3.1 Opening speech, chaired by Yuan Yue
3.2 First session, chaired by Yuan Yue
| Name | Time | Title |
|---|---|---|
| Elijah Zolduoarrati | 09:10 | Understanding diversity-driven disparities in software development: an empirical analysis of human aspects, contribution patterns, and code quality across various geographical regions. |
| Junlei Hong | 09:25 | Exploring head-worn LCD light actuators for perception modulation |
| Kushani Tharushika Perera | 09:40 | Ethical Implications of Pervasive Augmented Reality |
| Emma Collins | 09:55 | Exploring education technology for child health teaching |
| Cristhian Delgado | 10:10 | Satellite-based AI Emulators for Efficient Monitoring of Agricultural Methane Emissions |
| Shazia Gul | 10:25 | Localization and Tracking for Augmented Reality in large scale dynamic environments |
3.3 Second session, chaired by Abira Sengupta
| Name | Time | Title |
|---|---|---|
| Hayden McAlister | 10:55 | Prototype Learning in the Hopfield Network |
| Xiaojie Yu | 11:10 | White-box deep neural network based on rate reduction |
| Ying Pang | 11:25 | A simple noisy label detection approach for deep neural networks |
| Elliot Munro | 11:40 | Sample Efficient Reinforcement Learning |
| Wang Xianheng | 11:55 | Deep Learning in MI-EEG classifition with transformer |
| Yuan Yue | 12:10 | Finding obesity neural signature using machine learning based EEG analysis |
3.4 Thrid session, chaired by Ying Pang
| Name | Time | Title |
|---|---|---|
| Fei Gao | 13:20 | Scalable and Energy-Efficient Neural Network Accelerator with Photonic Interconnects |
| Lakmal Deshapriya | 13:35 | What are the gaps between academia and the industry in static code analysis tools, and how to overcome them? |
| Chengpeng Xia | 13:50 | STADIA: Photonic Stochastic Gradient Descent for Neural Network Accelerators |
| Hao Zhang | 14:05 | ESA: An efficient sequence alignment algorithm for biological database search on Sunway TaihuLight |
| Natalie Grattan | 14:20 | Automatically Informing the Cause of Software Defects |
| Sobhan Latifi | 14:35 | Optimization of 3d models in Engineering applications |
3.5 Forth session, chaired by Lakmal Deshapriya
| Name | Time | Title |
|---|---|---|
| Maryam Nakhoda | 15:05 | Quantifying and addressing uncertainty in the measurement of interdisciplinarity |
| Zahra Torabi | 15:20 | Using Machine Learning to Support the Development and Maintenance of Firewall Configurations |
| Garth Wales | 15:35 | End-to-end image segmentation with declarative neural networks |
| Abira Sengupta | 15:50 | Generalising Axelrod's Metanorms Game through the use of explicit domain-specific norms |
| Vaughan Kitchen | 16:05 | Dynamic Accumulator Selection for Disjunctive Query Evaluation |
3.6 Closing session, chaired by Hayden McAlister
| Name | Title | Abstract |
|---|---|---|
| Elijah Zolduoarrati | Understanding diversity-driven disparities in software development: an empirical analysis of human aspects, contribution patterns, and code quality across various geographical regions. | In software development, diversity across social, behavioural, and cultural dimensions significantly shapes collaborative dynamics. This research explores the challenge of integrating individuals from diverse backgrounds into cohesive teams, enabling their effective participation in tasks and decision-making—an obstacle demanding industry attention. Current research often dissects diversity into isolated elements, necessitating a more comprehensive perspective. Our work focusses on predicting contribution disparities in software engineering open communities. Commencing with a thorough tertiary analysis, we delve into nuanced human aspects within software development contexts. We investigate the alignment of various contribution attributes with diverse census indicators, exploring beyond quantitative measurements, including, contribution readability, polarity, as well as code snippet quality from reliability, readability, performance, and security standpoints. Our findings to date challenge assumptions, illuminating intricate disparities among contributors. |
| Junlei Hong | Exploring head-worn LCD light actuators for perception modulation | Computational Glasses are rapidly developed to enhance human visual perception abilities. Complex imaging technology for glasses is deeply studied to improve its performance. Much less explored is directly modulated perception by simple portable glasses, like head-worn LCD light actuators, which we address here. We first explore the prototype based on three potential applied daily life scenarios. We then ran a within-group user study to test the prototype's performance in an experimental environment given that there is little research implementing user studies to understand the potential of this technology. |
