PERVASIVE 2009 report

These are some raw notes that I took during PERVASIVE 2009, held in Nara, Japan.

TOSHIO IWAI presented a series of art+technology artifacts. His talk was titled: “Expanding Media Art from Flipbooks to TENORI-ON”. His work was based on Afterimages, namely the impression that an high-contrast image leaves on the retina. His work was deeply influenced by toys. Once, he was playing with The music box, which uses a piece of paper with holes to activate some drums. Toschio played “happy birthday” and then its inverse. The inverse was actually more interesting than the original and therefore he develop the idea that some visual patterns could be translated into melody. From this intuition, he developed TENORI-ON: a matrix of buttons and leds that can be used to generate a cellular automata. Positions in the matrix are associated with tones and therefore produce melodies.

During the first session of the conference, Jorg Muller and Florian Alt presented research on context-aware displays. Jorg equipped displays with cameras to understand who was paying attention to the ads displayed in the displays. They did not find a correlation between what people like and what they look at. Some participants actively looked away to tell the system they did not like the content being displayed. Florian presented research on “Users’ view on car advertisement”. Roof-mounted displays in US are location-aware. They asked which part of the car would they like to devote to advertisement. They asked particular form factor for the interactive display. They conducted an online survey. Their paper offers a number of technological implication on the design of car advertisement displays. They implemented a prototype for car advertisement.

JOHN KRUMM presented a study on “Realistic driving trips for location privacy”. A consistent amount of this work consisted in building a realistic simulation of trips to test out the system. His focus was on making false trips look more realistic.

DAGMAR KERN presented an interesting prototype for “Enhancing navigation information with tactile feedback in the steering wheel“. Their motivation is that the audio channel is often cluttered by many other information (e.g., chatting with another passenger, listening to the radio). Therefore multimodality can help. They used a vibro-tactile feedback system.

SOMAYA BEN ALLOUCH presented research on “Perceived benefits for using ambient technologies“. They used a path diagram and a validation technique similar to Structural Equation Modeling. They used a large-scale questionnaire to validate the adoption of a technology that is not yet on the market.

STEPHEN INTILLE presented a paper titled: “Adding GPS-Control to Traditional Thermostats: An experiment of Potential …”. Their starting assumption is that 50% of the thermostats is not effective. Furthermore, in the US 53% of the owner do not lower temperature during the day. They proposed a Context aware power management (CAPM) [Harris and Cahill 2005]. They proposed a context-aware thermostat that was informed of when the user was leaving home and coming back from work. The system used GPS signals detected with the mobile phone of the user. They tested with few families. The placed the system in their homes and placed a GPS tracker in their cars. The system switches the heating off when they leave and on when they commute back home. They simulated the saving (about 100 dollars for 70 days???). This study raises the discussion on whether we should talk about temperature or comfort.

JAMES SCOTT presented a new interaction modality with Mobile Devices using Force Sensing. The idea is to introduce force sensing as an input modality on mobile devices. The benefits are multiple: using unused parts of the device; avoid cluttering of the device; useful in combination with other inputs. They used bending motions to control the display of documents. Twist was used to turn web pages. Bending back/forward was used to emulate page up/down and stretching was used to emulate alt-tab. They conducted a user study to understand whether users could manage to use this modality for operating the device. They designed a task with some targets that needed to be used using one single modality. They needed to hold the selector on the target for 2 second. They found out that smal forces are not necessarily easier to apply and control. Force sensing seemed to be easy to learn and use.

JEFFREY HIGHTOWER presented a super interesting paper where they tried to Infer Identity of the user using Accelerometers in Television Remote Control. The motivation is that if you know who is watching then you can customize recommendations. This is not biometric authentication but more as biometric inference.Do people seem to use remote controls in physically different ways? They started with observations and notices that people hold their remote control in different ways. They used a classifier to infer who whas using the remote. they had 156 sensor features (3 windows x 4 lenghts x …). They tested using weka software. They used a decision-tree classifier. In a second attempt they tried to understand who was using by using acceleration on the device. They defined a session, which consist in the period of time between picking up the remote and setting it back. They got more accurate results. The most important features are: unique botton codes and sequences, button press rate, hand shake, angle the remote is held. There is a good literature review including commercial recognition, body worn accellerometers, etc.

STACEY KUZNETSOV focused on Human memory. The proposed an haptic bracelet that gives a specific vibration every time a similar concept is encountered. They used the device to support situation of blending where the meaning of two concept gets confused. The second study focused on auditory recognition. The paper present a controlled experiment. The haptic cues seemed to be helping recall but not so much free recall in low performer. Some participants found cues as distractive. This can be used as memory aid for people with memory problem.

JEFFREY HIGHTOWER presented a research titled: “Exploring Privacy Concerns wih Personal Sensing“. They basically conducted an in-depth study on privacy perception with the non technical participants that participated in the UbiFit study presented at CHI’08. They did not feel that their privacy was exposed by the data that was shared in the ubifit study. Acceptability of sensible data depends on the use you do of that data. Participants were not in favour of using raw audio even if it was filtered. Professionsl context can make audio recording unacceptable. Try to design core functionality with only minimally invasive sensors. When invasive sensors are usedm filtering or purging data may increase acceptability.

