### Abstract: Dynamic Convolutional Neural Networks - Academic Salon at Tsinghua University Institute for Artificial Intelligence (AIR) **Date:** April 1, 2021 (Thursday) **Time:** 2:00 PM - 3:30 PM **Location:** Turing Auditorium, 12th Floor, Building C, Tsinghua Science Park, Tsinghua University, Beijing, China #### Introduction To enhance academic exchange, promote the timely sharing of the latest industry information, and expand its influence in both academic and industrial circles, the Tsinghua University Institute for Artificial Intelligence (AIR) will regularly host various types of academic salons starting from April 1, 2021. The first salon in this series is scheduled for the afternoon of April 1, featuring Dr. Gao Huang, an assistant professor from the Department of Automation at Tsinghua University. Dr. Huang will present on the topic of dynamic convolutional neural networks (CNNs). #### Key Events - **Event Type:** Academic Salon - **Topic:** Dynamic Convolutional Neural Networks - **Speaker:** Dr. Gao Huang - **Host:** Tsinghua University Institute for Artificial Intelligence (AIR) - **Objective:** To discuss the advancements in designing efficient deep neural networks for edge computing platforms, the limitations of current methods, and future research directions. #### Background Deep convolutional networks have achieved significant success in various computer vision tasks, including image recognition, object detection, and scene understanding. However, the deployment of these models on edge computing platforms, such as smartphones, wearable devices, and IoT (Internet of Things) devices, is hindered by the limited computational power, storage space, and energy supply of these devices. This has led to a critical research direction focused on developing low-power, low-latency, and lightweight deep learning models tailored for edge computing environments. #### Report Summary In his presentation, Dr. Gao Huang will delve into the recent progress in designing dynamic and adaptive deep neural networks that are efficient and suitable for edge computing platforms. Dynamic models can adapt their computational resources based on the input data, thereby optimizing performance and reducing energy consumption. Dr. Huang will discuss the following key points: - **Advancements in Dynamic Models:** He will highlight recent research that has contributed to the development of dynamic CNNs, which can adjust their complexity and resource usage in real-time. - **Efficiency and Adaptability:** The talk will explore how these dynamic models can significantly reduce computational requirements and improve energy efficiency, making them ideal for deployment on edge devices. - **Limitations of Current Methods:** Dr. Huang will also address the limitations and challenges faced by existing dynamic CNNs, such as the trade-off between model complexity and accuracy, and the difficulty in maintaining real-time performance. - **Future Research Directions:** The presentation will conclude with a discussion on the potential future developments in this field, including new algorithms and hardware innovations that could further enhance the efficiency and adaptability of dynamic CNNs. #### Speaker Bio Dr. Gao Huang is an assistant professor and doctoral supervisor at the Department of Automation, Tsinghua University. He received his Ph.D. from Tsinghua University in 2015 and subsequently conducted postdoctoral research at the Department of Computer Science, Cornell University, from 2015 to 2018. Dr. Huang's primary research interests are in deep learning and computer vision, and he is particularly known for proposing the DenseNet model, a widely recognized and influential convolutional network architecture. He has published over 50 academic papers in top international conferences such as NeurIPS, ICML, and CVPR, as well as in multiple IEEE journals. His research has been cited more than 20,000 times. Dr. Huang has received several prestigious awards, including the Best Paper Award at CVPR, the DAMO Academy Qingcheng Award, the SAIL Pioneer Award at the World Artificial Intelligence Conference, the National Top 100 Most Influential International Academic Papers, the First Prize in Natural Science from the Chinese Association for Artificial Intelligence, and funding from the National Outstanding Youth Science Fund. #### Participation Information To attend the salon, interested participants are required to scan a QR code to make a reservation. Upon successful reservation, attendees should bring the necessary identification documents to the event. The final decision regarding attendance and any changes to the event will be made by the Tsinghua University Institute for Artificial Intelligence (AIR). #### Conclusion The Tsinghua University Institute for Artificial Intelligence (AIR) is committed to fostering a vibrant academic and industry ecosystem through regular academic salons. The first salon, focusing on dynamic convolutional neural networks, promises to be a valuable platform for researchers and practitioners to share insights and explore the future of efficient deep learning models for edge computing. Dr. Gao Huang's expertise and contributions to the field make him an ideal speaker to kick off this series, and his presentation is expected to provide a comprehensive overview of the current state and future prospects of dynamic CNNs.