Please enable JavaScript.
Coggle requires JavaScript to display documents.
A Survey on Federated Learning for Resource-Constrained IoT Devices -…
A Survey on Federated Learning
for Resource-Constrained IoT Devices
Existing Studies on FL Models
Definition of FL
Taxonomy of FL
Partitioning sample: Horizontal,
Vertical and Hybrid
Machine learning model:
Linear models, K-means,
Random forest
Federation scale:
cross-device, cross-silo
Privacy mechanisms:
Differential privacy,
Randomized perturbation,
Cryptographic methods
Encouragement towards FL:
Incentives, Regulations
Distributed Learning and
Optimization Algorithms
FL Algorithms: FedAvg,
Local gradient descent,
FedProx, q-FedAvg
Distributed Learning
Distributed and Federated Optimization
Learning on Resource-constrained Devices
Communication Overhead
Heterogeneous Hardware
Limited Memory and Energy Budget
Scheduling
Energy Efficient Training of DNNs
Fairness in FL
Scalability of FL
Privacy Issues
Potential Solutions
Deploying Existing Algorithms to Reduce Communication Overhead
Convergence Guarantee in Asynchronous FL
Quantification of Statistical Heterogeneity
Data Cleaning and Handling False Data Injection
Reducing Energy Consumption and On-device Training
Managing Dropped Participants
Privacy Preservation
Applications of FL
Resource-Sufficient FL Apps
Resource-Constrained FL Apps
Future Directions:
local training dataset,
non-convex problem,
energy efficiency,
device-centric automatic wake-up mechanism,
client mobility,
statistical heterogeneity,
effective incentives mechanism design,
lightweight blockchain,
trust model