5. Methodology

Compare the performance of XCS with an implementation of C4.5, a decision tree algorithm called J48, on the reminder generation task.

Exemplar from our context data-set:

Sensor value grouping : Sensor-Count, Sensor-All5, Sensor-Any5, Sensor-All1, Sensor-Any1 and Sensor-Immediate.

User1 – user name05.00 – appointment time 0, 0, 0, 0, 0, 0 – motion-sensor values 7, 0, 1, 0, 0, 0 – speech-sensor values20, 1, 1, 1, 1, 1 – process-1 sensor values20,1, 1, 1, 1, 1 - process-2 sensor values20, 1, 1, 1, 1, 1 - process-3 sensor values0, 0, 0, 0, 0, 0 - process-4 sensor values20, 1, 1, 1, 1, 1 - process-5 sensor values0, 0, 0, 0, 0, 0 – keyboard sensor values0, 0, 0, 0, 0, 0 – mouse sensor values3 – reminder type preferred by a user The same exemplar in a bit string representation for XCS

10001010101010101001110010110000100100001001101111010011011110100111111100000000000010011111110000000000000000000000:3

7. Discussions & Conclusions

2. Objectives• Apply machine learning

techniques to data gathered from simple context sensors to build improved human computer interfaces

• Use simple sensors to continuously gather data on a computer system’s internal and external environment

• Mine this context data for useful user-behavior patterns to better predict user preferences (behavior) and improve user interaction

• Apply XCS, a Genetics Based Machine Learning Technique to learn a mapping from user-related contextual features to reminder types

3. Our approach•Integrates approaches in context aware systems and data mining

•Considers a computer as a stationary robot with simple sensors for sensing the external and internal environments

•Builds user-interfaces on this stationary model of a computer

4. Sycophant•A simple calendaring application program that stores appointments and reminds a user using different reminder types

•Generates four types of reminders : visual (a pop-up window), speech (using a text-to-speech system), both, or neither

•Continuously gathers binary activity data from the keyboard, mouse, a motion detector, and a speech sensor; monitors the activity of five processes on the computer

•Generates a reminder and expects the user to indicate whether Sycophant used the correct reminder type

•Combine expert-generated rules with machine learned rules and use this combined knowledge to design better adaptive user interfaces

•Collect data from different users to scale up our research in the area of context learning applications and consider the possibility for personalization to individual users

Fig.1 Architecture

Fig.2 Sycophant – User Interface

Fig.3 XCS Architecture

Anil Shankar, Sushil J Louisanilk@cse.unr.edu sushil@cse.unr.edu

Evolutionary Computing Systems Lab (ECSL)Dept. of Computer Science and Engineering,

University of Nevada, Reno, USA

http://ecsl.cse.unr.edu

Acknowledgments: This work was supported in part by contract number N00014-0301-0104 from the Office of Naval Research.

6. Results

XCS outperforms J48 on the test sets

Current computer applications pay insufficient attention to a computer’s environment. These application programs lack context-awareness, and therefore they can only make weak attempts to adapt to individual user needs.

Our approach is to: •use simple sensors to collect data on a computer system’s internal and external environment. •mine this contextual information for useful user-behavior patterns to better predict user preferences (behavior) and improve user interaction

To substantiate our hypothesis, we have designed Sycophant, a context learning calendaring application program that learns a mapping from user-related contextual features to reminder actions

1. Motivation