On this thematic evening, students illustrate on their research project how open source and data matters to them.
- Erdem Yilmaz
- Mohammad T. Al-Antary
- Ilya Zaznov
- Frank Gadban
- Maha Abdelmohsen
- Samar Mahmoud
18:20 - Feel free to join the online meeting to chat with other participants
18:30 - Short introduction (5 min) of the evening by Julian Kunkel and Cornelia Boldyreff
18:35 - Presentations
20:35 - Closing Discussion
The livestream link will be open from 18:00 for networking, and the event will start at 18:30 prompt. We’ll keep the link open afterwards for discussion.
We organize this event annually to give students space to talk about their projects; the talks should be around 20 minutes (including 2-3 min discussion). If you are interested to give a talk, contact us!
Smart Mapping of Scientific Workflows onto Heterogeneous Resources
New and accelerated hardware is constantly being added to the HPCs and Data Centers, increasing the heterogeneity of the scientific simulation environments. To increase the performance of a simulation by better utilizing the allocated resources requires detailed knowledge about the underlying hardware. Mapping tasks to heterogeneous resources is a complex (NP-Complete) problem on its own. If we can create an abstract way to structure a workflow so that it can choose the best mapping by utilizing accelerated hardware, we can address both of the aforementioned problems. Having well-defined operators with associated performance models, we can predict a task’s makespan on a given host, accurately. This prediction can then be used within evolutionary mapping techniques to help explore a near-optimal solution where underlying heterogeneous hardware resources are better utilized with specialized implementations. A workflow, based on such an abstraction, will be able to adapt itself to new hosting environments with minimal change. Having this headspace, scientists can spend time on improving their model/simulation not on intrinsic details required to adapt their workflow to new environments.
Diabetic Retinopathy Classification using Deep Learning
Diabetic Retinopathy (DR) is a highly prevalent complication of diabetes mellitus, which causes lesions on the retina that affect vision which may lead to blindness if it
is not detected and diagnosed early. Convolutional neural networks (CNN) are becoming the state-of-the-art approach for automatic detection of DR by using fundus images. The high-level features extracted by CNN are mostly utilised for the detection and classification of lesions on the retina. This high-level representation is capable of classifying different DR classes; however, more effective features for detecting the damages are needed.
This research work proposes the multi-scale attention network (MSA-Net) for DR classification. The proposed approach applies the encoder network to embed the retina image in a high-level representational space, where the combination of mid and high-level features is used to enrich the representation. Then a multi-scale feature pyramid is included to describe the retinal structure in a different locality. Furthermore, to enhance the discriminative power of the feature representation a multi-scale attention mechanism is used on top of the high-level representation. The model is trained in a standard way using the cross-entropy loss to classify the DR severity level.
In parallel as an auxiliary task, the model is trained using the weakly annotated data to detect healthy and non-healthy retina images. This surrogate task helps the model to enrich its discriminative power for distinguishing the non-healthy retina images. The proposed method when implemented has achieved outstanding results on two public datasets: EyePACS and APTOS.
Breaking through the Chinese walls to the open-source and data for my research project
Open source plays important role in many different fields, especially in research. The research project I’m currently working on is not an exception, its goal is to explore the application of machine and deep learning methods in algorithmic trading with the main focus on the trading strategy. In this process I need at least three core components available in open access:
- Prior research papers in this field
- Code for the machine learning models described in these papers
- Data-sets used for training and testing these models
At the beginning of my project, I expected to easily open source all of these, however, in each of these three dimensions I encountered challenges of two kinds: availability and quality. I would like to share on the upcoming thematic evening what are the particular open sourcing challenges I am experiencing and how I’m overcoming these Chinese walls.
Performance of the S3 interface (Amazon Simple Storage Service) in an HPC environment
The line between HPC and Cloud is getting blurry. Performance is still the main driver in HPC, on the other hand, cloud storage systems assume to offer low-latency, high throughput, high availability, and scalability. The Simple Storage Service S3 has emerged as the de-facto storage API for object-storage in the Cloud. It is inevitable to assess its performance for HPC workloads and to check if it is already a viable alternative, or if further advances are necessary.
Enhancing Social Skills of Children with ASD by Assistive Technology
This research aims to investigate the potential of combining virtual environments with social robots as a novel approach to address some of their limitations and train the social skills of children with high-functioning autism (ASD). A non-immersive (desktop) virtual reality environment that employs a 3D robot has been designed. The developed environment aims to enhance the social skills of children with high-functioning ASD through a social skills training program guided by a parent or a teacher. The motivation of this research is to provide a tool that can be widely accessible, cheap and easily used by parents and teachers either at home or at school. The developed training program targets three social skills: imitation skills, emotion recognition skills, and intransitive gestures skill. The experimental sessions have been conducted online and on-site. 15 children with ASD (4-12 years) participated in this study. The participants were taught to recognize 6 basic emotions and 11 intransitive gestures (Phase I), to imitate these emotions and gestures (Phase II), and to produce them in appropriate social contexts (Phase III). Across all the three phases for each skill, significant differences were found between the results of the pre-test, post-test, and follow-up test.
Dynamic Image Representations for Crowd Anomaly Detection using Generative Adversarial Networks
Anomaly detection within crowded environments is a key challenge in the crowd behaviour understanding and computer vision fields. Application of crowd anomaly detection has improved recently, however, advancements of accuracy and computation (processing power and time) are still required. The proposed framework presents an approach to crowd behaviour anomaly detection using dynamic image representations as an alternative to optical flow extractions for temporal development feature extraction. The features are used in conjunction with image-to-image translation using conditional generative adversarial networks (CGANs) for anomaly detection within crowds. The proposed framework is evaluated on standard benchmark datasets as well as the high-density dataset (AHDCrowd). The experimental results obtained have demonstrated the efficacy of this approach in comparison to the state-of-the-art crowd anomaly detection methods.
About the speakers
Erdem Yilmaz is a software engineer, working at Ultromics, United Kingdom. Progressing on his research project to full-fill the requirements of a PhD in Computer Science Dept. of University of Reading, United Kingdom.
Mohammad T. Al-Antary
Mohammad T. Al-Antary received the B.Sc. degree in computer information systems from The University of Jordan, Amman, Jordan, in 2015, and the M.Sc. degree in enterprise systems and database administration/computing and information systems from the University of Greenwich, U.K., in 2017, where he is currently pursuing the Ph.D. degree in computing and information systems with the School of Computing and Mathematical Sciences. He is currently working as a Teaching Assistant with the School of Computing and Mathematical Sciences, University of Greenwich. His research and professional interests include data science, medical image processing, machine learning, business intelligence, and big data.
Ilya Zaznov is a PhD student in the department of Computer Science, University of Reading, UK. His PhD research area is an application of machine/deep learning methods to the algorithmic trading. The main topic of the thesis is formulated as “Application of Deep Learning methods to the Limit Order Book and Order Flow Data for Stock Price Prediction”.
Frank Gadban is a PhD student at the Hamburg University, currently investigating the Convergence between HPC and Cloud.
Maha Abdelmohsen is a PhD candidate at the University of Greenwich, London. She is conducting research related to autism and how assistive technology can enhance their social skills. She developed a free desktop virtual environment tool for training the social skills of children with ASD.
Samar Mahmoud is a PhD candidate at the University of Greenwich, London. She is currently investigating the possible enhancement of crowd anomaly detection using CGANs and dynamic image representations. She has also created a public abnormal high-density crowd (AHDCrowd) dataset for researchers to train and test crowd anomaly detection methods.
This event is brought to you by: BCS Open Source specialist group