3rd Panel Report
Applications of Memories for Life Research
The third panel looked at Applications of Memories for Life Research, with panellists Gareth Jones (Dublin City University), Hamish Cunningham (University of Sheffield), Ehud Reiter (University of Aberdeen) and Tim Thompson (i-Sho).
The panel discussed applications that have been implemented across several disciplines, including sensing and recording devices, biosensors and life-logging devices, as well as personal data organisation applications. Forgetting was seen as equally important as remembering.
The semantics of memory as a common structural abstraction for computer-based memory was discussed. The panel also predicted that richer models of statistical relevance beyond PageRank would be needed, and that workgroup memories may come to rely on semi-automatic semantic annotation.
The importance of how to explain data to different variations of people in a way that they can understand was emphasised (i.e. presenting semantics instead of raw data so that it can be useful instantly). Therefore data analysis and linguistic (graphical) communication were seen as important tools for memory-based applications.
Gareth Jones (download PowerPoint slides) from Dublin City University (standing in at short notice for his colleague Alan Smeaton) began by talking about Dublin City’s work using the SenseCam. He mentioned that they are looking at more ways to log memories by looking at ubiquitous techniques such as "smart material".
By looking at "smart materials", they are looking at ways of recording a person’s physical state. They can measure breathing, and whether the subject is moving their arms, and all this information can be used to makes guesses to what people are doing. The next thing on their agenda is to produce more advanced "Bio-sensors"; they would like to start mining and analysing people’s bodily fluids, PH values, and other similar forms of data, in order to start identifying people’s moods and other potentially interesting details.
Gareth is hopeful that technological advance will get us to the state where we can carry out ubiquitous life logging for the medical domain.
Hamish Cunningham, from the University of Sheffield, talked about his idea and dream of "Semantic Annotation Machines".
Hamish stressed the notion of capturing autobiographical metadata, envisaging a future of information structuring, with an abundance of Semantic Webs. A question he would like answered is: how can memory applications make it easier for people to identify, create, and exploit structures in artificial memories? How could we use these context-based annotations to aid tasks such as multimedia asset management?
Hamish then went on to describe the shift from databases and spreadsheets to semantic repositories. He presented a chronology showing how people are increasingly in-tune with "data-structures", and how people are getting better at using and exploiting such representations. He believes that these "new" semantics can be common structural abstractions, which could be shared across the population.
Hamish concluded by presenting some predictions about the importance and application of “Context and Structures” in the future of IT. He predicted that we will require richer models of statistical relevance beyond that of PageRank that has no notion of hyper-linking in personal memories. He wished to capture context to be used along with time to help us answer questions such as "what was that email I was reading when I was watching X?" or a world where del.icio.us data is associated with context. He also pointed out that users do not want to have to deal with creating the associations or defining their context, this needs to be done automatically.
His other prediction is related to "workgroup memories", that many systems will end up relying on semi-automatic semantic annotation; he gave the example of the Mechanical Turk system. He envisages a world where more, rich DIY structures are being created, using shared adopted controlled vocabularies, where automatic and manual hyper-linking is both supported and encouraged.
Ehud Reiter, from the University of Aberdeen, began by identifying the fact that a person’s memory does not only revolve around personal photographs or video data, it also includes what he called "technical data", i.e. medical, financial, legal, educational, and so on. Ehud presented a different view from the rest of the presenters, pointing out that it is often this "technical" data that is important, and that the M4L agenda should involve looking at ways of supporting and unifying this disperse information.
Ehud would like to investigate ways of explaining and passing this technical information to the right people, given the appropriate circumstances. How do you find ways of informing people about another person’s contextual information? How do you know when to make such information available? How do we inform people of all of the technical information stored about them?
An example given was that of scuba divers and how important their contextual information is. Divers all wear dive computers but kilobytes of raw numbers are not very useful. We need summaries, we need to look at methods are automatically summarising information so that it can be utilised in all walks of a persons life. In the scuba-diving example, the summaries should include information about your deepest dives, the most dangerous situations, and so on and so forth.
Ehud believes that the most important challenge faced by the M4L network is that of finding and identifying what is interesting and relevant in the swamp of personal data currently out there. He is also interested in identifying methods of communicating this information to those who do not have the expertise to exploit technical data, e.g. telling divers not to dive if it would be dangerous to do so.
Tim Thompson (download PowerPoint slides) from i-Sho, then went on to present their life-caching software. He described the system as a life-caching tool that allows its users to create "channels" which he described as a chronology of multimedia data that can be visualised and access via a timeline. The data can be organised, retrieved, tagged, hyperlinked, sorted, searched, and shared. This allows for people to have private, shared, or public channels. The system takes cues as visual mnemonics and exploits time as the vital contextual ingredient for organising and browsing through a personal multimedia library. A time-stream of multimedia items is presented allowing for communities to collaborate by sharing their data.
i-Sho attempts to get away from the "folder" structure apparent in current operating systems. The timeline is the "context" by which the system maybe navigated. The context can also be exploited in a graphical manner, for a user’s channel can be displayed alongside one of the public channels, allowing for the public channel to be used as a contextual cue. The system is presented as a “collaborative document authoring environment”, allowing users to trace and be aware of the provenance of any piece of multimedia on the timeline.
Questions:
There was much discussion about how could we get real people to give up their personal information so that it could be used for research. How do we acquire this personal data? After a number of comments the room came to the conclusion that it is hard and will be an issue facing all of us in the future, i.e. the collection of an initial data-set for applications is seen as a real problem.