Research
Efficient Machine Learning at the Edge
Recent advances in efficient machine learning seek to overcome the energy, latency, and scalability challenges that limit the deployment of deep models on edge and embedded systems. Our research develops a spectrum of innovations spanning algorithm, architecture, and circuit levels, from binary and combinational logic neural networks that minimize computation and memory movement, to unary and domain-specific accelerators that maximize energy efficiency through lightweight hardware primitives. We further introduce unified frameworks, which virtualize nonlinear operations and general matrix multiplication across diverse encoding schemes to achieve adaptive accuracy-energy trade-offs. Complementing these are techniques like content-aware computation offloading and latent-weight pruning, which dynamically tailor model complexity and data flow to resource availability. Collectively, these efforts pave the way for intelligent, low-power learning platforms applicable to wearables, healthcare monitoring, and other real-time edge-AI systems.
platforms without compromising user experience or reliability.
Energy-Quality Co-Design for Low-Power Computing
Modern computing faces an urgent need for low-power solutions as performance scaling increasingly confronts thermal and energy limits. Low-power computing leverages approximation, adaptivity, and workload awareness to minimize energy while maintaining acceptable quality across diverse subsystems, from processors and memory to storage and edge networks. Techniques such as approximate arithmetic, input-dependent iterative computing, and dynamic voltage scaling exploit the natural error tolerance of machine learning, signal processing, and multimedia workloads to achieve substantial efficiency gains. Complementary advances in approximate storage and energy-aware video transcoding further extend system lifetime and reduce operational cost by aligning computation and memory precision with application-level quality needs. Together, these innovations enable sustainable, high-performance computing across cloud, edge, and embedded platforms without compromising user experience or reliability.
Embedded Intelligence for Applied AI and Physical AI
Our research advances precision livestock farming through multimodal sensing, edge intelligence, and AI-driven analytics to improve dairy cattle health and welfare. We developed implantable and wearable biosensing systems that enable real-time, energy-neutral monitoring of core body temperature, allowing early detection of heat stress with minimal intervention. To support large-scale data-driven management, we created a multimodal dataset that synchronizes physiological, behavioral, and environmental data from dairy herds. Building on this foundation, we introduced an AI-powered video querying system that translates complex barn video data into intuitive, natural-language insights for farmers. Beyond agriculture, these technologies have strong potential for broader healthcare applications, such as continuous human vital sign monitoring and intelligent patient care systems that leverage edge AI and multimodal sensing.
User-Centric Security and Privacy for Smart Systems
Our research advances human-centered security and privacy by leveraging physical, physiological, and contextual signals to enable trustworthy and seamless authentication in connected environments. We design systems that sense and exploit ambient or device-specific phenomena, such as vibrations, power line noise, electromagnetic radiation, and microphone imperfections, to achieve zero-involvement, spoof-resistant authentication. Extending into immersive and emerging interfaces, our work safeguards sensitive sensing modalities like eye tracking and interaction data through real-time, privacy-preserving mechanisms. Beyond devices, we also examine users' evolving security and privacy attitudes in connected ecosystems, such as smart homes, using large-scale, in-the-wild behavioral analyses. Together, these efforts aim to build a privacy-preserving foundation for ubiquitous and embodied computing, balancing usability, trust, and resilience across the digital and physical worlds.
Acknowledgements
Our research is made possible through the generous support of various government agencies, corporates, and academic institutions.
Federal Agencies
Corporates
International Government Agencies
Physical AI for Smarter, Sustainable Farming, and Beyond
At the intersection of artificial intelligence and agricultural engineering, Physical AI for Precision Agriculture brings intelligence directly into the farm environment. Our research focuses on developing smart, sensing systems—from energy-harvesting wearable ear tags and implantable biosensors to edge-computing devices and vision-based AI—that continuously monitor animal health, welfare, and behavior in real time. By integrating data such as body temperature, movement, and environmental conditions, these systems can detect early signs of heat stress, disease, or discomfort in individual cows.
August 13, 2025
MmCows: A Multimodal Dataset for Dairy Cattle Monitoring
🌟 NeurIPS 2024 Spotlight Poster Paper Code Dataset Overview MmCows is a large-scale multimodal dataset for behavior monitoring, health management, and dietary management of dairy cattle. The dataset consists of data from 16 dairy cows collected during a 14-day real-world deployment, divided into two modality groups. The primary group includes 3D UWB location, cows’ neck IMMU acceleration, air pressure, cows’ CBT, ankle acceleration, multi-view RGB images, indoor THI, outdoor weather, and milk yield. The secondary group contains measured UWB distances, cows’ head direction, lying behavior, and health records.