Dusit Niyato

Research Overview

Dusit Niyato focuses on the optimization of next-generation wireless networks, semantic communication, and distributed edge intelligence. His work explores the integration of multi-agent reinforcement learning and generative AI to manage complex communication systems, including non-terrestrial and unmanned aerial networks. Recently, his research has expanded into federated customization of large models and secure, intent-driven semantic frameworks for autonomous systems.

Research Trajectory

Niyato’s research addresses the complex resource allocation and sensing challenges inherent in next-generation wireless systems, particularly those involving unmanned aerial vehicles (UAVs) and non-terrestrial networks (NTNs). Beginning with architectures for aerial swarms, he and his collaborators introduced an integrated sensing and communications (ISAC) framework in Pseudo-Random TDM-MIMO FMCW Based Millimeter-Wave Sensing and Communication Integration for UAV Swarm. This work leverages millimeter-wave signals and pseudo-random antenna switching to achieve simultaneous high-speed data transmission and high-precision radar sensing without traditional velocity ambiguity limits. Expanding this focus to global coverage architectures, he developed collaborative deep reinforcement learning schemes to manage the highly dynamic resource optimization required by low-earth-orbit satellite constellations in Toward Intelligent Non-Terrestrial Networks Through Symbiotic Radio: A Collaborative Deep Reinforcement Learning Scheme.

This interest in autonomous aerial systems naturally intersected with advanced multi-agent reinforcement learning. In Task Assignment and Exploration Optimization for Low Altitude UAV Rescue via Generative AI Enhanced Multi-agent Reinforcement Learning (and its subsequent journal expansion Task Assignment and Exploration Optimization for Low Altitude UAV Rescue via Generative AI Enhanced Multi-Agent Reinforcement Learning), his team proposed a cooperation framework integrating UAVs, ground-embedded robots, and airships for emergency rescue operations. This work contributes to the broader area of Diffusion Models for Reinforcement Learning Policies, utilizing generative AI to manage the high computational demands of dynamic task assignment and exploration across multiple autonomous agents.

Concurrently, Niyato has significantly advanced the paradigm of semantic communications, contributing to areas such as Task-Agnostic Semantic Communication Frameworks, which prioritizes the transmission of meaning and task-relevant information over raw data. To address the privacy vulnerabilities of these systems, he proposed integrating homomorphic encryption into joint source-channel coding models in Secure Semantic Communication With Homomorphic Encryption, enabling end-to-end encryption of semantic features. He further expanded on task-oriented architectures by proposing a shift from traditional high-fidelity digital twins to Goal-Oriented Semantic Twins (GOST) in Twinning for Space-Air-Ground-Sea Integrated Networks: Beyond Conventional Digital Twin Towards Goal-Oriented Semantic Twin. This approach optimizes massive, resource-constrained networks by prioritizing utility, aligning closely with the research direction of Intent-Driven Translation for Autonomous Systems.

Moving into 2026, his focus on decentralized and edge intelligence deepened. Recognizing the communication overhead and privacy concerns of centralized machine learning, he developed a lightweight framework in Lightweight Federated Learning Over Wireless Edge Networks that integrates wireless transmission power control, model pruning, and gradient quantization. Building on this decentralized approach, he addressed the challenge of adapting massive foundation models to private tasks in Federated Customization of Large Models: Approaches, Experiments, and Insights. By introducing Federated Prefix-Tuning, this work demonstrated how aggregating a small "prefix optimizer" network allows clients to achieve near-centralized performance with minimal communication overhead.

In parallel, Niyato continued to refine the physical and cognitive layers of 6G networks. He explored the integration of simultaneously transmitting and reflecting reconfigurable intelligent surfaces (STAR-RIS) with UAVs in HCLN: A Hierarchical Collaborative Learning Network for UAV-STAR-RIS-Assisted Wireless Communications, designing a hierarchical collaborative learning network to overcome the high-dimensional, nonconvex challenges of aerial orientation fluctuations. At the cognitive layer, he addressed the complexities of multimodal data transmission in mobile edge computing through Edge Large AI Model Agent-Empowered Cognitive Multimodal Semantic Communication, proposing an agent capable of generating personalized communication policies tailored to diverse user intents.

His recent work also extends into sophisticated multi-agent reasoning and physical layer security. In CLAUSE: Agentic Neuro-Symbolic Knowledge Graph Reasoning via Dynamic Learnable Context Engineering, he structured context gathering for knowledge graph question answering as a budgeted learning problem, coordinating reinforcement learning-trained agents to curate information under strict latency constraints. Finally, he investigated physical layer security in Covert Communication Assisted by Movable Time-Modulated Arrays, leveraging the spatial position of antennas as a new degree of freedom and optimizing covert throughput via an attention-enhanced deep neural network.

Current Directions

The most recent papers in the evidence indicate a strong focus on deploying large AI models and agentic frameworks at the network edge. Current work investigates how cognitive agents can manage multimodal semantic communications, dynamically generating personalized transmission policies based on user intent and bandwidth constraints. Additionally, Niyato is exploring the federated customization of foundation models, seeking methods like prefix-tuning to adapt massive networks across decentralized clients without prohibitive communication costs. At the physical and network layers, his ongoing research addresses the optimization of 6G architectures, utilizing hierarchical collaborative learning to stabilize UAV-assisted reconfigurable intelligent surfaces and employing movable time-modulated arrays to enhance covert communications.