Connected Devices Weekly Update/2016-09-29

• Team is working to define product and hardware plan by Oct 1.
• Performed initial Legal briefing and review
• Ran and completed Project SensorWeb Digital Ads Test in Taipei
  • 7-Days of Data, Facebook Ads, Google Ads, Search/Keywords
• UX on track to provide by EOW wireframes out of box user journey by Friday, Sept 30
• Team preparing deck for Tues, Oct 4th Gate 1 Pitch to PIB.
• Project Goals:
  • Evidence from the experiments to show the consumer needs. (consumer service)
  • Defined product (kit) to be prototyped by partner (hardware solution)

Working on user study for new product idea

Progress Report

• Prototype
  • Prototype under development.
  • Using FxOS on a Z3C phone for prototype
  • 3d printing an enclosure for the phone and speaker

• User Survey
  • User recruitment in progress with external user recruiter support
  • SurveyGizmo survey in progress with Panel support

• Timeline
  • Participants recruited - Oct 5th
  • Prototype Finished - Oct 7th
  • User Study - Oct 12th - 16th
  • Synthesis/Findings - Oct 19th

Platform

• Team reviewing the common blocks & frameworks needed for each "track-1" product as they get defined/scoped.
• Assessing options (& pros/cons) for OTA framework, security policies, build systems etc. Keeping generic to benefit all products & be a complete offering.
• Weekly Notes: https://docs.google.com/document/d/1yo1AtuozukgfkZwzO745LdEsB1fdZrEWoIYr8F9eydo/edit#heading=h.qgtmzupjdx35

Deep Speech

• Re-wrote deep learning training algorithm (Adam optimization) to work across multiple GPU's
• Integrated CPU based CTC algorithm into the neural network
• Benchmarked GPU based CTC algorithm vs CPU based CTC algorithm looking at execution time and validation error
  • GPU algorithm much better - less validation error, less time variance and we don’t have to move data between CPU and GPU memory.
• Wrote code to periodically serialize the model to disk while training to allow for archiving performant models
• Started writing the multi-threaded data loading code for the 300 hour TED data corpus

• General agreement in the team to focus on a software Dev Kit first.
  • Part 1 - Start cataloguing a “hub” of components & utilities that we already have on deck within CD teams (see examples in link below).
  • Part 2 - Add product specific developer framework components as we learn about each track-1 product's developer involvement.
• Expect 1st devkit (for sensorweb & other products) around Feb'2017. Schedule of milestones being built.
• Weekly Notes: https://docs.google.com/document/d/1BvXPyCRWZxVpi0fpdhOXdkdDti32jXK68a8IJUWnJ7o/edit?ts=57e448b7#heading=h.oxrerlr2zojn
• Weekly Meeting - Tuesday 9AM Pacific. CD Vidyo Room.

• This week Project Abigail passed Gate 0.

• Team held a kick-off meeting to plan activities around hypothesis testing. Discussion was held around what use case(s) would be most appropriate for user testing.
• Team discussed lessons learned from previous prototype testing. Re-use of code already developed by Vaani and Cue teams is being considered. Discussed options for how team might develop a hardware prototype that would maximize the amount of feedback collected yet be the simplest and quickest to put into homes.
• The use of a video is planned to introduce the product vision and full capabilities of Abigail to gauge interest and willingness to purchase.
• Prototype development is expected during the month of October, with devices tested in homes and feedback collected during November.

View Source experiment

• Final version of the View Source Research insights.
  • Some answers this week but not change in the conclusions
  • Shared with the View Source team and planned a debrief meeting with them next week. They are interested in repeat the experience for two product days (~50-100 ppl events) they have scheduled,

Product Definition

Vision

• Democratize story telling

Why?

• Deploying digital content into physical spaces requires excessive time, money and expertise.
• Today nearby content is siloed, requiring a different app for each experience. A single open platform is more convenient for users and accessible to all content owners.

For whom?

• An easy and affordable way for any storyteller to deploy content to physical locations for others to discover.
• A single immersive experience for curious and inquisitive learners to discover the hidden stories behind their surroundings, wherever they happen to be.

Team's focused on:

• Analyze the Competition Landscape for the Client, Service and Beacons
• Meetings with Google Physical Web team, Mozilla Campus Club (as there is a lot of interest in our work) and other possible users of our product
• Working week in MTV next week to work on the deck
• Gate 0 pitch scheduled 12th October

Problem Hypothesis

"We believe that early technology adopters have a desire to monitor their home over the Internet so that they can have peace of mind that their home, family, pets and belongings are safe."

Solution Hypothesis

"We believe that for early adopters who are currently using analog or cloud-based home monitoring solutions with a recurring subscription, if we create a simple web based system hosted in the home itself we can provide a more private, secure, cost-effective and future-proof alternative."

• Ramping up on market and technical research
• Team meeting all next week to refine product definition in time for gate 0 pitch on October 12th

• Good progress has been made on the product definition.
• Team has been evaluating technology of different types. Challenge is to keep the experience very similar to playing physical board games yet enhance the experience enough to make a compelling case for the additional cost the technology will add.
• Use case development is well underway, with several technology enhanced experiences having been identified for simple children's games that involve the use of a spinner. A prototype spinner that uses bluetooth has been developed.
• Team is in the process of putting together a gate 0 pitch deck and is collecting feedback on it.

• This week Project Lighthouse passed Gate 0
• The team is proceeding with testing hypotheses that 1) people who are blind would trust and actually use new object identification technology to increase their independence and 2) it's technically feasible to build an accurate image recognition solution with voice identification.
• There are two key problems that the team needs to solve: Image acquisition and image matching
  • Difficulty in getting a good image
  • Difficulty in separating the image from the background
  • Can a portable solution that's easy to use be developed
  • Can a reference image be matched to a new image taken of an object to identify it successfully
• Team will spend a week on feasibility testing for a solution that could be more mobile (enable use outside the home) but will move on to a simpler but less flexible solution (home based) as a backup plan for the prototype
• Goal is to complete hypothesis and prototype testing by end of November

• Work on the User Journey and filled the Mozilla IoT Product definition

• No Update

• Smart Home
  • Apple is rumoured to be prototype testing its own Amazon Echo competitor.
  • Google's Home smart home hub is rumoured to be $50 cheaper than the Echo when it hits the market this fall.
  • The Alexa Fund has led a $5.6 million Series A investment round in Nucleus, a smart home intercom which works with Alexa. The Alexa Fund provides up to $100 million in venture capital funding to fuel voice technology innovation.
  • Amazon’s voice recognition platform, Alexa, has reached 3,000 skills. The platform has been available for a little over a year in the U.S., providing developers with a way to integrate their services and smart home devices into the Echo, Dot, Tap, and Fire TV. Amazon also announced new skills including Food Network and GE Appliances available now, with Yahoo Sports Fantasy Football, Bloomberg, and What to Expect When Expecting all coming by the end of this month.

• Artificial Intelligence
  • Google released its v3 of its AI image captioning system in TensorFlow. The new version is much faster to train, and produces more detailed and accurate descriptions compared to the original system - it achieves 93.9% accuracy on the ImageNet classification task.
  • Apple acquired an Indian start-up focused on AI, its third buy in this space this year.
  • Facebook and Intel won first place in 2 categories of VizDoom, an AI challenge to play the video game Doom the same way that humans do. Just like Google DeepMind's successes playing boardgame Go, VizDoom is an accessible way to demonstrate to the world the nascent possibilities of artificial intelligence tech.