| Kushani Tharushika Perera | Ethical Implications of Pervasive Augmented Reality | Augmented Reality emerges as the next step of mobile and wearable computing. Even more, a continuously maturing technology is eventually turning AR headsets into an everyday, casual, always-on commodity. However, this omnipresent and continuous augmentation of our environment, generally referred to as Pervasive AR, might also introduce changes in behaviour and social interaction. We designed a PAR technology probe and exposed 40 participants in pairs of two to an environment emulating a near-future scenario displaying virtually augmented posters. We explored behavioural changes that arise from using Pervasive AR in such a public setting and in particular highlight the differences to traditional notions of public displays and how this will affect our interactions with one another, trust in information, and the attitude towards Pervasive AR technology. We conclude our exploration with recommendations for designing future Pervasive AR systems. |
| Emma Collins | Exploring education technology for child health teaching | Using technology in the delivery of healthcare is essential to patient care. Therefore, as health care educators we need to ensure that the learners that we interact with as undergrads, are technologically capable. This mindset led me into a research interest in digital health and educational technologies. Whilst working as a lecturer at Otago Polytechnic I used holograms and the Microsoft HoloLens to engage the nursing students in learning clinical reasoning. This led to numerous research outputs. Since moving to the University of Otago to work in child health, I have continued growing my passion for digital health and educational technologies. Hence my PhD topic of exploring extended realities for teaching child health. For child health, the gold standard is clinical experience with real children. Due to the pandemic and other factors, this is very difficult to facilitate, and students are left with lectures and textbooks. I am hypothesing that a child health teaching scenario using some form of extended reality, will bridge the gap between didactic teaching and real children. This presentation will explore partly how I got here but I will also demonstrate how a hologram can be useful in clinical teaching. |
| Cristhian Delgado | Satellite-based AI Emulators for Efficient Monitoring of Agricultural Methane Emissions | The latest New Zealand Greenhouse Gas (GHG) inventory submitted to the UNFCCC highlights that half of the nation's GHG emissions are attributed to Agricultural methane (CH4), primarily from livestock production accounting for 77.5%. Given agriculture's economic and emissions significance, New Zealand is actively developing technologies to curtail CH4 emissions from livestock in line with the Paris Agreement. Implementing and verifying these technologies at scale pose challenges due to diverse agroecological factors affecting emission patterns. Traditional verification methods like closed chambers and the eddy covariance technique are limited by scale and complexity |
| Shazia Gul | Localization and Tracking for Augmented Reality in large scale dynamic environments | Localization, which determines a user’s position and orientation is crucial in accurately overlaying virtual content in Augmented Reality (AR). Our research investigates localization methods suitable for AR in largescale environments and integrates them into a mobile system for evaluation. Specifically, we evaluate the performance of two state-of-the-art approaches: the Expert Sample Consensus Applied to Camera Re-Localization (ESAC), and Accelerated Coordinate Encoding (ACE), focusing on their feasibility for AR in large-scale environments. As an experimental platform, we developed a client-server framework coupled with a mobile AR application for onsite feasibility testing. Using three trained large sample datasets --- a stadium, a clocktower, and a courtyard; our approach processes a query image from the mobile AR client and provides pose computation using either ESAC or ACE. The pose data is then used in the mobile AR client to render 3D registered content. |
| Hayden McAlister | Prototype Learning in the Hopfield Network | We explore prototype formation in the Hopfield network. We formalize formation behavior under Hebbian learning and derive a stability condition for prototype states. We experimentally test the formalization using a network under standard Hebbian learning. For a more practical example, we extend the experimentation to a network using the error correcting Hebbian learning rule, and investigate the formation of multiple prototypes. We also make a link between spurious states and prototype formation. |
| Xiaojie Yu | White-box deep neural network based on rate reduction | Deep Learning, as a branch of artificial intelligence, has been applied in many fields, such as computer vision, speech recognition, and natural language processing. While deep learning is effective in many scenarios, the design principle behind deep learning hasn’t been well clarified. It works like a black box. Hence, it is necessary to develop deep learning to be more understandable and more useful for more scenarios. |
| Ying Pang | A simple noisy label detection approach for deep neural networks | In federated learning, more participants may bring more knowledge but also more risk of label noise. Label noise is a common issue in real-world applications, owing to the blankness of annotation expertise or malicious attack. A significant performance drop will be caused when the clients contain inevitable noisy labels. Most of the existing solutions are no longer applicable to FL due to the inaccessibility of local data. Our main work is to provide a solution for FL to identify noisy labels at each client without compromising privacy-preserving. The key idea is to leverage output logits of local models to predict the possibility of noisy labels and introduce the curriculum learning strategy to expand the gap of output logits between noisy samples and clean samples. |
