My interests in the similarity of motion started after having a few discussions with Roberto Tamassia regarding the properties of rigid transformations and while studying the theory of Davenport-Schinzel sequences. Subsequently, there was the ACM GIS 2007 work which soon led to broadening the interest into the efficiency of detecting such similarity on a larger-scale datasets, which became part of the dissertaion of Hui Ding. As luck may have it, the search for literature and observations of certain claims, motivated the investigation into different distance functions and similarity measures, and the observations that there is no unique comparative evaluation led to collaboration with Eamonn Keogh and his students (at that time).
More recently, I have applied the concept of similarity of motion to the topic of efficient detection of theft of mobile devices (PhD work of Sausan Yazji). The crux was to augment the dynamics-based similarity of motion with probabilistic values about possible-trajectories being taken, and efficiently detect when certan threshold is(not) preserved. As part of the work, the findings were extended to incorporate the concept of data reduction, in order to enable quicker analysis with a bounded error on the result.
Motivated by a technical report from Ralf Guting (who's always kind enough to send me the pre-prints of his works that he deems may be of interest), I started looking into the similarity of the, so called, semantic/symbolic trajectories and similarity at a larger-scale, at the level of warehouses. I had discussions on the field of semantic relatedness (which was the topic of her MS thesis) with Ivana Donevska and followed up with investigating how that "context" of similarity could be brought into a multi-criteria function for similarities of trajectories at a database level (i.e., seamless integration of coupling the motion-features and semantics features of locations, in order to evaluate how (dis)similar are two symbolic trajectories (ongoing...)); and also investigate what would be the consequences of "infusing it" in trajectories warehouses (Alejandro Vaisman thought it was an interesting idea...) and how it would change the semantics/answer-sets of some traditional queries.
Most recently, some of the works have targeted the problem of efficient/effective maintenance of differential privacy (with Gabriel Ghinita and Mihai Maruseac) and a novel spatio-temporal query -- a continuous variant of the good-old-MaxRS (with Mas-ud Hussain).
Below are some specific problems and the corresponding sample-publications:
- Similarity of Motion Under Ridig Transformation
- Efficiency of Similarity Evaluation for Large Datasets of Trajectories
- Experimental Comparison of Methods and Distance Functions
- Similarity in Time-Series, an Extended Experimental Comparison
- Similarity of Motion applied to Mobile Devices Intrusion Detection
- You can read a brief description of the EDBT 2012 tutorial here (with Dimitris Gunopoulos). The slides are available at the official website of the conference...
- A recent survey on clustering time-series/trajectories
- Some most-current results related to adding the semantic-awareness to warehousing symbolic trajectories are summarized in this paper (appeared at ACM IWGS 2015) and here you can find a zip-ed folder containing the source-codes and the datasets (developed by Besim Avci; data generated by Di Tian and Tian Zhang) used in the experimental evaluations of the benefits of the proposed methodologies (NOTE: the zip-ed version of the samples of the trajectories data is ~0.5GB; please read the README_and_DISCLAIMER.txt file).
- Intrusion detection for mobile devices
- Differential Privacy for crowd-sourced environmental sensing
- Continuous Maximal Range-Sum (Co-MaxRS) for trajectories
Teaching
Research
Publications
Students
