Towards Wearable Cognitive Assistance

Motivation: 
Many people in the US are suffering from cognitive decline which include Alzheimer's disease, stroke survivors and mild brain injury. Cognitive decline is the inability of recognising people, locations and objects etc.

Key Ideas/Contributions: 

This paper describes a complete system for cognitive assistance for users with cognitive decline. Their system was designed using Google Glasses.

Their primary design constraints include bounded responses, overhead and graceful degradation of offloading services.

The authors of the paper try to build a system to address some very important aspects of a cognitive assistance system such as functionality, usability, performance and customization while guaranteeing real time responses on battery and computing power constrained devices.

Gabriel is a prototype of their multi tiered assistance system
They employ VM based scalability model which guarantees low latencies by offloading expensive computations to cloudlets.
They mention that their architecture can support a wide variety of cognitive engines which are already out there making it is a flexible and pluggable architecture.
 The wrappers written over these engines are encapsulated within the cognitive VMs.

Others:

- This use case is both useful and in my opinion very important. The applications could extend to the older population who experience cognitive impairment. Any technologies which assist them are a good investment of effort and money.

- This paper addresses only applications/use cases which can be solved using camera sensors. Thus, only problems like face recognition, OCR, object and motion detection. I wanted to see if other sensors could be easily incorporated into the Gabrielle system. Sensors which would be able to support a variety of use cases for example, biosensors would have been truly unique system since I see glass as an extension of a mobile phone and don't necessarily find the system extremely unique when compared to the other mobile phone offloading systems. This system could maybe in theory, be extended to people with physical and other kind of impairments.

- However, their experiments do show the drawbacks of the current wearables. This was unsurprising but interesting to note.

- Alert based mechanisms weren't discussed in this paper. We should also consider providing safety mechanisms in such systems especially for the use cases mentioned by the paper. If the user is not using the technology for a prolonged period, should family or friends be alerted etc.
- In continuation, how would their react when there is network failure?

- Customization wasn't discussed in detail. How much can be tuned, what depends on the user's preference and interaction. Not all interactions have to be real time.

- Real prediction and user guidance algorithms might show different results since the reported results are run on simple versions.

- The paper detailed the systems performance in terms of overhead of VMs, scaling times, and even motivated the need for cloudlets very well.
But of course, the battery gain seen when offloading to a cloudlet which is predecided will be much better. Since cloudlets architectures are not yet a reality. We still really know how much time the routing and cloudlet selection takes? The success of this system depends on how the edge cloud and MAUI like computing architectures spans out.