Research Centre for Data Analytics and Cognition La Trobe University, Bundoora, Victoria 3086, Australia.

HSI Deep Learning Workshop – ACCEPTED SESSIONS

Session DL01: Global Models for Time series Forecasting – presented by Dr kasun bandara and team cdac

Thursday 28 July 9:00 am  – 10:30 am

Session Summary: Forecasting models that are trained across sets of many time series, known as Global Forecasting Models, have shown recently promising results in forecasting competitions (CIF2016, M4, M5, and Wikipedia) and real-world applications, outperforming many state-of-the-art univariate forecasting techniques. While developing global models is increasingly gaining interest among forecasting practitioners, the opportunities in these models to improve the forecast accuracy and use for various real-world forecasting applications are far reaching. In this talk, the focus will be on the evolution of global forecasting models and to discuss some recent research I have conducted in this space.

Speaker Bio: Dr Kasun Bandara is a Forecasting Analytics Analysis at Energy Australia and an Honorary Research Fellow at  University of Melbourne, Australia. Kasun has developed forecasting architectures in collaboration with research scientists from Uber, Walmart, Facebook and has published his research work in journals, such as IEEE Transactions on Neural Networks and Learning Systems, Pattern Recognition, International Journal of Forecasting, and conferences such as PAKDD, IJCNN, and ICONIP. Kasun also has been a reviewer for leading ML conferences, such as ICML, NeurIPS, and ICLR. Kasun regularly participates in data science related competitions. This includes, IEEE CIS Energy Forecasting Competition (4th Position), Fuzz-IEEE Competition on Explainable Energy Prediction (2nd Position), M5 Forecasting Competition (Kaggle Gold Medallist), Air-Liquide Future Ready Data Challenge (4th position).

Session DL02: Convolutional Neural Networks for Computer Vision – Presented by Professor Jacek Rumiński

Thursday 28 July 11:00 am  – 12:30 pm

Session Summary: Currently, Convolutional Neural Networks (CNNs) are the fundamental backbones of modern computer vision solutions. The aim of the lecture will be to familiarize participants with the basics of CNN, features extraction with CNN, popular CNN architectures and other practical aspects of deep neural networks. The interactive exercises will be presented, explained and made available to participants in a Jupyter Notebook format. This content will allow the participants to acquire basic knowledge and skills of practical application of CNN in computer vision. As part of the workshop, the presented material will enable active participation in subsequent hands-on gesture recognition. This session will be conducted as a 90 minute lecture with interactive demonstrations using Jupyter Lab and Google Colab tools.

Speaker Biography: Professor Jacek Rumiński (Ph.D. in Computer Science, habilitation in Biocybernetics and Biomedical Engineering) is Head of Biomedical Engineering Department at Gdańsk University of Technology. He has spent about 2 years working on projects at different European institutions. Prof. Rumiński was a coordinator or an investigator in about 20 projects receiving a number of awards, including for best papers, practical innovations (7 medals and awards) and also the Andronicos G. Kantsios Award. He is the author of about 210 papers, and several patent applications and patents. Recently he was a main coordinator of the European eGlasses project focused on HCI using smartglasses. His research is focused on application of machine learning in healthcare, image processing and human-system interaction methods. Prof. Jacek Rumiński cooperates with Intel (San Diego, USA), Sima.AI (USA) and many other companies working on innovative products, research problems, etc. He is the organizer of series of International Summer School of Deep Learning (www.dl-lab.eu).

Session DL03: Gesture-based Interaction – Presented by Dr. Tomasz Kocejko

Thursday 28 July 1:30 pm  – 3:00 pm

Session Summary: In this course, participants will learn how static and dynamic gestures can be used for human-machine interaction. For example, the detection and interpretation of static gestures can be used to convert sign language to text, while dynamic gestures can help navigate the computer interface in a septic hospital environment. The concept of gesture-based interaction relies on three principles: detection, tracking and recognition. Computer vision technology along with deep learning algorithms can help to understand the underlying pattern. Participants will learn how artificial intelligence can accelerate the building of hand gesture-based interfaces and how to implement pretrained established deep learning models (such as VGG, ResNet50, Inception) for gesture recognition tasks. The practical application of transfer learning will be demonstrated with a few examples. Participants will learn how such models can be utilized to initialize their own model of human-machine interaction with a gesture-based interface.

