Yongzhi Huang (黄勇志)

Brief: Don’t want to trek to a clinic or rent bulky tympanometry gear to check your ears? Want your own ear-canal digital twin for long-term ear health? Here’s the rub: every canal is unique, and without a sealed, expensive probe, reconstructing the acoustic path is hard. EarCSI fixes that. With everyday earbuds and a short audio ping, our physics-guided sensing-and-computing pipeline rebuilds your canal’s acoustic channel. It tracks subtle tympanic reflections in real time—no clinic, no sealed probe.

Brief: By making a simple alteration to an electric toothbrush, it can be transformed into a smart toothbrush. This innovative toothbrush can detect your brushing habits and prompt you to correct any bad habits.

Brief: Up until now, diagnosing kidney disease at home has been a challenge. Traditionally, urine testing requires bulky equipment found in hospitals. However, our team has created a system that can accurately and quickly identify impurities like uric acid and protein in the urine, making it a more convenient and efficient option.

Brief: Want your LoRa IoT network to gain environmental sensing without extra hardware or protocol changes? SenLoRa mines confidence symbols from uplink LoRaWAN frames to boost both decoding reliability and per-packet sensing. When traffic is sparse, an entropy-driven trigger injects just a few incentive packets—keeping batteries happy.

Brief: Factory workers often perform highly complex actions during their tasks, frequently interspersed with incidental movements such as wiping sweat or checking a watch. Traditional approaches to address this challenge rely on large-scale generalization networks, which result in significant computational overhead. By decomposing these actions, we propose a self-supervised, adaptive, low-computation clustering approach to tackle this issue, making action recognition in factory operations more feasible for real-world deployment.

Brief: Detecting biological changes in the Industrial Internet of Things can be a challenging task. However, a solution we are exploring for the beverage industry to monitor microbial changes in liquids, including spoilage.

Brief: Wineries sometimes introduce other bacteria in unique ways, like stomping on them, to create special flavors. However, this complex microflora can lead to winemaking failures. Testing sealed wine for microbial changes carries a risk of contamination and is also time-consuming. To address this, we developed a real-time system that monitors microbial changes. Our system eliminates the need for microbial culture methods, making microorganism measurement faster and safer.

Brief: Did you know that it's possible to obtain humidity information without a hygrometer? Yes! Our system utilizes wireless signals to gather this data, making it incredibly convenient.

Brief: Upgrading the infrastructure and implementing fire prediction systems in older high-rise residential buildings can be a challenging task. Our research has revealed that flames are a type of plasma, which can obstruct signals and produce erratic interference with signal channels. Utilizing this knowledge, we have developed a novel fire detection system.

Brief: In areas where water infrastructure is lacking, such as Africa and Southeast Asia, it may be difficult to detect when water quality may be subpar. To address this issue, we have developed a system that utilizes mobile phone sensors to identify the quality of water. With this technology, it is possible to quickly and accurately determine whether the liquid is safe for consumption.

Brief: Have you ever wondered if there is a way to type or chat with text in virtual reality that is similar to using a keyboard in real life? With our technology, we utilize a watch capable of detecting the vibrations generated by finger when typing, enabling replication of keystroke activity within the virtual world.

Brief: Our goal was to create technology that assists visually impaired individuals in locating objects. We made adjustments to the design to eliminate any stigma and negative connotations associated with using the device due to their disability.

Brief: Federated learning breaks data silos by coordinating local training and aggregation from a central server without exposing raw data or local models. In practice, client data—especially in IoT and finance—often contain many rare classes, exhibiting long-tail characteristics and client-specific distributions, which introduces heterogeneity. Training directly under these conditions can collapse, because long-tail heterogeneity sculpts a bumpy loss landscape where different data types pull against each other. We propose FedWMSAM. Intuitively, it works like riding a snowboard: when the surface sticks up, the board “lifts and glides” over it. Moreover, the roughness varies at different elevations. Therefore, we adopt an adaptive scheme: when the terrain is relatively flat in the early stage, everyone moves quickly in the same direction; as bumps accumulate, we gradually “widen the board” to span uneven regions, guiding the model toward a better global solution.

Brief: Real-world federated data are typically long-tailed and highly imbalanced, leading to unstable convergence and minority-class collapse. We diagnose the root causes and introduce FedWCM, a distribution-aware weighting scheme that markedly improves accuracy and convergence stability under extreme tail distributions.

Brief: The sharing of data resources has long been a significant challenge for enterprises, leading to the creation of numerous data silos. However, homomorphic encryption technology offers an ideal solution for ensuring data privacy. This article provides a comprehensive review of this field.

Brief: Existing keystroke detection technologies face significant challenges in real-world deployment, especially in complex environments like offices. Factors such as overlapping keystroke sounds, variations in keyboard positioning, and differences in typists can significantly impact detection accuracy, rendering traditional techniques ineffective. "Chameleon" addresses this issue by introducing an adaptive system that dynamically adjusts recognition features based on the environment, enabling detection technologies to achieve robust, environment-adaptive performance.

Brief: Medical data is highly sensitive and classified as private information; however, malicious actors may exploit a physician's privileges to infringe on patient privacy. To address this issue, we designed a novel encryption scheme that requires user key authorization before accessing medical data, effectively preventing unauthorized access through forgery or other malicious means.

Brief: Spy cameras? Sneaky, but no match for us! Forget those bulky, specialized gadgets—just grab your smartphone, plug in a pair of headphones, and voilà! Our solution turns your everyday gear into a powerful spy camera detector, leaving no hiding place for those prying eyes!

Brief: A side-channel attack is a hacking technique that exploits physical information leakage. Using this method, we can eavesdrop on the audio content transmitted through your headphones.

Brief: Are you never concerned about the safety of your headphones? Surprisingly, someone can easily eavesdrop on what you are listening to with just a few dollars of equipment.