Red Team: tehruhn

[tehruhn]

Summary

• Submit a Red Team JSON file at red_team_submissions/tehruhn.json with 1,000 differentiating CIFAR100 stimuli.
• Each entry follows the required schema (dataset_name, image_identifier) and uses unique (dataset_name, image_identifier) pairs.
• The submitted set is generated via an iterative disagreement-focused selection process that increases cross-model divergence relative to random baselines.

Method (brief)

• Models used: resnet18, resnet34, resnet50, densenet121, vit_b_16.
• Scoring objective per image: 0.4*mean_pairwise_js + 0.2*top1_disagreement + 0.4*resnet_pair_js.
• Diversity constraint: class cap of 20 images/class during top-1000 selection.

Validation

• 1,000 total entries.
• Unique (dataset_name, image_identifier) pairs.
• JSON keys exactly match required submission fields.

[tehruhn]

Closing this PR in favor of #4, which uses the correctly named head branch .