Up to Four PhD scholarships, Visiting Research Fellow / Student, and a Postdoc position are available!

Research Postdoc in Microsoft-RMIT Cortana Intelligence Institute

A postdoc position available in the new Microsoft RMIT Cortana Intelligence Institute. If you have strong track record in machine learning, and have experience working with the following, and would like to be part of the team advancing the next generation of intelligent assistant for working professionals, please contact me.

  • spatial/temporal data,
  • text,
  • user data,
  • perhaps a little NLP,
  • signal processing,
  • audio processing.

Visiting Research Fellow/Student in Trajectory Data Mining

A 3-4 month visiting fellow position is available for an excellent postdoc/PhD student in trajectory data mining. The Visiting Research Fellow and/or PhD will work in an exciting research project partnered with Northrop Grumman Corporation, a US-based global aerospace technology company. The candidate’s role is primarily to plan, develop and engage in spatio-temporal data mining and applied machine learning research using several large-scale airport trajectory datasets. The  candidate will be expected to produce high quality outputs in A/A* journal / conference outlets. The outcomes of the research will have a significant impact in the domain of airport operations.

The candidate needs to be highly skilled in machine learning in R, Matlab, or other platforms and have sound knowledge in statistics, data modelling, and have published papers related to unsupervised and semi-supervised learning techniques. Experience with processing and analysing large scale spatio-temporal data is desirable.

The scheme is primarily intended for PhD students (who have already passed their Qualification Exam or Confirmation of Candidature at their respective university) or for postdocs who seek to conduct joint research in trajectory data mining and machine learning.

We will cover return airfares and living allowance at competitive rates.

To apply, contact Dr. Flora Salim Flora.Salim@rmit.edu.au.

PhD Scholarship

Scholarships in four PhD topics are available:

  1. Scalable and Transferrable Occupant Behaviour Learning with Multi-Sensor Data from Multi-region Living Labs co-supervised by Prof Mikkel Kjaergaard, University of Southern Denmark
  2. Efficient Temporal Segmentation in Mining Big Time Series for Accurate Prediction of Heterogeneous Activities and Events, co-supervised by Prof Eamonn Keogh, UC Riverside
  3. Spatio-temporal Analytics and Deep Learning of Big Trajectory Data for Situation Awareness, co-supervised by Dr. Jeffrey Chan, RMIT University
  4. Cross Domain Data Fusion and Analytics of User Contexts and Behaviours for Personalized Recommendation Systems, co-supervised by Dr. Yongli Ren, RMIT University

PhD and Scholarship Application Process

Closing dates:
All applicants need to first submit their documents to Flora.Salim@rmit.edu.au by 15th September 2018 as there is a PhD admission process prior to the scholarship application process and both process needs to be completed by the central deadline (31st October 2018).

Research scholarships information for topics 1-4 is available at RMIT scholarship website.

The positions will commence in Semester 1, 2019 (March 2019).

Required documents:
CV (which includes a list of publications), academic transcripts, and a research proposal.
For international students from non-English speaking countries, when submitting the application to RMIT system, a proof of English proficiency may also be required (IELTS or TOEFL).

Why RMIT?
RMIT is a global university of technology and design, focused on creating solutions that transform the future for the benefit of people and their environments. RMIT is ranked in the top 100 in 2018 QS World Ranking for Computer Science and Information Systems. RMIT has also achieved two ERA 4 rankings (above the world standard) in the fields of Artificial Intelligence and Information Systems.

Why Melbourne?
Melbourne is the most liveable city in the world, for six years in a row, according to The Economist’s liveability ranking.

For more information, contact Flora Salim flora.salim@rmit.edu.au

Scalable and Transferrable Occupant Behaviour Learning with Multi-Sensor Data from Multi-region Living Labs

The focus of this PhD project will be on data-driven occupant research methods to support evidence-based decision-making. The development in machine learning and data mining methods that are generic and robust to multiple sensing modalities will enable a rich insight about occupant behaviors. This wealth of building sensor data opens new opportunities for extracting knowledge from data and data-driven modeling of occupant behavior. Among others, the data offers opportunities for creating models that are more individually customized to the particular occupant, building or climate zone. This wealth of data also creates new threats to the occupant in terms of the violation of the individuals’ right to privacy that has to be addressed.

The goal is to demonstrate how in situ data can be converted to meaningful information with key traits (e.g., occupant diversity, temporal trends, triggers, patterns and relationships). The project will investigate the potential applications of data-based methods for knowledge discovery and modeling of occupant behavior. The project will establish and evaluate new data-driven research opportunities in the light of the rich variety of models already developed in the field to understand the advantages and disadvantages of the different modeling paradigms.

