PHD scholarships: Making Computers Understand Place

Your PhD research will contribute to a project at the forefront of fundamental AI and spatial information research: Making computers understand common language about place. This project, funded by the Australian Research Council (2017-2019), will deliver the computational methods to capture, model, process, and interact on human place knowledge, which conceptualizes places and their relations, instead of using coordinates and maps. The outcomes will enable human-machine interaction that is capable of responding to human intuition.

The project engages in research in artificial intelligence, language technologies, and spatial information science – you will work in an interdisciplinary environment. The project is also geographically spread between The University of Melbourne, the Australian National University in Canberra, and the University of California, Santa Barbara. The scholarships, however, are allocated to the two Australian universities, which are Australia’s leading research universities.

We seek applicants covering one of the following tasks:

1. Capture (Melbourne: led by Tim Baldwin). This task addresses the extraction of place knowledge in conversational context from large-scale text corpora such as social media and web sources, or crowd-sourced from mobile apps.

2. Modelling (Melbourne: led by Stephan Winter). This task addresses the representation of extracted place knowledge in place graphs, especially the representation of context with places and their relationships.

3. Reasoning (Canberra: led by Jochen Renz). This task addresses context-sensitive spatial reasoning beyond maintaining a locally consistent database. We expect to make progress on this long avoided challenge by considering the spatial context derived from the integration of place knowledge and maps.

4. User Interaction (Melbourne: led by Martin Tomko). This task addresses the user interaction with a place graph, enabling users to query or modify stored data by developing methods answering user queries taking into account the context of the querying user.

Selection Criteria

* Masters or Honours Degree in Computer Science, Geodesy/Geomatics/Geoinformatics, or another relevant discipline;

* A weighted average mark at least satisfying the entry requirements of the PhD programs at the respective universities;

* Demonstrated ability to perform independent research (e.g., by a research component in the master degree);

* Excellent written and verbal communication skills to technical and non-technical audiences (e.g., demonstrated in the written application and interview);

* Software development skills;

* Ability to work in (at least one of) spatial data handling and analysis, database systems, machine learning, natural language understanding, and artificial intelligence;

* Ability to work co-operatively in a multi-disciplinary environment and team.


Scholarships are for three years, with the expectation to complete a PhD in this time-frame. The universities have excellent environments and frameworks to support PhD research. Also, successful candidates will be provided a tuition fee waiver. Scholarships are sufficient to cover living costs in Melbourne or Canberra. The project also provides funds for conference travels commensurate with opportunity.

How to apply

Send your letter of application (addressing the selection criteria, and identifying the particular task of interest), a CV, and a transcript of your last degree per email to the project leader, Prof Stephan Winter (

The application deadline is 20 March 2017, for a start later in 2017. Selection is pending admission at the respective university and visa processes.

Vacancy: Research Fellow in Autonomous Spatial Data Integration

All #AI #ML #DB talent out there!

A Research Fellow position is available within a project funded by the Australian Research Council, “Self-Healing Maps”. The project aims to expand our ability to automatically integrate real-time data in map databases of high integrity, enabling adaptive and autonomous protection of maps from the insertions of erroneous or malicious data, and the detection and correction of inconsistencies.

You will work within under the guidance of Dr Martin Tomko and other chief investigators across all aspects of Geomatics and spatial information. You will be in charge of conducting research and coordinating the efforts leading to a new generation of mechanisms supporting spatial data integration. You will devise novel methods enabling the autonomous and evolving discovery and rectification of errors, discrepancies and inconsistencies in spatial databases. You will apply your outstanding computational skills to demonstrate the validity of your theoretical contributions through the implementation of proof-of-concept software.

You will conduct collaborative and independent research, leading to the preparation and publication of research outcomes in conferences and journals. You will be located in the Department of Infrastructure Engineering in the Melbourne School of Engineering.


For more information, contact Dr Martin Tomko (

“Ripe for the Picking? Dataset Maturity Assessment based on Temporal Dynamics of Feature Definitions”

Our paper that has been recently accepted for publication in the International Journal of Geographical Information Science investigates a thus-far unexplored aspect of spatial data of particular relevance to VGI usability (OSM as a case) – the differences between (geometric) feature definitions within a feature class. I am particularly pleased by this paper as it is the outcome of a short, but intense and very satisfying collaboration with Stephen Maguire, a Masters of IT (Spatial) student here at the University of Melbourne. Congratulations Stephen!