| Elliot Munro | Sample Efficient Reinforcement Learning | Reinforcement Learning (RL) is getting stronger as a field year by year, and its applications are vast, from simulated game mastery to autonomous vehicles. However, sample efficiency still remains a large problem, we still lack the well crafted architecture designs that will allow agents to train significantly faster, both in simulation and in the real-world. Real-world learning is very desirable because it allows robots the ability (1) to adapt from the simulator to the real-world in terms of the exact environment dynamics and representation and also (2) to learn to better navigate around previously unseen obstacles. My work has been focused on RL for real-world quadruped robot locomotion through sim-to-real transfer. Over the last 5 years sim-to-real has become the go to approach for real-world performance. However none of these works display any learning in the real-world after the transfer is made. The aim of my work is to show learning in the real-world that fine tunes the simulator trained policy online. |
| Wang Xianheng | Deep Learning in MI-EEG classifition with transformer | Currently, Transformer-based models have been dominant in various fields, e.g., computer vision, natural language processing, etc, thanks to their remarkable performance. However, Transformer-based methods have not been widely explored in EEG classification. This leads to a question: Is Transformer still a good design choice in EEG classification? This presentation will conduct a comprehensive assessment and comparative analysis of Transformer-based models in conjunction with representative EEG classification models based on other popular deep learning techniques. Two common EEG classification tasks, i.e., Motor Imagery EEG and Depression EEG are chosen and used for model evaluation and analysis. This presentation aims to help researchers better design their EEG classification models. |
| Yuan Yue | Finding obesity neural signature using machine learning based EEG analysis | Obesity is a serious issue in modern society since it associates with a significantly reduced quality of life. This study aims to develop machine learning-based frameworks to understand the difference between the obese brain and non-obese brain. We will also explore how brain activities of certain brain regions, and the connectivity and interaction between them, associate with an individual’s hunger/satiety state. |
| Fei Gao | Scalable and Energy-Efficient Neural Network Accelerator with Photonic Interconnects | Matrix-vector multiplication, a crucial foundation for information processing in diverse scientific and technological domains, accounts for a significant portion of the computational overhead in contemporary signal processing and artificial intelligence algorithms. However, the prevalent dependence on matrix computations through traditional metal-based interconnects in accelerators increasingly poses considerable challenges to performance, energy efficiency, and scalability. The photonics-based accelerator, equipped with photonic interconnects and computing units, is specifically designed to expedite computations in the optical domain, particularly matrix multiplication, addressing the escalating demand for advanced computing resources and capacity. we propose a scalable accelerator for matrix-vector multiplication with photonic interconnects and a photonic computing unit named AMMP. |
| Lakmal Deshapriya | What are the gaps between academia and the industry in static code analysis tools, and how to overcome them? | Static code analysis (SCA) tools are essential for software development, aiming to reduce both the cost and time incurred during code reviews. Both industry and academia evaluate SCA tools, albeit through different methods. Notably, some SCA tools, highly regarded in the industry, remain unexamined within academic contexts. Currently, only approximately 12.7% of the SCA tools reported in the community have undergone academic scrutiny. Consequently, there is a pressing need for academia to investigate more SCA tools that are prevalent in the software engineering community. Conversely, the best tools acknowledged within the software engineering community exhibit elevated rates of false positives and false negatives. To address this challenge, various approaches, including the combining of results from existing SCA tools, can be employed. |
