Research Seminar

Speaker: Mr. Xin Lin and Mr. Minjie Lyu

Abstract: We trained artificial neural network (ANN) models to classify peripheral blood mononuclear cells (PBMC) in chronic lymphoid leukemia (CLL) patients. The classification task was to determine differences in gene expression profiles in PBMC pre-treatment (with ibrutinib) and on days 30, 120, 150, and 280 after the start of treatment. Twelve datasets represented clinical samples containing a total 48,016 single cell profiles were used to train and test ANN models to classify the progress of therapy by gene expression changes. The accuracy of ANN classification was >92% in internal cross-validation. External cross-validation, using independent data sets for training and testing, showed the accuracy of classification of post-treatment PBMCs to more than 80%. To the best of our knowledge, this is the first study that has demonstrated the potential of ANNs with 10x single cell gene expression data for detecting the changes during treatment of CLL.

Speaker: Mr. Xiang Zhang

Abstract: Pregnancy induces additional stress on a woman's body. Pregnancy-related complications affect the health of both mother and baby during the pregnancy itself, and at the time of birth. Wearable sensors enable vital signs collection and health monitoring outside the clinic setting. With monitoring and prediction of health events using wearable devices, better pregnancy outcome can be achieved by early diagnosis and early intervention. However, several challenges need to be addressed on health monitoring system implementation, such as data accuracy, accuracy and adequacy of medical knowledge, privacy, standards and governance. The proposed conceptual solution combines data streaming, filtering, cross-calibration, use of medical knowledge for system operation and data interpretation, and IoT-based calibration using certified linked diagnostic devices. Integration of blockchain technologies and artificial intelligence offers a solution to the increasing needs for higher accuracy of measurements of vital signs, high-quality decision-making, and dependability, including key medical and ethical requirements of safety and security of the data.

Speaker: Mr. Hao Wang and Dr. Anthony Bellotti

Abstract: Discrete time survival models (DTSM) are used to build dynamic credit risk models; i.e. modelling changes in risk over time. Such models are valuable for forecasting borrower behaviour over time, expected default rates, macroeconomic risks and for stress testing in banking. Recently, there has been great interest in developing DTSM in a machine learning framework. We contribute to this research theme and address the problem of explainability which is an important concern for financial practitioners and regulators. We present some preliminary results based on US mortgage data.

Speaker: Mr. Shihe Wang

Abstract: The Bayesian classification framework has been widely used in many fields, but the covariance matrix is usually difficult to estimate reliably. To alleviate the problem, many naive Bayes (NB) approaches with good performance have been developed. However, the assumption of conditional independence between attributes in NB rarely holds in reality. Various attribute weighting schemes have been developed to address this problem. Among them, class-specific attribute weighted naive Bayes (CAWNB) has recently achieved good performance by using classification feedback to optimize the attribute weights of each class. However, the derived model may be over-fitted to the training dataset, especially when the dataset is insufficient to train a model with good generalization performance. This paper proposes a regularization technique to improve the generalization capability of CAWNB, which could well balance the trade-off between discrimination power and generalization capability. More specifically, by introducing the regularization term, the proposed method, namely regularized naive Bayes (RNB), could well capture the data characteristics when the dataset is large, and exhibit good generalization performance when the dataset is small. RNB is compared with the state-of-the-art naive Bayes methods. Experiments on 33 machine-learning benchmark datasets demonstrate that RNB outperforms the compared methods significantly.

Speaker: Mr. Yiming Zhang

Abstract: Biomedical images are widely collected from various applications, which are used for patients' screening, diagnosis and treatment. The dark regions of biomedical images may play an important role in the bright regions. The enhanced details in the dark regions of biomedical images simultaneously maintain the quality of the rest of the images and reveal more information for doctors and surgeons in medical procedures. This work proposes a fast method to adaptively enhance the details in the dark regions of biomedical images, including X-rays, video frames of laparoscopy in minimally invasive surgery (MIS).

Speaker: Mr. Hejia Qiu

Abstract: Tensor-power (TP) recurrent model is a family of non-linear dynamical systems, of which the recurrence relation consists of a p-fold (a.k.a., degree-p) tensor product. Despite such the model frequently appears in the advanced recurrent neural networks (RNNs), to this date there is limited study on its memory property, a critical characteristic in sequence tasks. In this work, we conduct a thorough investigation of the memory mechanism of TP recurrent models. Theoretically, we prove that a large degree p is an essential condition to achieve the long memory effect, yet it would lead to unstable dynamical behaviors. Empirically, we tackle this issue by extending the degree p from discrete to a differentiable domain, such that it is efficiently learnable from a variety of datasets. Taken together, the new model is expected to benefit from the long memory effect in a stable manner. We experimentally show that the proposed model achieves competitive performance compared to various advanced RNNs in both the single-cell and seq2seq architectures.

Speaker: Dr. Bo Tang (Southern University of Science and Technology)

Abstract: Big Data Analytics is about efficiently extracting knowledge from data. It has a wide range of applications in exploratory analysis applications. The components of current big data systems are data collection, storage, integration, indexing and analytics. In this seminar, I will introduce the history of big data analytics, then we know what exactly big data is. Next, I will elaborate the advantages and limitations in current big data analytical systems/techniques. Last, I will briefly present the recent big data analytics research works in my group and summarize the technical challenges of big data analytics both in academia and industry. 