SHIN’ICHI KONOMI presented ASKUS: Amplifying global actions. They conducted a study on Location Based Services trying to extend the study of information needs to action needs. They conducted a good literature review and a diary study. They designed ASKUS, a platform that can help people to ask favour to remote strangers. The field trial revealed several variables that can have an impact on the deployment of LBS services: like awareness and accountability, cost and motivation.

ERICH TUNTEBECK presented an interesting approach for Detecting batteryless Tags Through the Power Lines in a Building (Pl-Tags). Motivation: reducing the need for infrastructure. A broad class of devices produces powerline voltage transients. Transients are radiated from the power-line as broadband RF energy. The author show that basically without any radio receiver, the presence of a RDidtag near the powerline is detected any time a device produces a transient. Application: detecting who activated what electrical device. Detecting the presence or avsence of objects such as medicine cabinet, batteryless sensor.

ANMOL SHETH presented a paper titled: “Geo-Fencing: Confining WiFi in physical places“. Flexible Acess Control is Challenging (e.g., provide wifi access only to the patrins of the cafe). The GeoFencing approach creates a controlled overlap between multiple directional antenna patterns. Tie connectivity to the intersection of these ovelapping patterns. Antenna pattern is an iscoscele triangle with an radiation angle of 28 degrees. As a threshold they used the number of packets received. Less than 70% of the packets received makes most of the WiFi systems unusable.

NATHAN EAGLE presented a paper titled: “Methodologies for Continuous Cellular Tower Data Analysis“. Tower transitions into meaningful locations. We are good at analyzing small static graphs. What we are not so good at is in analyzing weighted large graphs with dynamic covariates and outcomes. This study studies IMMI Data collected with a number of phones that were instrumented with software that was collecting bluetooth scans, tower transitions, and sampling audio. They recruited 215 participants within Los Angeles. Time Series of Visible Cellular Towers. No information about tower location. This information can be represented as a graph. Nodes are towers and edges are the probabilities that two antennas can be seen at the same time. To segment this graph he used different algorithm. They used a bluetooth beacon to identify “home”. Then they segmented the graph and they trained a bayesian predictor to calculate the transition probabilities between different group of antennas. Nathan is trying to caraterize behavior with an entropy metric. Youngsters are much more entropic in their movements while senior staff is more regular in the mobility patterns.

STINA NYLANDER presented a paper titled: “It’s just easier with the phone: A diary study of Internet access fro m celll phones“. One week diary study with 19 participants. The average age was 30 years. 28% of news reading and 21% of email access. Other information 16% and travel and contact infor 15% (this comes from a Swedish service that provides this kind of information). They found some specific services used only on PC and some used only on mobile phones (e.g., microblogging service). News and email is read frequently because is usually available in small chuks which are frequently updated and it is good to pass time. They used the phone even if they were at home (31%) and with a computer access (51%). Most of the time when they were using the phone they were relazin (38%) and they were conducting home activities (18%). It is easier to avoid spending too much time with a phone. It supports mobility.

KAI KUNZE introduced a thoughtful quantitative study on context-aware information for supporting workers. The paper is titled: “Does context matter? A quantitative evaluation in a real world …”. Wearable computing. The question is that context awareness might not be beneficial. Is context useful in real life evanlaation? in tasks long/difficult enouh to need contextual help? with subjects that work on such tasks? They wanted to test context awareness with real tasks. The selection of the task took a long time. They used regular workers that did similar tasks every day. They used a wizard of oz software that they developed that they are planning to open source: They used a test training with lego where the technician had to use the head-mounted display with the goal of building the lego model. They found that context-aware system was the most efficient system to finish the task (time used as dependent variable). The fastest user with speech only. Context-aware system seems to help less proficient technicians.

BO BEGOLE presented a nice study “On the Anonymity of Home/Work Location Pairs“. Location traces are useful for personalized location-based services. However, they can be sensitive because they can reveal business connections, political afficiliation, medical conditiron, risky behaviors, etc.Some ways to mitigate the risk –> only my trusted network provider knows my location. Sometimes it is possible to infer identity by comgining data sources. 87% of US population have unique date of birth gender and postal code. With 2-week woth of GPS data collected you can infer person home location (Krumm, 2007). Obfuscation techniques does not prevent entirely reverse-engineering this data to disclose sensitive information. K-anonymity (Sweneey, 2000): data safe to release if at least k other people share the same data. What level of accuracy of location information could result in small sizes of k anonymity for some portion of the population. 7% of the US population live and work in a unique combination of postal codes. Approximate home and work location can result in a small anonymity set. If you combine this information with other information sources can reveal the identity of a person. Obfuscation is not necessarily providing anonymity.

AMY KARLSON presented a nice study of PC and phone transitions. The paper is titled: “Working Overtime: Paterns of Smartphone and PC usage in the Day on an Information Worker“. Workers are sourranded by a multitude of devices that often work in isolation. What opportunities exists for Mobile-PC interaction exists? They recuited 16 participants and installed a PC logger and a logger on the smartphone. They distinguished interaction patterns where workers were using their PC and their phone. Simple analysis of these patterns reveals that workers use the PC most of the time, and the mobile during the intervals and after work. However time to time they receive phone calls during the working hours. Besides time of use they were able to tell what ind of activity people were conducting. Using this information they were able to detect “handoff patterns” where participants started interaction on the phone and then they could move to the PC. They observed many situations where the phone was preferred but they did not observe many situations in which a task was started in a mobile phone and then handed over the PC. They concluded that people are not handing over tasks because there is not much support for that.

Leave a Reply