Title: Towards Responsible AI-powered Products
 

Dr Morteza Saberi is currently a Senior Lecturer (Associate Professor) at the School of Computer Science, UTS, Sydney. He has an outstanding research record and significant capabilities in the area of business intelligence, data mining and applied machine learning. He has a broad interest in the emerging forms of societal-scale human-computer systems that currently govern and facilitate knowledge exchange among individuals and organizations. 

Existing research has primarily focused on enhancing the performance of these systems through the development of machine learning models. However, notable deficiencies and failures persist. These shortcomings involve the absence of a robust feedback loop connecting machine learning outputs with human decisions and a lack of attention to user reactions towards machine learning-based decisions. Dr Saberi’s research endeavors strive to address these limitations by constructing human-centric machine learning models that enhance the decision-making processes through collaborative partnerships between algorithms and system users. Rather than simply replacing humans with algorithms, his approach aims to augment human capabilities within the design process of machine learning models. To achieve this goal, Dr Saberi leverages interdisciplinary techniques encompassing optimization, large-scale data mining, and traditional machine learning. This approach has yielded conceptual innovations and technical advancements. The common thread underlying his research interests is the pursuit of enhancing the functionality of these systems for the betterment of society. He actively participates in various related projects and activities, including initiatives such as Smart Farm and the Knowledge-Driven Solution Support System for intrusion detection.

Contact: morteza.saberi@uts.edu.au

Plants are complex 3D objects, continuously growing, with self-occlusion and self-similarities. As such they constitute challenges for computer vision. These challenges meet the need for automated phenotyping for understanding plant stress response, climate resilient plant selection, biocontrol, ... In this talk we will provide a panorama of computer vision and machine learning problems that we recently addressed while developing a network of RGB-Depth cameras connected to minicomputers on our plant phenotyping platform. This includes the learning of spatio-temporal patterns during plant growth, the transferability of the learning from one crop to another or an environment to another, the self-occlusion of leaves, ... We discuss the remaining open challenges and opportunities for the future of this low-cost imaging system available with open-source codes and hardware setup.

1. Cordier, M., Torres, C., Rasti, P., & Rousseau, D. (2023). On the Use of Circadian Cycles to Monitor Individual Young Plants. Remote Sensing, 15(11), 2704.
2. Garbouge, H., Rasti, P., & Rousseau, D. (2021). Enhancing the Tracking of Seedling Growth Using RGB-Depth Fusion and Deep Learning. Sensors, 21(24), 8425.
3. Samiei, S., Rasti, P., Ly Vu, J., Buitink, J., & Rousseau, D. (2020). Deep learning-based detection of seedling development. Plant Methods, 16(1), 1-11.

Prof David Rousseau heads the Bioimaging group - ImHorPhen - at Université d'Angers IRHS-INRAe, France. He develops, in an interdisciplinary spirit, imaging and machine learning-based computer vision solutions for plant phenotyping applied to seed, seedling and horticultural crops. He cares about the teachability of his work and makes it available via video tutorials at:
https://www.youtube.com/channel/UCsd9Dt6N7O-fydynsWEfkww