Mobile User Behavior Modeling for Anomaly Detection and Behavior Prediction


Carnegie Mellon University - CyLab Mobility Research Center
Advisors : Anind K. Dey and Martin Griss
2011 - Present

We are developing a framework for modelling the behavior of a mobile user based on the rich set of information sources present on a mobile device, and using the models in the framework to detect anomalies in the user’s behavior.

The framework allows applications to leverage the anomaly detection mechanisms as service, and developers can extend the models present in the framework for custom scenarios.

We are prototyping the framework for the Android platform.


Hermes Framework

A Context-Aware Application Development Framework and Toolkit for the Mobile Environment.

Carnegie Mellon University - CyLab Mobility Research Center and Ericsson Research Silicon Valley
Advisors : Anind K. Dey and Martin Griss
2009 - 2012

Developed a software toolkit which provides a framework, including context inferencing, communication, storage, power management, security and intelligibility support, for developing more powerful context-aware applications for the modern mobile environment.

The Hermes toolkit and framework are designed around a loosely coupled component-based architecture that facilitates the decomposition of context-aware applications into multiple smaller components, each of which captures, transforms or aggregates pieces of context information to produce the high-level context used by applications. Implemented for the Android platform and Java.

Rapid Acoustic Model Training using GPUs for Speech Recognition

Rapid Acoustic Model Training using GPUs for Speech Recognition

Carnegie Mellon University
Advisors : Ian Lane and Jike Chong
2010 - 2012

Developed a parallel implementation of Viterbi training optimized for training Hidden-Markov-Model based acoustic models using highly parallel graphics processing units. Implemented in CUDA and C.

Using a single NVIDIA GTX580 GPU our system was shown to be 94.8x faster than a sequential CPU implementation, enabling a moderately sized acoustic model to be trained on 1000 hours of speech data in under 7 hours. Our implementation on a two-GPU system can perform 3.3x faster than a standard parallel reference implementation on a high-end 32-core Xeon server at 1/15th the cost.

Our GPU-based training platform lets research groups to rapidly evaluate new ideas and build accurate and robust acoustic models on very large training corpora at nominal cost.

Other Projects



A Hybrid Virtual-Physical Collaboration Environment

Carnegie Mellon University
Advisor : Joy Zhang

Developed a remote collaboration system to using a hybrid between the physical world and virtual worlds, where a group of people collaborate in the real world and others join them remotely via a virtual world.

HyPhIVE uses non-intrusive mobile sensors to detect real world users' collaboration context such as their position, direction of gaze, gestures and voice. HyPhIVE projects the sensed real world collaboration into a virtual world in a way that collaboration patterns are preserved. Remote users join the collaboration using virtual world clients and interact with other users' avatars.

Lunar Image Classification for Terrain Detection

Lunar Image Classification for Terrain Detection

Carnegie Mellon University and NASA Ames Research Center
Advisors : Joy Zhang and Ara V. Nefian

Investigated several image features and classifiers for lunar terrain classification, and extended a histogram of gradient orientation approach to discern the characteristics of various terrain types.

Experimental results showed that the proposed system achieves 95% accuracy of classification evaluated on a large dataset of 931 lunar image patches from NASA Apollo missions.


PMA - Personal Messaging Assistant

Carnegie Mellon University
Advisor : Martin Griss

Developed an advanced rule-based email management system which considers user context and message content. PMA uses separate scales of importance and urgency to prioritize emails and to decide on an appropriate action, such as SMS to user, defer to later, file or forward.

Context information is gathered from various sources including mobile phones, indoor and outdoor locationing systems, and calendars.

Real-Time Ray-Tracing on GPUs

Real-Time Ray-Tracing on GPUs

Undergraduate Research Lab - University of Peradeniya, Sri Lanka
Advisor : Manjula Sandirigama
2006 - 2007

Modified and implemented ray tracing algorithms to achieve real-time rendering speeds using the graphics processing unit as an alternative processor for ray-tracing calculations. Created an extension library for DirectX. Implemented in C++ and High-level shader language (HLSL).