[ 🏠 Home / 📋 About / 📧 Contact / 🏆 WOTM ] [ b ] [ wd / ui / css / resp ] [ seo / serp / loc / tech ] [ sm / cont / conv / ana ] [ case / tool / q / job ]

/q/ - Q&A Central

Help, troubleshooting & advice for practitioners
Name
Email
Subject
Comment
File
Password (For file deletion.)

File: 1782165327679.jpg (84.78 KB, 1024x1024, img_1782165318758_qu43bvif.jpg)ImgOps Exif Google Yandex

b59bf No.1844

just stumbled across this breakdown on data poisoning techniques like label flipping and backdoors. it's pretty wild how easy it is to corrupt an entire training pipeline w/ gradient manipulation if you aren't careful.

full read: https://www.infoq.com/articles/understanding-ml-model-poisoning/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=global

b59bf No.1845

File: 1782165479834.jpg (264.92 KB, 1024x1024, img_1782165465660_kz5iblty.jpg)ImgOps Exif Google Yandex

the real nightmare is when u cant even audit the training set because its being pulled from an unverified upstream source. always run a data_drift_check script or use something like Great Expectations to catch those distribution shifts before they hit ur weights.


[Return] [Go to top] Catalog [Post a Reply]
Delete Post [ ]
[ 🏠 Home / 📋 About / 📧 Contact / 🏆 WOTM ] [ b ] [ wd / ui / css / resp ] [ seo / serp / loc / tech ] [ sm / cont / conv / ana ] [ case / tool / q / job ]
. "http://www.w3.org/TR/html4/strict.dtd">