In conjunction with IEEE BigMM 2016
Big Data Small Footprint for Detecting Transportation Modes
As the “biggest big data,” multimedia big data introduces many technological challenges, including compression, storage, transmission, analysis, and recognition. Each year, the BigMM conference organization committee organizes an algorithmic competition to address one grand challenge in the field of multimedia big data, which is open to all tool vendors, academics and corporations. The major aim is to recognize the most innovative, efficient and methodologically advanced BigMM tools.
The BigMM 2016 Challenge is to address the problem of Big Data Small Footprint for Detecting Transportation Modes. Detecting transportation modes of a user is a critical subroutine of many mobile applications. The detected mode can be used to infer the user’s state to perform context-aware computing. However, sensors on mobile devices bring forth a couple of new challenges to big data research. First, the power consumption for analyzing and classifying sensor data must be low, since most wearables and portable devices are power-strapped. Second, the velocity of analyzing big data on these devices must be high; otherwise the limited local storage may overflow. Tackling these challenging problems has both scientific and practical importance.
The topic of the BigMM 2016 Challenge is to predict transportation modes through analyzing signals from motion sensors (accelerometer and gyroscope). The contestants are asked to develop algorithms to predict a person’s transportation modes including still, walking, running, biking, and on a vehicle. A contesting system will focus on addressing the big-data small-footprint requirement by designing a classifier that is high in prediction accuracy and low in both computational complexity and memory requirement.
For the consideration of small footprint limitations and the real use case, the system needs to meet following requirements:
- Prediction latency for an instance does not exceed 8 seconds.
- The execution file of the system does not exceed 16KB.
- The power consumption should be low. Note that gyroscope consumes 6mA and accelerometer only consumes 0.1mA at 30Hz sampling rate.
- Memory usage (15%)
- Power consumption and computation time to predict an instance (25%)
- Prediction accuracy of the five modes: still, walking, running, biking, and on a vehicle (60%)
In this challenge, a transportation mode dataset collected by HTC is available for offline training and a dataset is available for testing. Detailed descriptions of the dataset can be found at the web site of this challenge. In addition, the organizer will make available a web service accessible from the web site of this challenge for online test runs three months before the final submission deadline. Each contestant can enter the URL of the web service to upload the results. Upon receiving your results, the Challenge web site will schedule a job to call the web service, evaluate the results and post the result obtained in the “Team” and “Leaderboard” sections of the web site.
It should be noted that in order to be eligible to receive the data, each contestant must first apply to participate in the BigMM 2016 Challenge. The application will be acknowledged by the organizing committee and receive a team ID, an active participant’s password, and instructions about how to obtain the data. To avoid any issue related to copyright, each contestant will be asked to email a scanned image of “Data Use Agreements and Distribution” to the organizing committee.
The dataset consists of training data and test data. Details about the dataset are as follows:
The transportation status includes ten modes: still, walking, running, biking, (riding) motorcycle, car, bus, metro, train, and high speed rail (HSR). For this competition, please combine modes of motorcycle, car, bus, metro, train, and HSR into the vehicle mode.
|Transportation Mode||Collection Time(hour)|
Each of the top three teams receives a latest HTC device.
Meng-Chieh Yu, Tong Yu, Shao-Chen Wang, Chih-Jen Lin, and Edward Y. Chang. 2014. Big data small footprint: the design of a low-power classifier for detecting transportation modes. Proc. VLDB Endow. 7, 13 (August 2014), 1429-1440.