Kunhe Yang
University of California, Berkeley
kunheyang@berkeley.edu
Bio
Kang Yang is a Postdoctoral Scholar in Electrical and Computer Engineering at UCLA, advised by Professor Mani Srivastava, where he focuses on AI-driven, robust, and efficient decision-making in humancyberphysical systems. He received his Ph.D. in Electrical Engineering and Computer Science from the University of California, Merced, advised by Professor Wan Du. During his doctoral studies, he was funded by the U.S. Department of Commerce Economic Development Administration to develop reliable and energy-efficient Long Range (LoRa) networks for smart irrigation and groundwater recharge in orchards.
Areas of Research
- Machine Learning
Distortion of AI Alignment: Does Preference Optimization Optimize for Preferences?
My research addresses the challenge of developing intelligent infrastructures that remain reliable, energy-efficient, and adaptable under severe resource constraints through the integration of networking, sensing, and optimization as mutually reinforcing pillars. Real-world applications such as expansive agricultural deployments or global energy grids must cope with unreliable communication links, strict power budgets, and constantly changing physical environments. Networking provides the foundational backbone by enabling dependable data transmission across long distances despite variable channel quality, and I strengthen this layer with advanced error-correction codes and adaptive link protocols that sustain robustness in low-power networks. Sensing supplies essential visibility into the environment, where I design propagation models and localization algorithms that account for obstacles and variability, allowing precise signal-strength prediction and accurate positioning in obstructed or evolving settings. I further extend this capability through generative synthesis of radio-frequency data, which reduces the cost and energy required for large-scale measurements. Optimization completes the cycle by transforming networked sensing data into actionable decisions for energy and carbon management; using time-series foundation models for global carbon-intensity forecasting and imputation, I enable carbon-aware scheduling and resource allocation without dependence on proprietary grid information. Together these three aspects form a complete stack in which networking secures reliable connectivity, sensing delivers high-fidelity awareness, and optimization drives sustainable, intelligent operation. By combining artificial intelligence, domain-specific wireless physics, and large-scale predictive modeling, my work addresses the core problem of designing infrastructures that achieve long-term viability with minimal environmental impact and maximum resilience.