Speaker Biography: Dr. Tomasz Kocejko received the M.Sc. degree in electronics and the Ph.D. degree in medical informatics at the Gdańsk University of Technology. He is an assistant professor at Faculty of Electronics, Telecommunication and Informatics, Gdańsk University of Technology. His research focuses mainly on image processing, human-computer interfaces and interaction, machine learning and biomedical engineering. He is involved into active aging programs organized by the faculty and in the situation of disabled students, acting as a dean’s representative in this matter. Speaker, consultant and promoter of artificial intelligence in healthcare and in ambient assisted living. His passion for HCI has allowed him to develop all kinds of assistive technologies, including an eye tracking interface for people with Amyotrophic Lateral Sclerosis (ALS), an interactive cube for kids with cerebral palsy, an EMG-based extension for Scratch, or a prototype of a gaze-controlled arm prosthesis.

Session DL04: Women in STEM: Opportunities and Challenges – convened by Dr. achini adikari

Thursday 28 July 3:30 pm  – 5:00 pm

Session Summary: The diversification of Science, Technology, Engineering and Mathematics (STEM) largely focuses on gender segregation as women continue to be underrepresented. Despite the nuances of modern society and increased efforts in improving gender diversity in STEM, women often encounter challenges due to entrenched biases, misbeliefs, stereotyping and families. Although there has been a steady rise in women in STEM and improved diversification in recent years, the underrepresentation of women continues to be prominent in technology-intensive fields such as engineering, mathematics, computer science and physics. Past research shows even though women occupy close to 50% of the overall employment, in most countries, women make up less than 30% of employment in research and experimental development. Due to this gender gap in STEM, women often face challenges in choosing a career path, career progression and assuming leadership positions than men. The existing gender gap and challenges impact the career choices of women anticipating a profession in STEM. To address these challenges and improve the diversity in STEM, many organizations have currently undertaken practices to encourage and facilitate the involvement of women in STEM. Research studies show that a significant factor in overcoming barriers faced by women in STEM is through networking and mentoring by which the challenges and opportunities are discussed with a wider community.

The forum Women in STEM: Opportunities and Challenges will identify and discuss several key challenges faced by women at different stages of their careers in order to motivate fellow researchers and industry practitioners in STEM.

Speaker Biography: Dr Achini Adikari is a Postdoctoral Research Fellow in the Centre for Data Analytics and Cognition (CDAC), La Trobe University, Australia. Her research interests include Emotion modelling, Social media analysis, Natural Language Processing and Self-structuring AI, particularly related to digital healthcare applications. She is currently involved in a healthcare research grant focused on developing AI solutions for people with communication disabilities. Prior to her work as a postdoctoral research fellow, she worked as a Technology Lead in several projects related to emotion modelling using conversation and social media data. Achini graduated from the University of Moratuwa, Sri Lanka in 2015, with first-class honours in Information technology and completed her PhD at La Trobe University, Australia. 

Session DL05: Hyperdimensional Computing and Neuromorphic AI – by team cdac 

Thursday 28 July 5:00 pm – 6:30 pm

Session Summary: Contemporary efforts to build an Artificial General Intelligence (AGI) have had limited success due to fundamental limitations in representation and computation. Neither the classical symbolic AI nor its recent successor, deep learning, have been able to realize an AGI. A cognitive approach that has so far proved effective is a hybrid between the two, symbolic AI for representation and machine learning for computation. Hyperdimensional Computing (HC) is effective at representing complex symbolic data structures in high-dimensional embedding spaces. Neuromorphic Computing (NC) mimics the brain computational operations by using spiking neural networks (SNN) while also addressing the von Neumann bottleneck. In this session, participants will learn thetheory and practice of HC and NC, including a hands-on session for building SNNs for practical applications. Building SNN models using the Nengo framework will also be demonstrated. 

Speaker Biography: Sachin Kahawala and Dilantha Haputhanthri are members in the Centre for Data Analytics and Cognition (CDAC), La Trobe University, Australia, working on the development of new algorithms and systems in Neuromorphic Computing and Hyperdimensional Computing.