Speaker: Dr. Xianxu Hou

Abstract: Although significant progress has been made in synthesizing high-quality and visually realistic face images by unconditional Generative Adversarial Networks (GANs), there still lacks of control over the generation process. In this talk, our recent works on semantic face editing based on pretrained StyleGAN were introduced. We proposed a novel learning framework, called GuidedStyle, to guide the image generation process of StyleGAN with a knowledge network. Furthermore, we allow an attention mechanism in StyleGAN generator to adaptively select a single layer for style manipulation. As a result, our method is able to perform disentangled and controllable edits along various facial attributes, including smiling, eyeglasses, gender, mustache and hair color. Both qualitative and quantitative results demonstrate the superiority of our method over other competing methods for semantic face editing. Moreover, we show that our model can be also applied to different types of real and artistic face editing, demonstrating strong generalization ability. 

Speaker: Dr. Xinjia Jiang (Nanjing University of Aeronautics and Astronautics)

Abstract: This talk shared the basic concepts and trends of maritime logistics, especially in the area of container yard management. We further discussed an intelligent cross-sectional yard crane deployment problem in a busy transhipment hub, where a special "consignment strategy" is applied. Under this strategy, containers to the same destination vessel are usually stored together to facilitate the loading process and reduce the long distance travel of yard cranes. The storage allocation is planned considering the practical requirements from traffic control, space capacity and yard crane workload, etc. However, most of the studies on space allocation assumed that yard cranes are always ready whenever needed, while the yard crane deployment problem was neglected. This may lead to not only unnecessary operational cost, but also the infeasibility of some storage allocation plan. In this work, the intelligent cross-sectional deployment of yard cranes is studied to improve the situation. 

Speaker: Dr. Qianyu Liu

Abstract: With the foreseeable ultra-dense deployment of femtocells and rapid mobility characteristics in 5G, the traditional mobility management method employed in LTE will cause an increased number of unnecessary and delayed handovers. It will contribute to the lower performance of handover latencies and reliability of network connection. A functional handover for 5G scenario needs to be fast, reliable, accurate, high-precision and easy to maintain. To achieve this, the algorithm will need to have adaptive, predictive and self-organised characteristics in its core design. In this seminar, a project aiming to develop novel predictive and adaptive handover methods based on reinforcement learning (RL) for the application of 5G was shared.

Speaker: Dr. Yuchang Zhang and Prof. Ruibin Bai

Abstract: The ports have a great influence on import and export trade, so it is of significant importance to economic development of a country. Meanwhile, by optimizing the dispatch of container trucks in the port, a series of optimization goals (such as minimizing the total distance traveled by vehicles, total costs, vehicle idling rate, or maximizing equipment utilization and total profits) can be achieved. The overall efficiency of the port can thus be greatly improved. Deep Reinforcement Learning (DRL) achieved high notoriety in academia when AlphaGo won a Go match against the world champion Lee Sedol in 2016.This method has natural advantages for solving sequential decision problems, especially problems with uncertainties. In this seminar, a DRL based hyper-heuristic framework was introduced to solve a real-world dynamic truck dispatching problem with uncertain service time.

Speaker: Xinan Chen

Abstract: The number of international and domestic maritime trade has been expanding dramatically in the last few decades, and seaborne container transportation has become an indispensable part of international trade since the efficient and easy-to-use characteristics of containers. As an important hub of container transport, container terminals use a range of metrics to measure their efficiency, among which the hourly container throughput (i.e., the number of Twenty-foot equivalent unit containers, or TEUs) is the most important objective to improve. In this seminar, a GP (Genetic Programming) approach to build a dynamic truck dispatching system that is trained in real-world operation data with stochasticities was introduced.

Speaker: Prof. Ruibin Bai

Abstract: Full truckload transportation (FTL) in the form of freight containers represents one of the most important transportation modes in international trade. Due to large volume and scale, in FTL, delivery time is often less critical but cost and service quality are crucial. Therefore, efficiently solving large scale multiple shift FTL problems is becoming more and more important and requires further research. In one of our earlier studies, a set covering model and a three-stage solution method were developed for a multi-shift FTL problem. We extended the previous work and presents a significantly more efficient approach by hybridising pricing and cutting strategies with metaheuristics (a variable neighbourhood search and a genetic algorithm). The metaheuristics were adopted to find promising columns (vehicle routes) guided by pricing and cuts are dynamically generated to eliminate infeasible flow assignments caused by incompatible commodities. Computational experiments on real-life and artificial benchmark FTL problems showed superior performance both in terms of computational time and solution quality, when compared with previous MIP based three-stage methods and two existing metaheuristics. The proposed cutting and heuristic pricing approach can efficiently solve large scale real-life FTL problems.

Speaker: Dr. Jiawei Li

Abstract: Fuzzy logic is a form of many valued logic, different from the classical two valued logic in mathematics. As a methodology to handle uncertainty, it has been broadly used in building intelligent decision-making systems. Typical applications include control systems of subway, flight aid of helicopters, and medical decision making. In this talk, a brief introduction of fuzzy logic and fuzzy systems was given, and then a fuzzy controller was discussed as an example of fuzzy systems.