Keynotes

Edge AI for Wildlife Conservation

Abstract:

Biodiversity is declining globally at unprecedented rates. We need to monitor species in real time and in greater detail to quickly understand which conservation efforts are most effective and take corrective action. Current ecological monitoring systems generate data far faster than researchers can analyze it, making scaling up impossible without automated data processing. However, ecological data collected in the field presents a number of challenges that current methods, like deep learning, are not designed to tackle. Biodiversity data is correlated in time and space, resulting in overfitting and poor generalization to new sensor deployments. Environmental monitoring sensors have limited intelligence, resulting in objects of interest that are often too close/far, blurry, or in clutter. Further, the distribution of species is long-tailed, which results in highly-imbalanced datasets. Finally, there exist significant bandwidth bottlenecks from the field to the cloud that significantly slow down processing times. Human-AI systems that incorporate edge-based processing are needed for time-sensitive challenges like mitigating human-wildlife conflict, reducing poaching, and monitoring escapement to ensure sustainable fishing.

Speaker: Sara Beery, Caltech

Sara Beery is a senior PhD candidate in Computing and Mathematical Sciences at Caltech. Her research focuses on computer vision for conservation and sustainability challenges. Sara was awarded both PIMCO Data Science Fellowship and the Amazon AI4Science Fellowship, which recognise her remarkable impact on machine learning, data science and their application. She has been actively engaged in research communities as invited speakers and panellists in prestigious venues including CVPR, NeurIPS, ICML, ICCV, and WACV.

Efficient Neural Architecture Search

Abstract:

In this talk, I will introduce our recent work on neural architecture search. To begin with, I will briefly introduce the challenges of neural architecture search. I will summarize the existing approaches and their corresponding limitations. After that, I will introduce our proposed approach to modularize the large search space of NAS into blocks to ensure that the potential candidate architectures are fully trained. This reduces the representation shift caused by the shared parameters and leads to the correct rating of the candidates. Then, we will talk about dynamic slimmable network, which aims to achieve good hardware-efficiency via dynamically adjusting filter numbers of networks at test time with respect to different inputs, while keeping filters stored statically and contiguously in hardware to prevent the extra burden. Finally, I will introduce an unsupervised NAS method that addresses the problem of in- accurate architecture rating caused by large weight-sharing space and biased supervision in previous methods.

Speaker: Prof. Xiaojun Chang, University of Technology Sydney

Dr Xiaojun Chang is a Professor in Australian Artificial Intelligence Insitute, Faculty of Engineering and Information Technology, University of Technology Sydney. He is also an Honorary Professor in the School of Computing Technologies, RMIT University, Australia. He has focused his research on exploring multiple signals (visual, acoustic, textual) for automatic content analysis in unconstrained or surveillance videos. Xiaojun’s team has won multiple prizes from international grand challenges which hosted competitive teams from MIT, University of Maryland, Facebook AI Research (FAIR) and Baidu VIS, and aim to advance visual understanding using deep learning.

Knowledge Distillation and Model Quantisation

Abstract:

We consider transferring the structure information from large networks to compact ones for dense prediction tasks in computer vision. Previous knowledge distillation strategies used for dense prediction tasks often directly borrow the distillation scheme for image classification and perform knowledge distillation for each pixel separately, leading to sub-optimal performance. Here we propose to distill structured knowledge from large networks to compact networks, taking into account the fact that dense prediction is a structured prediction problem. Specifically, we study two structured distillation schemes: i) pair-wise distillation that distills the pair-wise similarities by building a static graph; and ii) holistic distillation that uses adversarial training to distill holistic knowledge. The effectiveness of our knowledge distillation approaches is demonstrated by experiments on three dense prediction tasks: semantic segmentation, depth estimation and object detection.

Speaker: Prof. Chunhua Shen, Zhejiang University

Professor Chunhua Shen is currently a professor of computer science at the Zhejiang University. His research interests are computer vision and pattern recognition. He was awarded an ARC Future Fellowship in 2012. Professor Shen was listed as a top researcher in the Australian’s Lifetime Achievers Leaderboard. He has received 35,000+ citations with high-quality papers in prestigious conferences and journals including CVPR, ICCV, ECCV, ICML, NeurIPS, TPAMI, IJCV, JMLR, TOG.

AI for Koala Road Crossing Behaviour Monitoring

Abstract:

From 1997 to 2018, an average of 356 koalas entered care facilities due to vehicle collisions each year. Mitigating koala fatalities and injuries caused by vehicles is one of the most important tasks for koala conservation. This requires a deeper understanding and a better prediction of koala road crossing behaviour, which relies on koala monitoring and tracking technology. This talk introduces a pilot study to expand the pool of koala monitoring technology, applying AI-powered observation networks to investigate koala road crossing behaviours. A network of interconnected devices has been deployed in the vicinity of road crossing structures, with each device integrating a camera, a motion sensor, a wireless/mobile network module, and a solar panel. An animal movement triggers image capture, with images transferred to a server at Griffith University. Computer vision and machine learning systems are then used to process images, allowing for automatic detection and recognition of individual koalas and other animals. These data will be further analysed to provide a greater understanding of koala use of crossing structures and help to design and optimise the location of fauna mitigation measures on roads.

Speaker: A/Prof. Jun Zhou, Griffith University

Dr Jun Zhou is an associate professor in School of Information and Communication Technology in Griffith University. His research focuses on hyperspectral imaging, computer vision, pattern recognition and their applications to remote sensing, agriculture, environment, and medicine. Dr Jun is also the deputy director of ARC Hub for Driving Farming Productivity and Disease Prevention. He was a recipient of the Australian Research Council Discovery Early Career Researcher Award in 2012.