{
  "id": 5085,
  "url": "https://arxiv.org/abs/2605.08137v1",
  "title": "Weight Pruning Amplifies Bias: A Multi-Method Study of Compressed LLMs for Edge AI",
  "summary": "Weight pruning is widely advocated for deploying Large Language Models on resource-constrained IoT and edge devices, yet its impact on model fairness remains poorly understood. We conduct a controlled empirical study of three instruction-tuned models (Gemma-2-9b-it, Mistral-7B-Instruct-v0.3, Phi-3.5-mini-instruct) across three pruning methods (Random, Magnitude, Wanda) at four sparsity levels (10-70%) on 12,148 BBQ bias benchmark items with 5 random seeds, totaling 2,368,860 inference records. O",
  "authors": "Plawan Kumar Rath, Rahul Maliakkal",
  "category": "research",
  "topics": "bias-fairness",
  "orgs": "mistral",
  "regions": null,
  "published_at": "2026-05-02T05:27:40.000Z",
  "fetched_at": "2026-07-14T16:31:31.210Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/5085",
  "original_url": "https://arxiv.org/abs/2605.08137v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}