There is no excerpt because this is a protected post.
Innovation Wing Two
September 21 2023 (Thursday) 4:30-5:30pm
Can machines sense without cameras or sensors? Computer vision allows machines to “see,” but their perception capabilities based on cameras are fundamentally limited to a specific field of view and good lighting conditions – they cannot see through any occlusions or in the dark. In this talk, I will introduce Wireless AI Perception that opens a new sense for machine perception to decipher the physical world, even in absolute darkness and through walls and obstacles. To achieve this, Wireless AI leverages ambient wireless signals for sensing and turns any Wi-Fi devices from a pure communication medium into a ubiquitous all-in-one sensing platform. We will first introduce the concepts, principles, and grand challenges of Wi-Fi sensing, and then share our unique solution of Wireless AI, which has been commercialized and deployed as real-world products, such as motion sensing, sleep monitoring, fall detection, indoor tracking, just to name a few. We foresee that Wi-Fi Sensing will enter billions of devices and millions of homes, and today is just the beginning of this revolution.
September 19, 2023 (Tuesday) 4:30-5:30pm
Driven by the remarkable success of artificial intelligence (AI) and edge computing, the deployment of well-trained private AI models on third-party edge devices for mission-critical applications has become increasingly prevalent. Safeguarding these private models on untrusted devices, while simultaneously speeding up model serving (i.e., inference) through accelerators like GPUs, has escalated in urgency.
We introduce SOTER, a new AI serving system that, for the first time, achieves both high security and high performance. Harnessing the associativity property of AI operators, SOTER presents an innovative approach—transforming computationally expensive AI operators into parameter-morphed equivalents for secure execution on untrusted but fast GPUs, and losslessly restoring inference results within trusted execution environments (TEEs) in CPUs. Experimental results on six prevalent AI models in the three most popular categories show that, even with stronger model protection, SOTER achieves comparable performance with baselines while retaining the same high accuracy as insecure AI model inference.
September 14, 2023 (Thursday) 4:30-5:30pm
On-demand ride services or ride-sourcing services, offered by transportation network companies like Uber, Lyft and Didi, have been experiencing fast development and steadily reshaping the way people travel in the past decade. Various mathematical models and optimization algorithms, including reinforcement learning approaches, have been developed in the literature to help ride-sourcing platforms design better operational strategies to achieve higher operational efficiency. However, due to cost and reliability issues (implementing an immature algorithm for real operations may result in system turbulence), it is commonly infeasible to validate these models and train/test these optimization algorithms within real-world ride sourcing platforms. Acting as a useful test bed, a simulation platform for ride-sourcing systems will thus be very important for both researchers and industrial practitioners to conduct algorithm training/testing or model validation through trails and errors. While previous studies have established a variety of simulators for their own tasks, it lacks a fair and public platform for comparing the models/algorithms proposed by different researchers. In addition, the existing simulators still face many challenges, ranging from their closeness to real environments of ride-sourcing systems, to the completeness of different tasks they can implement. To address the challenges, we propose a novel multi-functional and open-sourced simulation platform for ride-sourcing systems, which can simulate the behaviors and movements of various agents (including drivers and passengers) on a real transportation network. It provides a few accessible portals for users to train and test various optimization algorithms, especially reinforcement learning algorithms, for a variety of tasks, including on-demand matching, idle vehicle repositioning, and dynamic pricing. Evaluated by experiments based on real-world datasets, the simulator is demonstrated to be an efficient and effective test bed for various tasks related to on-demand ride service operations.
September 12, 2023 (Tuesday) 4:30-5:30pm
The rapid development of communication technologies enables real-time information exchange between vehicles, thus being virtually connected. However, the full connected vehicle (CV) deployment will take a long time and may never be achievable, due to privacy, security, and willingness. Knowledge of the CV penetration rate is thus crucial for realizing numerous beneficial applications during the prolonged transition period. Although several novel models have been proposed for CV penetration rate estimation, they are solely point estimators. Direct use of these point estimators without considering their variability can lead to biased models or suboptimal solutions. To bridge this research gap, this study proposes a series of analytical models to accurately estimate the variability of CV penetration rate. Comprehensive VISSIM simulations, real-world applications, and a simple CV-based adaptive signal control scheme demonstrate the readiness of the models for use in real-world situations and the potential of the models to improve system optimizations.
Aerogels are lightweight materials with extensive microscale pores, which could be used in thermal insulation, energy devices, aerospace structures, as well as emerging technologies of flexible electronics. However, traditional aerogels based on ceramics tend to be brittle, which limits their performance in load-bearing structures. Due to restrictions posed by their building blocks, recently developed classes of polymeric aerogels can only achieve high mechanical strength by sacrificing their structural porosity or lightweight characteristics.
August 28, 2023 (Monday) 2-3pm
Thermodynamics is a science concerning the state of a system, whether it is stable, metastable, or unstable. Its derivatives to natural variables give fundamental physico-chemical properties of the system. It is historically divided into four categories: equilibrium thermodynamics by Gibbs, statistical thermodynamics by Gibbs and Landau, irreversible thermodynamics by Onsager and Prigogine, and quantum mechanics. The development of density function theory (DFT) enabled the quantitative prediction of properties of the ground state of a system from quantum mechanics. Their integration into predictive theories will be discussed in this presentation along with future perspectives. It will be shown that the zentropy theory combines the bottom-up DFT predictions with the revised top-down statistical thermodynamics, while the theory of cross phenomena keeps the entropy production due to irreversible processes in the combine law of thermodynamics to revise the Onsager flux equations. The zentropy theory is capable of quantitatively predicting free energy landscape, singularity and emergent divergences of properties at critical point free of parameters, while the theory of cross phenomena can predict the coefficients of internal processes between non-conjugate variables.
High strength lightweight steel can effectively save energy and cut emissions of vehicles, although it is more brittle than traditional low strength steel, which causes safety issue. This project tailors the coating structure of high strength lightweight steel to make this steel more ductile and ensure the safety.
Professor Komura’s research focuses on creating diverse human movements for computer animation, games, and virtual environments. His research team designed a new type of neural network called the Periodic Autoencoder that can identify repeating patterns in large sets of motion data without any additional information. This allows us to create unique movements in various styles, such as dance motions synchronized to music or dribbling movements in soccer. Additionally, the system can help find similar motions in large database, produce natural movements between a small number of key frame poses and estimate human movements even when the body is partially obscured in videos.
“HKU PERfECT” is the first wearable platform that can simultaneously acquire the following three merits.
1. Highly sensitive: By combing electrochemical technology with microelectronic technology, the highest sensitivity is reached.
2. Smallest and lightest: By using the smallest possible electronic units and marrying emerging stretchable bioelectronic technologies, coin-sized and light (0.5 grams) PERfECT wearables have been used for diagnosis and treatment of various diseases, and rehabilitation.
3. Energy efficient: By using interdisciplinary research strategies spanning analytical chemistry, low-power microelectronics, and low-power wireless communication, PERfECT achieves the highest accuracy with the lowest power consumption, ideal for long-term using.