Dataset Viewer
The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Reproduction: Selecting Samples on Graphs (UGIES)
Independent reproduction for the ICML 2026 Agent Reproduction Challenge.
- Paper: Selecting Samples on Graphs: A Unified Dataset Pruning Framework for Lossless Training Acceleration
- arXiv: 2606.12913 · OpenReview: a3GdvuPItd · Max points: 6
- Agent: CLAUDE · No official code (independent reimplementation)
Outcome
- Claims 1–2 (theory): FULL numerical audits — MWCP objective is well-posed; submodularity + (1−1/e) greedy bound verified to double precision, control-checked.
- Claims 3, 6 (CIFAR pruning): INCONCLUSIVE (toy) — UGIES 31.2% < random 62.3% < full 74.6% at reduced scale; did not reproduce (negative signal, not a paper-level falsification).
- Claims 4–5 (ImageNet): INCONCLUSIVE — infeasible under the ≤55-min / ≤$0.75 budget.
One-command rerun
# theory audits (CPU, ~70s)
python src/experiments.py
# CIFAR pruning experiment (GPU; ran on HF Jobs t4-small)
python src/cifar_job.py --per-class 800 --ratio 0.5 --proxy-epochs 10 --train-epochs 35 --alpha 0.5
# figures
python figures/make_figures.py
Layout
src/ UGIES + experiments · outputs/ raw JSON results · figures/ Plotly HTML + CSV + poster PNGs · poster/ poster.html/png/embed · protocol.md pre-registered protocol.
Data: https://huggingface.co/datasets/uoft-cs/cifar10 · Job: https://huggingface.co/jobs/MarxistLeninist/6a654458db23d7a7ec1cd5a4
- Downloads last month
- 11