Machine Learning and Social Protocols for Enhancing Spectrum Access for Wireless Communications
Goal: Enhancing access to RADIO spectrum by taking advantage of human behavior
This interdisciplinary project applies and develops expertise from areas of social computing, machine learning, wireless technology, security engineering, physical analogs, mobile systems, and user-centered design. We study human social protocols that are cooperative yet discretionary methods that allow to users distribute access more fairly using inherently natural decision-making processes as opposed to externally imposed ones.
This project may lead to substantial societal impact by allowing more work to be done with the same resources. Success will also benefit the environment by limiting the infrastructure needed for the required data traffic; both energy and infrastructure investment can be minimized.
When to interrupt users?
We have studied the timing of messages towards the end of implementing effective social protocols. This is based on the hypothesis that people would be more receptive to cooperate with their wireless access if they are asked to cooperate at an opportune moment. We collected data from participants in during their daily lives, designed, implemented and evaluated machine learning models to predict the opportune times to send them messages (Interruptibility). This work led to a tier-1 publication published in proceedings of CHI’17.
1) We proposed a two-stage hierarchical interruptibility prediction model. In the first stage, our model predicts (with 75% accuracy) whether a user will react to an interruption or notification based on mobile sensor data and personality traits. If the user reacts, it further predicts the user’s interruptibility intensity for various tasks (requiring user involvement) in the second stage based on mobile sensor data and user’s self-reported contextual information.
2) The evaluation results showed that our model can achieve an overall accuracy of 66% for interruptibility intensity prediction (with 61% mean accuracy).
3) We were the first to introduce people’s personality into an interruptibility prediction model. On average, it improves the major measures (accuracy, precision, recall and F- measure) of tested classifiers over 10 percentage points in the first stage.
4) Our model solves the initial prediction problem, that is, how to predict when you do not have user data. To achieve this, in the second stage, our model uses the data of people who share similar personality with the user. Compared to the models using all the data of other people, this reduces the training time significantly while maintains comparable prediction accuracy.
5) We implemented a smartphone platform for this study. We collected over 5000 interruptibility records from 22 participants over four weeks.
Xianyi Gao, Gradeigh D. Clark, and Janne Lindqvist. "Of Two Minds, Multiple Addresses, and One Ledger: Characterizing Opinions, Knowledge, and Perceptions of Bitcoin Across Users and Non-Users," CHI'16 Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems, 2016. doi:10.1145/2858036.2858049
Fengpeng Yuan, Xianyi Gao, and Janne Lindqvist. "How Busy Are You?: Predicting the Interruptibility Intensity of Mobile Users," Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (CHI '17). ACM, New York, NY, USA., 2017. doi:https://doi.org/10.1145/3025453.3025946
Gradeigh D. Clark, Swapnil Sarode, and Janne Lindqvist. "No time at all: opportunity cost of Android permissions (invited paper)," Proceedings of the 3rd Workshop on Hot Topics in Wireless (HotWireless '16). ACM, New York, NY, USA,, 2016. doi:http://dx.doi.org/10.1145/2980115.2980117
Janne Lindqvist, PI
Richard Howard, staff scientist
Meghan McLean, research manager
Graduate students: Hua Deng, Demetrios Lambropolous, Can Liu, Christos Mitropolous
Undergraduate students: Asmaa Ahmed, Jasmine Elbana, Zachary Foley, Hyewoon Jeong, Jocelyn Macaraeg, Natalie Shultis, Gali Zaborowski
This material is based upon work supported by the National Science Foundation under Grant Number 1546689. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.