Affinity-Graph Subset Audit for Spectral Representation Bounds¶
- Task ID:
computer_science.affinity_subset_bound_audit - Domain:
computer_science - Subdomain:
self_supervised_learning_theory - Status:
test - Tags:
contrastive_learning,self_supervised_learning,spectral_clustering,graph_diagnostics,generalization_bounds,difficult_examples
Public Summary¶
This page is generated from task metadata and selected public-safe excerpts.
Example B1 Prompt Excerpt¶
You are given an anonymized affinity-graph instance derived from a similarity-graph model for unsupervised contrastive representation analysis. Build a reproducible audit from the files in `data/`. Write the executable script as `analysis.py` at the workspace root, and write every required artifact under `results/`.
The public instance has `{{ n_items }}` items. The plausible number of latent affinity groups is between `{{ candidate_group_count_min }}` and `{{ candidate_group_count_max }}`.
## Input files
- `data/affinity_matrix.npy`: a square symmetric nonnegative float matrix. Rows and columns follow item id order `item_0000`, `item_0001`, ...
- `data/screening_table.csv`: row-level affinity summaries for the same item ids.
- `data/metadata.json`: file schema, candidate group counts, and required output columns.
The latent labels, disruptive subset, and graph coefficients are not supplied.
## Model to recover
Notes¶
- This page is a generated site artifact.
- Higher-level prompt details and internal benchmark specifics may remain intentionally undisclosed.