Invited Talks

Extreme Edge Intelligence: Moving Artificial Intelligence Closer to Sensors

Edge intelligence holds immense importance in enabling real-time decision-making and reducing latency by processing data closer to sensing. This approach enhances efficiency, privacy, and security by minimizing the need for constant data transmission to centralized servers. However, the tradeoffs lie in the limited computational power and memory at the edge, potentially restricting the complexity of algorithms and the scale of data processed compared to the vast resources available in the cloud.

This talk presents the fusion of AI with sensors showcasing the integration of AI at an embryonic stage with sensors, leading to a profound impact on sensor functionality and application. The talk will illustrate this transformative integration of AI at the sensor level using three compelling case studies and illustrating the advantages of edge intelligence:

The first case study, within the realm of olfaction, the convergence of bio-inspired classification algorithms with gas/smell sensors presents a breakthrough in realizing an electronic nose. The seamless embedding of AI augments sensor capabilities, enabling training, recognition and classification of diverse odors, mirroring the sophistication of human olfactory system.

The second case study delves into the domain of vision sensors, illustrating how minor modification of pixel structure can accommodate convolutional layers for deep learning. This adaptation amplifies the capacity of vision sensors, empowering them to interpret and process visual data with the sophistication characteristic of deep learning models, enabling efficient integration of deep learning at the sensor level.

Last but not least, the integration of AI with wearable sensors in order to classify signals derived from wearable devices, enabling the discrimination of various gestures, biomechanics, and vital signs of individuals. By embedding AI within these sensors, a new paradigm in healthcare monitoring and patient care unfolds, revolutionizing the real-time analysis and interpretation of physiological data.

In a nutshell, this talk unravels the paradigm shift where AI converges with sensors at the edge, augmenting their capabilities and reshaping possibilities across olfaction, vision, and wearable sensor domains. The talk not only unveils technical advancements in the area but also underscores the transformative potential of extreme edge intelligence in diverse fields.

Talk title to be announced

Talk abstract to be announced.

Hardware-Software Co-Design for Robotic Perception System

Robotic perception systems are fundamentally constrained by Size, Weight, and Power (SWaP). This work presents a multi-level hardware-software co-design methodology to solve three key resulting challenges. To resolve the "impossible trinity" between accuracy, speed, and generalization, a self-adaptive stereo system co-optimizes its algorithm and a dedicated accelerator. To address the "insect paradox" of diverse computing tasks, a flexible instruction-driven architecture for SLAM unifies and accelerates the operators of the frontend and backend. To align component tuning with application-level goals, an automatic Design Space Exploration (DSE) framework optimizes the entire stack for an application-specific metric, further enhancing performance. This integrated methodology provides a systematic path to designing highly efficient perception systems for real-world, SWaP-constrained robotics. 

AI at Sea: How Unmanned Surface Vessels Are Driving Maritime Autonomy

While public attention often focuses on terrestrial or aerial AI-driven systems, the application of AI in oceanic environments is making significant strides, with Unmanned Surface Vessels (USVs) emerging as one of the most innovative and practical examples. 

Among the global pioneers in this field is OceanAlpha, a Chinese company that began developing civilian unmanned surface vessels as early as 15 years ago. Today, OceanAlpha stands as China's largest USV enterprise and has developed multiple global or first-in-China USVs, each designed for diverse specialized missions.

In this speech, the speaker will unveil OceanAlpha's core technological strengths in USV development, including autonomous navigation, obstacle avoidance, mission planning, and real-time environmental perception, which involve the use of AI algorithms.

USV applications in marine survey, offshore energy, and maritime safety will also be examined, highlighting how AI-powered USVs can significantly reduce human risk and cost.