From the abstract:

Map databases traditionally capture snapshot representations of the world following strict data collection and representation guidelines. The content of these map databases is often assessed using data quality metrics focusing on accuracy, completeness and consistency. The success of volunteered geographic information, supporting evolving representations of the world based on fluid guidelines, has rendered these measures insufficient. In this paper, we address the need to capture the variability in quality of a map database. We propose a new spatial data quality measure — dataset maturity — enabling assessment of the database based on temporal trends in feature definitions, specifically geometry type definitions. The proposed measure can be (1) efficiently used to identify feature definition patterns reflecting community consensus that could be formalised in community guidelines; and (2) deployed to identify regions that would benefit from increased editorial activity to achieve greater map homogeneity. We demonstrate the measure based on the content of the OpenStreetMap database in four regions of the world, and show how the proposed dataset maturity measure captures a distinct quality of the datasets, distinct to data completeness and consistency.

Read the full thing in pre-print, supplementary materials here.

Maguire, S., & Tomko, M. (accepted 2017). Ripe for the Picking? Dataset Maturity Assessment based on Temporal Dynamics of Feature Definitions. International Journal of Geographical Information Science. doi:10.1080/13658816.2017.1287370

Our paper on D-Log accepted to Pervasive and Mobile Computing

A new output from our TRIIBE project is now available: “D-Log: A WiFi Log-based differential scheme for enhanced indoor localization with single RSSI source and infrequent sampling rate“, by our TRIIBE team (Yongli Ren, Flora Salim, Martin Tomko, Brian Bai, Jeffrey Chan, Kyle Qin and Mark Sanderson) is now online here. It presents a way to post-process large amounts of single AP+RSSI fixations to better estimate the approximate location of users in an indoor environment. I will publish a pre-print version soon here.



(Re)Starting at The University of Melbourne

“I am looking for PhD students with topics related to computational urban GIScience” is one of the main pieces of content I have updated on this website. Being back in Melbourne is exciting, and I am keen to sink my teeth in some ideas that have been germinating for a while. Please, spread the word, or get in touch. Note that only students with an outstanding profile from their Masters studies may be eligible for local funding. But if you really see yourself in the profile outlined here, get in touch anyway.

Exploring patterns of individual transcontinental oscillation between Australia and Europe. A subjective study.

The cryptic title hides a prosaic content: I will be wrapping up here in Zurich by the end of the year and I will be returning to the University of Melbourne and the Geomatics group at the Department of Infrastructure Engineering, from January 2016. I am looking forward to rekindling my existing ties in Melbourne and developing new ones, and I hope to continue my collaboration with my amazing Swiss colleagues.


And I am looking for some great PhD students interesting in exploring urban GIScience with me…

Information Retrieval. What are the temporal trends?

Part II of series, that started by looking at GIScience (my core interest)

This is, as far as I know, the first publication of temporal trends in Google Metrics.

For the last three years, Google Scholar has been releasing their Google Scholar Metrics. Recently, they released the 2015 batch.

These Metrics provide an insight into the most successful/impactful publication outlets for individual disciplines (and subdisciplines) and also allow one to explore the most cited papers by their h5 index (h5 for a venue/author is the n (number) of papers with at least n citations, in this case for a 5 year period).

There are problems with the way these data are collected (not all venues are monitored, and the coverage may not be 100%, see here). The coverage has been slowly improving over the years. While Computer Science is relatively well covered, some conferences/workshop published in the well respected Springer Lecture Notes in Computer Science series are not monitored by Google and the individual volumes can not be well sorted into disciplines anyway.

Anyway, this project has been running for 3 years now and we can start looking at some trends (without any statistical insights, for this the series are too short). It is worth to note some separation of the journals into tears ( purely visually). Note that this may not say anything about the quality of the venue itself but maybe the audience is smaller/more niche).

It would be worth to compare these trends with the sibling disciplijne of data mining/knowledge discovery, where many venues are used by both communities.

Also note the discussion of the h5 index in here (Vrettas and Sanderson, 2015), suggesting that the size of the venue tends to lead to an over-inflation of its h5index. I would be happy to include additional venues into this, and share data for deeper investigation. I acknowledge the seed list of IR venues from @IR_oldie for this analysis.

I am looking forward to comments!

GIScience Google Metrics trend


Vrettas, G. and Sanderson, M. (2015), Conferences versus journals in computer science. Journal of the Association for Information Science and Technology. doi: 10.1002/asi.23349


The R Hadleyverse for rvest, tidyr, stringr, dplyr and ggvis! Great little problem to learn these!