Research Interests

Wireless communication and networking
Distributed/on-device machine learning (ML)
Knowledge-aided ML for wireless communication and networking
Edge computing/caching
Resource optimization

Continual Federated Learning for Resource-constrained Wireless Applications

Continual data arrival: delete old training samples to make space for new ones (fixed/limited storage)
Quantizes gradients for communication efficiency
Uses gradient-similarity and participation-based online score function and global update step size to control error due to data distribution shift

Federated Learning for Autonomous Ground Vehicles

Impact of mobility in vehicular edge federated learning: new aggregation rules - guided by mobility and sojourn time
Joint optimization of wireless (scheduling and power allocation) and learning parameters (client selection, CPU frequency, and local rounds) to facilitate the convergence

Federated Learning for Resource-constrained Edge Clients

Model pruning-aided hierarchical federated learning (PHFL) to address resource constraints
Optimize joint resources to facilitate convergence bound - optimize wireless resources to minimize pruning errrors
Comparable performance with non-pruned upper bound with significantly lower bandwidth and energy expense

Federated Learning for Resource-constrained Unmanned Aerial Vehicles (UAVs)

Communication- and computation-efficient online federated learning (C2EOFL) for resource-constrained UAVs
Uses model pruning and gradient quantization

Domain-Informed Machine Learning for Wireless Channel Modeling

Uses small language models (built on Transformer) and learnable graphs (graph neural networks)
Injects propagation knowledge to update the optimizer

Distributed Machine Learning for Beam Management / CSI Compression

FL for Beam Management

Continual multimodal sensing with onboard sensors
Continual federated learning for multi-step beam prediction

FL for CSI Compression

CSI compression from continual (multimodal) sensing data
Continual distributed learning for CSI compression
Domain-knowledge injection

Machine Learning for Content Caching

User-centric radio access technology (RAT) for delay-sensitive applications
Reinforcement learning (RL) for cache placement policy
Integrated solution for cache placement and content delivery to minimize long-term service delay

Hierarchical FL for Video Caching

Privacy-preserving 3-party (UE-ISP-CSP) collaborative solution for wireless video caching
Resource-aware hiearcharical federated learning (RawHFL) for content demand prediction

"It is during our darkest moments that we must focus to see the light." - Aristotle

Google Sites

Report abuse