The rich datasets and online repositories from the two living labs– RMIT and University of Southern Denmark – will be used by the PhD candidate. OU44, the 8500m2 3-storey building, is a highly energy efficient teaching building at the University of Southern Denmark in Odense, and a living lab for research in energy informatics and occupancy behavior, with full capability to monitor, manage and control the building operation. The building is equipped with energy efficient technologies including, ventilation units with heat recovery, LED lights, underfloor heating, PV modules, and heating, lighting and electricity consumption sub-meters, and temperature, humidity, CO2, Lux and PIR sensors on the room level. RMIT and a project on activity based working in multiple offices of Arup Melbourne and Sydney provide another living lab with multi-sensor, multi-device tracking data of occupants.

This project will extend the existing research by Dr. Salim and Prof Kjaergaard on occupant sensing, tracking, and prediction. The candidate needs to have a strong background in algorithms and data mining/machine learning. This project will contribute to the development of a new International Annex by IEA EBC (International Energy Agency’s Energy in Buildings and Communities) on “Occupant behaviour-centric building design and operation”, particularly the Sub Task 2 on “data-driven methods for occupant behavior modelling”.

References:

[1] A.J.R. Ruiz, H. Blunck, T.S. Prentow, A. Stisen, M. B. Kjærgaard, Analysis methods for extracting knowledge from large-scale WiFi monitoring to inform building facility planning. PerCom 2014, pp. 130-138.

[2] I.B.A-Ang, M. Hamilton, F. D. Salim, RUP: Large Room Utilisation Prediction with carbon dioxide sensor, Pervasive and Mobile Computing, Volume 46, 2018, Pages 49-72.

Contact Details:

To discuss this project further please contact Dr. Flora Salim flora.salim@rmit.edu.au

1st Supervisor – Dr. Flora Salim, School of Science, RMIT University.
2nd Supervisor Prof. Mikkel Kjaergaard, University of Southern Denmark

Return to Scholarships

 

Efficient Temporal Segmentation in Mining Big Time Series for Accurate Prediction of Heterogeneous Activities and Events

Temporal segmentation is the key to processing big time-series data from diverse sources, low-level and high-level human activities. Many techniques have been applied for activity recognition using various kinds of sensor data. However, there is still a need for efficient and generic method to segment the data prior to recognition or prediction tasks, as the transition times between the activities are unknown. Temporal segmentation can also be used to understand humans’ mobility patterns across the whole day. To deal with time-series data from diverse range of sources, a generic and fast temporal segmentation method that is applicable to heterogeneous data is required [1]. As the input data is not just a single time series, an efficient method that is applicable to high-dimensional data from multivariate channels is required. In addition the optimal time window for finding temporal segments that minimize the noise while maximising the accuracy of the recognised is also required [2].

Matrix Profile [3] is recently introduced as an efficient method for time-series processing. It annotates a time series by recording the location of and distance to the nearest neighbor to every subsequence. This information trivially gives the answer to both time series motifs and time series discords, perhaps the two most frequently used primitives in time series data mining. One attractive feature of the Matrix Profile is that it completely divorces the high-level details of the analytics performed, from the computational “heavy lifting”. The Matrix Profile can be computed using the appropriate computational paradigm, CPU, GPU, FPGA, distributed computing, anytime computation, incremental computation, etc., but this can all be hidden from the analyst.

This project will extend the existing research by Dr. Salim and Prof Keogh on time series data mining. The candidate will expand the existing MP and the temporal segmentation method for multiple time-series. In this project, the candidate will design and develop a new method for mining multiple time-series for multi-view time series mining and multi-task learning. The candidate needs to have a strong background in algorithms and data mining/machine learning.

References:

[1]. Sadri, A., Ren. Y., Salim, F.D., (2017), ‘Information gain-based metric for recognizing transitions in human activities’, in Pervasive and Mobile Computing, vol. 38, part 1, July 2017, Elsevier, pp. 92–109.

[2] Liono, J., Qin, K, Salim, F.D., “Optimal time window for temporal segmentation of sensor streams in multi-activity recognition”, 13th Annual International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services (MobiQuitous 2016).

[3] C-C M. Yeh, Y. Zhu, L. Ulanova, N. Begum, Y. Ding, H. A. Dau, D. F. Silva, A. Mueen, E. Keogh (2016). Matrix Profile I: All Pairs Similarity Joins for Time Series: A Unifying View that Includes Motifs, Discords and Shapelets. IEEE ICDM 2016.

Contact Details:

To discuss this project further please contact Dr. Flora Salim flora.salim@rmit.edu.au

1st Supervisor – Dr. Flora Salim, School of Science, RMIT University.

2nd Supervisor –Prof. Eamonn Keogh, University of California – Riverside

Return to Scholarships

 

Spatio-temporal Analytics and Deep Learning of Big Trajectory Data for Situation Awareness

Situation awareness is vital to effective decision making in complex and dynamic sensed environments. The pervasiveness of distributed sensor systems in public transits and precincts such as airports is generating significant amounts of heterogeneous, spatio-temporal data that, correctly handled, can offer significant insights for situation awareness across a wide range of scenarios.

