes.
Next Additions
The next pass can add richer bars and derived observables: component counts, degree loss, matching size, or E/I block summaries. That keeps the same SVG comparison frame while making the consequences of each schedule more quantitative.