| Chengpeng Xia | STADIA: Photonic Stochastic Gradient Descent for Neural Network Accelerators | Deep Neural Networks (DNNs) have demonstrated great success in many fields such as image recognition and text analysis. However, the ever-increasing sizes of both DNN models and training datasets make deep leaning extremely computation- and memory-intensive. Recently, photonic computing has emerged as a promising technology for accelerating DNNs. While the design of photonic accelerators for DNN inference and forward propagation of DNN training has been widely investigated, the architectural acceleration for equally important backpropagation of DNN training has not been well studied. In this paper, we propose a novel silicon photonic-based backpropagation accelerator for high performance DNN training. Specifically, a general-purpose photonic gradient descent unit named STADIA is designed to implement the multiplication, accumulation, and subtraction operations required for computing gradients using mature optical devices including Mach-Zehnder Interferometer (MZI) and Mircoring Resonator (MRR), which can significantly reduce the training latency and improve the energy efficiency of backpropagation. To demonstrate efficient parallel computing, we propose a STADIA-based backpropagation acceleration architecture and design a dataflow by using wavelength-division multiplexing (WDM). We analyze the precision of STADIA by quantifying the precision limitations imposed by losses and noises. |
| Hao Zhang | ESA: An efficient sequence alignment algorithm for biological database search on Sunway TaihuLight | In computational biology, biological database search has been playing a very important role. Since the COVID-19 outbreak, it has provided significant help in identifying common characteristics of viruses and developing vaccines and drugs. Sequence alignment, a method finding similarity, homology and other information between gene/protein sequences, is the usual tool in the database search. With the explosive growth of biological databases, the search process has become extremely time-consuming. However, existing parallel sequence alignment algorithms cannot deliver efficient database search due to low utilisation of the resources such as cache memory and performance issues such as load imbalance and high communication overhead. In this paper, we propose an efficient sequence alignment algorithm on Sunway TaihuLight, called ESA, for biological database search. ESA adopts a novel hybrid alignment algorithm combining local and global alignments, which has higher accuracy than other sequence alignment algorithms. Further, ESA has several optimizations including cache-aware sequence alignment, capacity-aware load balancing and bandwidth-aware data transfer. They are implemented in a heterogeneous processor SW26010 adopted in the world’s 6th fastest supercomputer, Sunway TaihuLight. The implementation of ESA is evaluated with the Swiss-Prot database on Sunway TaihuLight and other platforms. Our experimental results show that ESA has a speedup of 34.5 on a single core group (with 65 cores) of Sunway TaihuLight. The strong and weak scalabilities of ESA are tested with 1 to 1024 core groups of Sunway TaihuLight. The results show that ESA has linear weak scalability and very impressive strong scalability. For strong scalability, ESA achieves a speedup of 338.04 with 1024 core groups compared with a single core group. We also show that our proposed optimizations are also applicable to GPU, Intel multicore processors, and heterogeneous computing platforms. |
| Natalie Grattan | Automatically Informing the Cause of Software Defects | Increasingly, AI is replacing or changing existing technical structures. With that, we must invest resources into ensuring AI and machine-learning algorithms have the capacity to be explainable, especially if these are to be utilized by a non-technical audience. Software defect prediction, an application of machine-learning, uses predictive algorithms to identify software modules likely to contain defects. In a software project, the number of defects usually outweighs the resources available to find and fix these defects. Existing models often achieve high performance, but lack actionable information, such as the severity or the cause of the defect. More informative defect prediction models would help developers allocate testing resources more efficiently. To begin to bridge this gap, we present our causal comments model. We use BERT (Bidirectional Encoder Representations for Transformers) to transform causal comments into numerical vectors. These vectors are used as features to build a model to automatically predict whether a comment from an issue log in an issue tracking system informs the cause of a software bug or not. With this model, we later intend to predict whether comments are causal or not, providing a crucial step towards a more informative defect prediction model. |
| Sobhan Latifi | Optimization of 3d models in Engineering applications | With the utilization of 3D printing technology, we have the capability to seamlessly incorporate digital shapes and objects into our physical environment. The process of translating digital objects into the tangible world requires to know their real-world applications and features. For instance, when these objects need to exhibit qualities such as strength, resistance to heat or pressure, or specialized functionality within industrial contexts, it's essential to refine these attributes within the digital model prior to the printing process. This research is centered around chromatography, a field concerned with the separation of material components. Objects intended for use in chromatography must deliver the highest level of separation quality, necessitating meticulous optimization. In this presentation, I will explain about a specific technique that can produce high-performance objects customized for chromatographic applications. |
