AI Paper Atlas

Research map / Continual learning / Class-incremental

A Data-Free Approach to Mitigate Catastrophic Forgetting in Federated Class Incremental Learning for Vision Tasks

Sara Babakniya, Zalan Fabian, Chaoyang He, Mahdi Soltanolkotabi, Salman Avestimehr

NeurIPS 2023 · Continual learning › Class-incremental · approach: Class-balanced knowledge retention · 9 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add batch statistics loss to generator training (ablation)MFCL ResNet18 · CIFAR-100 (10 tasks, 50 clients)average accuracy (%): 38.51 -> 44.98Improved
add diversity loss to generator training (ablation)MFCL ResNet18 · CIFAR-100 (10 tasks, 50 clients)average accuracy (%): 38.44 -> 44.98Improved
add image prior loss to generator training (ablation)MFCL ResNet18 · CIFAR-100 (10 tasks, 50 clients)average accuracy (%): 43.87 -> 44.98Improved
restrict cross-entropy to current-task heads (ablation)MFCL ResNet18 · CIFAR-100 (10 tasks, 50 clients)average accuracy (%): 26.78 -> 44.98Improved
add classifier fine-tuning loss on real plus synthetic data (ablation)MFCL ResNet18 · CIFAR-100 (10 tasks, 50 clients)average accuracy (%): 37.33 -> 44.98Improved
add importance-weighted feature distillation loss (ablation)MFCL ResNet18 · CIFAR-100 (10 tasks, 50 clients)average accuracy (%): 30.86 -> 44.98Improved
add client-side modifications on top of FedAvg + generator (ablation)MFCL ResNet18 · CIFAR-100 (10 tasks, 50 clients)average accuracy (%): 25.69 -> 44.98Improved
add client-side modifications (CE, FT, KD) vs plain FedAvg + Gen (ablation)MFCL ResNet18 · CIFAR-100 (10 tasks, 50 clients)average forgetting (%): 60.46 -> 28.3Improved
add image prior loss (ablation)MFCL ResNet18 · CIFAR-100 (10 tasks, 50 clients)average forgetting (%): 29.47 -> 28.3No clear effect
use server-side data-free generator (MFCL) instead of client-trained GAN (FedCIL)MFCL · CIFAR-100 (10 tasks, 50 clients)average accuracy (%): 26.8 -> 44.98Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy