|09:00 – 10:00
||Keynote Talk #1: Myra Spiliopoulou – Learning from Multiple Correlated Sensor Signals in Medical Research Applications
Intelligent sensor technology is widely used in healthcare. One goal is to assist patients with chronical diseases, e.g. by monitoring activity, recognizing emergencies and raising alerts. Before any use in healthcare, sensor technology is first tested in medical research, where the goals are to contribute to better diagnostics, to assess the effectiveness of some treatment or to design prevention measures. In this talk, I discuss examples of sensor mining for diagnostics and prevention.
The first example is on analyzing tumor enhancement kinetics in Dynamic Contrast-Enhanced Magnetic Resonance Images. Here, the goal of sensor mining is to distinguish between benign and malignant breast tumors. The relative enhancement curves differ, depending on whether a voxel belongs to a malignant region or not. However, the curve of any single voxel is not adequate to decide on malignancy. Signals from proximal voxels are evidently correlated, and this fact can (and should) be exploited to identify regions that exhibit similar enhancement curves.
The second example is on understanding how patients with diabetic foot syndrome apply plantar pressure. Medical research has shown that the likelihood of foot amputation among patients with diabetic foot syndrom is up to 40 times higher than among non-diabetics, that increased foot temperature may indicate the onset of an ulceration, and that plantar pressure modulates temperature. Intelligent wearables thus monitor pressure in different regions of the feet. Here, a goal of sensor mining is to derive pressure profiles, exploiting the correlation among proximal sensors, and to optimize the number and placement of sensors in the wearable.
|10:30 – 11:30
||Session I – Sensory Data Analysis – Chair: Ashfaqur Rahman
• Predicting Phone Usage Behaviors with Sensory Data using Hierarchical Generative Model, by Chuankai An and Dan Rockmore
• Learning Multi-faceted Activities from Heterogeneous Data with the Product Space Hierarchical Dirichlet Processes, by Thanh Binh Nguyen, Vu Nguyen, Svetha Venkatesh and Dinh Phungy
|11:30 – 12:30
||Session II – Clustering and Applications – Chair: Yuan Jiang
• Image Segmentation With Superpixel Based Covariance Descriptor, by Xianbin Gu and Martin Purvis
• Phishing Detection on Twitter Streams using Unsupervised Learning, by Se Yeong Jeong, Yun Sing Koh and Gill Dobbie
• Rigidly Self-Expressive Sparse Subspace Clustering, by Linbo Qiao, Bo-Feng Zhang, Shiqian Ma and Jinshu Su
|14:30 – 15:10
||Session III – Action Recognition – Chair: Jeremiah Deng
• A Comparative Evaluation of Action Recognition Methods via Riemannian Manifolds, Fisher Vectors and GMMs: Ideal and Challenging Conditions, by Johanna Carvajal, Arnold Wiliem, Chris McCool, Brian Lovell and Conrad Sanderson
• Joint Recognition and Segmentation of Actions via Probabilistic Integration of Spatio-Temporal Fisher Vectors, by Johanna Carvajal, Chris McCool, Brian Lovell and Conrad Sanderson