Federated Averaging

Implement client partitioning, local SGD, weighted aggregation, partial participation, and non-IID experiments.

6 parts26 individual lessons2.2 estimated hours130 XP available

0%0 of 26 complete

No compressed chapters.

Every source step is its own lesson with intuition, concepts, correctly rendered MathJax mathematics, implementation, tests, mistakes, and a checkpoint.

PART 1

Model and Dataset Setup

Define the MLP classifier, build a synthetic labeled dataset, and split it into train and test sets.

0/3

001build mlp classifier+5 XP 002build synthetic dataset+5 XP 003train test split dataset+5 XP

PART 2

Client Data Partitioning

Distribute the training data across clients using IID and non-IID strategies and report per-client sample counts.

0/3

004partition data iid+5 XP 005partition data non iid+5 XP 006count client samples+5 XP

PART 3

Local Client Training

Iterate over a client's batches and run forward, loss, backward, and optimizer steps to train locally for several epochs.

0/4

007iterate client batches+5 XP 008compute batch loss+5 XP 009local sgd step+5 XP 010train client local+5 XP

PART 4

Parameter Utilities and Aggregation

Clone, load, initialize, add, and scale model state dicts to build the sample-weighted FedAvg aggregation.

0/6

011clone model state+5 XP 012load model state+5 XP 013initialize global state+5 XP 014add state dicts+5 XP 015scale state dict+5 XP 016aggregate weighted average+5 XP

PART 5

Communication Rounds and FedAvg Loop

Select participating clients, run a full communication round, evaluate accuracy, and drive the multi-round FedAvg training loop.

0/4

017select round clients+5 XP 018run communication round+5 XP 019evaluate accuracy+5 XP 020run fedavg+5 XP

PART 6

Experiments and Analysis

Compare against a centralized baseline and study IID vs non-IID gaps, the effect of local epochs, and the impact of client fraction.

0/6

021train centralized baseline+5 XP 022run fedavg iid+5 XP 023run fedavg non iid+5 XP 024compute non iid gap+5 XP 025rounds to target vs local epochs+5 XP 026accuracy vs client fraction+5 XP