List view
Task 1: Dataset Non-IID Partitioning Strategy and Non-IID Visualization (Data Heterogeneity) Task 2: Comparison between different FL algorithms: FedAvg, FedOPT (FedAdam, FedAdagrad). Backup: FedProx, FedRS Task 3: Centralized baseline. Task 4: Comparison between 2 different non-IID levels. Task 5: Client Sampling Ratio (random sampling): 10% (low sampling rate), 20% (high sampling rate). Customized experiments for resource-constrained IoT devices in real-world deployments. Task 6: Error labels. Task 7: Missing labels. (see FedMultimodal section 4.4) Task 8: Fine-tuning a pre-trained model vs Transfer Learning vs Train from Scratch and their memory and communication usage vs. accuracy (see FedCV). Task 9: Varying number of frozen layers vs communication cost vs. accuracy (see FedNLP table 3). Task 10: Batch size vs. memory vs. accuracy (see FedCV). (Optional) Task 11: Gradient quantization: number of bits per gradient: 2-32 vs. accuracy. Gradient sparsity vs. gain in bandwidth vs. accuracy. (see FLUTE Table 5 and Table 6).
Overdue by 2 year(s)•Due by April 30, 2023•0/4 issues closedWISDM WIDAR VISDRONE UTHAR EMOGNITION D1 EMOGNITION D2 KITTI NUSCENE
Overdue by 2 year(s)•Due by April 30, 2023•0/2 issues closed