Building Trust for Users and Regulators Alike: A Cost-Efficient PyTorch Pat...
summary
Claims a reproducible PyTorch pattern that embeds executable regulatory controls into training and inference using existing primitives, with zero new dependencies and code on GitHub. Uses Dispatcher, custom Autograd functions, and forward hooks to log lineage, enforce a fairness gate that halts training on group disparity, and trigger drift alerts (example: drift firing around day 50). Saves augmented checkpoints (ResNet18 example) as SHA-256 chained checkpoints. Reports addressing 34 of 37 EU Act items; three duties remain human or organizational.
AI-written summary · about
chapters (13)
- 0:00PyTorch compliance framework intro
- 2:16Model compliance roles
- 3:42Privacy governance gaps
- 5:07PyTorch primitives for compliance
- 7:17Runtime compliance checks
- 8:43t-comply engine orchestration
- 10:52Checkpointing with audit trails
- 12:15Fairness gates in training
- 13:39Micro-level consent enforcement
- 14:42Model drift monitoring
- 15:49Compliance across training runs
- 16:33EU Act coverage
- 17:39Legal framework adaptation
related talks
30mHow to Trust Your Deep Learning Code
17mPreserving integrity and trust in AI
08mOpen Source Al and PyTorch
27mControlling the Chaos: Reproducible Deep Learning Experiments Using Lightning and Hydra
44m$1 AI Guardrails: The Unreasonable Effectiveness of Finetuned ModernBERTs
47mGet to Know PyTorch - A Cutting Edge Open Source AI Framework for Deep Learning
36mAvoiding 5 missteps that undermine your AI readiness and success
19mSecuring AI Pipelines from Development to Production