Paper formulates split federated unlearning and proposes SURE
SURE uses dual-importance estimation and selective perturbation, matching retraining at modest extra cost.
Stories per day · 30 days
2 RSS
Relatedprivacy3machine-unlearning2batch-normalization1edge-devices1edge-learning1federated-learning1red-teaming1text-reconstruction1
7 stories, newest first · one card per story
Stories tagged #split-learning
BatchNorm buffers can fake deletion in split-learning unlearning
CIFAR-10 forget accuracy fell 61.50 points from BN recalibration with every learnable weight fixed.
LLM-Attacker standardizes privacy tests for split language models
In a GPT-2 test, returned gradients raised token recovery from 94.20% to 97.38% and exact-document recovery from 13.71% to 37.77%.
What Gradients Add to Text Leakage in Split Language Models, Counted per Token and per Document
Split-learning observers recover 97% of GPT-2 tokens; gradients lift exact document recovery from 14% to 38%.
Importance-Aware Feature Sparsification for Wireless Split Learning
ICS sparsifies wireless split-learning features with server-side Grad-CAM scores, cutting communication cost.
LLM-Based Schema-Aware Split Learning for Privacy-Preserving Mental Distress Prediction Across Heterogeneous Surveys
Schema-aware split learning uses LLaMA-3.2-3B-Instruct as shared semantic encoder to harmonize heterogeneous mental-health surveys while raw data stays local.
Don't Send What You Don't Need: Question-Guided Token Pruning as a Privacy Defense for Vision-Language Models
QPriv-VL prunes privacy-sensitive visual tokens in federated/split VQA, cutting membership-inference success on VQA-RAD from 0.99 to 0.76-0.79 using ~40% of tokens.