| Maryam Nakhoda | Quantifying and addressing uncertainty in the measurement of interdisciplinarity | A common method for quantifying the interdisciplinarity of a publication is to measure the diversity of the publication’s cited references based on their disciplines. Here we examine the criteria that must be satisfied to develop a meaningful interdisciplinary measure based on citations and discuss the stages where uncertainty or bias may be introduced. In addition, using the Rao-Stirling diversity measure as an exemplar for such citation-based measures, we show how bootstrapping can be used to estimate a confidence interval for interdisciplinarity. Using an academic publication database, this approach is used to develop and assess a reliability measure for interdisciplinarity that extends current methods. Our results highlight issues with citation analysis for measuring interdisciplinarity and offer an approach to improve the confidence in assessing this concept. Specific guidelines for assessing the confidence in the Rao-Stirling diversity measure and subsequently other similar diversity measures are presented, hopefully reducing the likelihood of drawing false inferences about interdisciplinarity in the future. |
| Zahra Torabi | Using Machine Learning to Support the Development and Maintenance of Firewall Configurations | Managing network firewall structures can be a challenging task, especially when dealing with complex configurations filled with numerous firewall rules, and error rates may be higher than desirable when trying to maintain rulesets left by other staff members. Our research focuses on simplifying the development and maintenance of firewall configurations. To address these challenges, we are developing a tool that utilizes machine learning algorithms like the Bayesian network model and contextual analysis to predict the intended purpose of components and the decision tree to optimize the resulting rule-set within firewall configurations. We intend for our tool to aid in understanding and documenting legacy configurations, making transitions between administrators smoother. In this talk, I will present how a Bayesian network can predict the intended purpose of firewall components and how a decision tree can optimize the rule-set. |
| Garth Wales | End-to-end image segmentation with declarative neural networks | Classical image segmentation methods take a mathematical basis and apply it to the image domain. We are interested in the feasibility of combining the inductive biases of normalized cuts, a graph theoretic approach, with the representational power of deep neural networks. |
| Abira Sengupta | Generalising Axelrod's Metanorms Game through the use of explicit domain-specific norms | Achieving social order in societies of self-interested autonomous agents is a difficult problem due to lack of trust in the actions of others and the temptation to seek rewards at the expense of others. In human society, social norms play a strong role in fostering cooperative behaviour - as long as the value of cooperation and the cost of defection are understood by a large proportion of society. Prior work has shown the importance of both norms and metanorms (requiring punishment of defection) to produce and maintain norm-compliant behaviour in a society, e.g. as in Axelrod’s approach of learning of individual behavioural characteristics of boldness and vengefulness. However, much of this work (including Axelrod’s) uses simplified simulation scenarios in which norms are implicit in the code or are represented as simple bit strings, which limits the practical application of these methods for agents that interact across a range of real-world scenarios with complex norms. This work presents a generalisation of Axelrod’s approach in which norms are explicitly represented and agents can choose their actions after performing what-if reasoning using a version of the event calculus that tracks the creation, fulfilment and violation of expectations. This approach allows agents to continually learn and apply their boldness and vengefulness parameters across multiple scenarios with differing norms. The approach is illustrated using Axelrod’s scenario as well as a social dilemma from the behavioural game theory literature. |
| Vaughan Kitchen | Dynamic Accumulator Selection for Disjunctive Query Evaluation | When searching large quantities of text, efficiency is considered highly important. Despite many attempts to improve the efficiency of search systems by improving the accumulation of document impact scores into the result, accumulation still largely governs the search efficiency. Recent research suggests that there may be no one size fits all accumulation algorithm and that in different contexts certain algorithms or parameter tuning perform better. I believe it may be possible to classify queries to chose between accumulation algorithms. The first step in this is to analyse why one algorithm or tuning may perform better than another and in what cases. Once this knowledge is found, dynamic algorithm selection may be possible and push forward the efficiency frontier. |