Existing time-series mining techniques cannot be directly applied to spatio-temporal interval data [1]. Existing spatio-temporal clustering methods also fail to measure similarity across multiple domains of data (e.g. spatial, temporal, data domain). The gaps in existing research demonstrate the need for new methodologies for effective fusion and pattern recognition methods that can utilise the proliferation of heterogeneous data.

We will use multiple trajectory and mobility datasets in this project. One particular dataset that will be used is a large-volume trajectory data from civilian airports, provided by a partner of this project. This research aims to develop robust techniques and tools to discover spatio-temporal patterns from a variety of sensor and trajectory data and complementary data sources to inform and improve situation awareness in different scenarios.

The candidate will need to perform data processing and cleaning of a large volume real-world data, and then explore the characteristics and patterns in the data. The candidate will investigate suitable machine learning techniques to develop models that will deal with the problem of the complexity of the scale and dimension of the trajectory and sensor data.

The key objectives of this research are as follows:

  1. To investigate novel spatio-temporal clustering techniques for discovering trends and characterising ground transport vehicle operations based on multiple similarity measures and objectives.
  2. To develop a deep learning model from large scale unlabeled data, potentially with help from some labeled data, enabling multi-task learning and situation recognition.

The candidate needs to have a strong background in algorithms and data mining/machine learning.

References:

[1].  Shao, W., Salim, F.D., Song, A., Bouguettaya, X., (2016) “Clustering Big Spatio-temporal Interval Data”, IEEE Transactions on Big Data, vol. 2, no. 3. pp. 190-203.

Contact Details:

To discuss this project further please contact Dr. Flora Salim flora.salim@rmit.edu.au

1st Supervisor – Dr. Flora Salim, School of Science

2nd Supervisor – Dr. Jeffrey Chan, School of Science

Return to Scholarships

 

 Cross Domain Data Fusion and Analytics of User Contexts and Behaviours for Personalized Recommendation Systems

 Project Description – This project will investigate human behaviour using datasets from multiple domains, to construct accurate representations of their behaviour patterns. With the popularity of human sensing and Internet of Things, traces of human behavior are captured in multiple data sources across different domains. Cross-domain data fusion has recently became an important problem to solve given the rise of big data with disparate sources [1]. Characterizing contexts and behaviors of users is essential to provide personalized services and recommendations. By integrating context-sensing, activity recognition, pattern recognition, and recommendation techniques, the expected outcomes of the project will include:

  1. a theoretical framework for cross-domain fusion and analytics of human behaviours;
  2. a suite of algorithms and techniques for characterizing user contexts, for profiling and predicting user behaviors, and for issuing personalized recommendations to users.
  3. Evaluation of the framework and the personalized recommendation systems for two case studies: 1) Activity Based Working (ABW); 2) Retail sentiment analysis.

ABW encourages collaboration and better knowledge sharing through flexible working environments and a more mobile workforce. We will analyse a challenging multi-sourced dataset from Building Management Systems (BMS), Wi-Fi infrastructures, project assignment systems, and a customised app given to a sample of workers. The overlay of fine-grained human behaviour patterns with building operational data will allow a rich model of physical (movement), social grouping, and spatial contexts to be constructed [2]. This study will generate the first integrated analytics of workers’ trajectories, engagement, collaboration, and work behaviours, and productivity patterns from human tracking, space utilisation and BMS data.

In retail marketing, mapping a person’s physical behaviour to their opinion is a difficult problem to solve, i.e. there is no obvious way to continually measure a person’s opinion in the same manner as it is possible to measure their location. It is hypothesised that social sentiment data is able to provide statistically significant predictions by measuring the change in location events which coincide with changes in social sentiment. This research aims to discover the links between any social media sentiment changes and the changes in visitation patterns within the chosen retail venues of study. The study will use cross-domain logs from Wi-Fi and social networks. The candidate needs to have a strong background in algorithms and data mining/machine learning.

References:

[1]. Zheng, Y., “Methodologies for Cross-Domain Data Fusion: An Overview,” in IEEE Transactions on Big Data, vol. 1, no. 1, pp. 16-34, March 1 2015.

[2]. Ren, Y., Tomko, M., Salim, F.D., Ong, K., Sanderson, M. (2015). “Analyzing Web Behavior in Indoor Retail Spaces”. Journal of the Assoc. for Info. Sci and Tech. (JASIST). v. 68, 1.

Contact Details:

To discuss this project further please contact Dr. Flora Salim  flora.salim@rmit.edu.au

1st Supervisor – Dr. Flora Salim, School of Science

2nd Supervisor – Dr. Yongli Ren, School of Science

Return to Scholarships