arXiv proposes FedReason for LLM-based aggregation in federated learning
The LLM analyzes weights in a sandbox and submits short Python aggregation programs every ten rounds.
Stories per day · 30 days
1 RSS
Relatedbyzantine-robustness1celeba1differential-privacy1diffusion-models1federated-learning1llm-aggregation1model-distillation1model-extraction1parameter-recovery1privacy-accounting1relu1weight-space1
10 stories, newest first · one card per story
Stories tagged #cifar-10
Paper recovers neural-network parameters by extraction and distillation
On the largest synthetic net, Leaky ReLU recovered 51-57 deepest-layer neurons while ReLU recovered none.
Paper accounts privacy loss during diffusion-model sampling
On CIFAR-10 and CelebA-64, loss rises sharply while semantic content forms, motivating hybrid denoisers.
Space Over Statistic, Retraining Over Ranking: Auditing Training Labels Under Label-Only Backdoors
Label-only backdoor audit shows feature space, not the detector score, decides whether retraining removes the attack.
Feature Selective Model Collapse in Diffusion Models: Total Replacement versus Fixed-Budget Training
Fixed-budget diffusion retraining loses only fragile features, unlike full replacement with synthetic data.
Data Unlearning via Inverse Distillation
Inverse Distillation Unlearning distills diffusion and flow models into one-step generators while suppressing a designated forget set.
Common-Mode Collapse and Recovery in Direct Feedback Alignment
Direct feedback alignment can stall when a shared error mode saturates tanh hidden units.
Weight Pair Encoding: Inducing a Smaller Grammar in Neural Network Weights
WeightPE finetunes int8 weights so Re-Pair grammars shrink to 0.38–0.43x of QAT, costing about 1–2 accuracy points.
Learning Holographic Reduced Representations with Clifford Variational Autoencoders
Clifford-VAE learns torus embeddings that beat Gaussian VAEs on vector-symbolic binding and capacity tests.
Reinforcement Learning Inspired Black-box Adversarial Attacks for Computer Vision
A reinforcement-learning black-box attack fools vision models with fewer queries than earlier methods.