Google I/O 2017 day 3

It's the third and final day of Google I/O 2017. Having observed and absorbed the mood over a couple of days, and with temperatures reaching some 30+ degrees Celcius, I'm more inclined to refer to the event as a developer festival than a conference. Thankfully Google provides plenty of cold drinks, sun lotion and shaded areas just as the tents are well (some would say too well) air conditioned. Cognitive fatigue and saturation is slowly but surely starting to set in, here on the third day of intense technical stimulation.

Thankfully, the laid-back lab and demo tents, called Sandboxes, allowed for some physical experimentation and admiration - like the 180 Teraflop (that's 180000000000000 calculations per second) custom TPU processor which is doing the heavy lifting of Google's increasing use of AI and machine-learning. I could not get anyone from Google to disclose who is manufacturing this beast, but I'm going to guess it's Nvidia. Google is giving away 1000 of these to universities for educational purposes, but otherwise the only way to get your code to run on one of these, will be to use the appropriate Google Cloud services.

The themes of the talks are much the same as the day before, and one is once again left with the classic problem of not being able to attend every technical talk you really want to. To put a big fat line under Google's move to "AI first", machine-learning snug into all kind of talks you otherwise wouldn't really expect it to - such as the talk about Android location API and Sensor Hub. Google continues to push for lower battery consumption and higher location precision, even if that sounds contradictive to common sense. These optimisations are achieved by using dual-frequency GNSS and a whole slew of low-level tricks involving machine-learning algorithms. Thankfully, to the developer, not much is changing except for some policy tweaking regarding location aware processes in the background. It's clearly Google's goal to let their so-called "fused" location provider be able to provide best-of-breed always-on navigation both indoor and outdoor.

Staying with the location theme, the fused location API is now also able to detect whether you are weight lifting, traveling by train and other actions and transitions, all in an effort to drive more context-aware apps.

We don't leave the machine-learning theme just yet, because Google also announced TensorFlow Lite through the Android Neural Network API. The API is suppose to bring machine-learning directly onto the handheld device by using a slimmed down version of TensorFlow with focus on running models (prediction) rather than learning (training) - which as you may know, is really compute intensive. Google envisions hardware support for this becoming mainstream (akin to their own server TPU's) but until then, the ANN API is exposing a uniform interface around TPU, GPU, CPU, DSP and other SoC components. You can see this talk here.

There were also dedicated talks about the Kotlin language, which isn't surprising given how Google is now embracing it head on for Android development. I have seen talks about Kotlin before, but Hadi Hariri really did a great job showing just how verbose and clumsy Java is. You can watch his introductory talk here and if you are warming up to Kotlin on Android, as I admittedly am, you should also watch this more advanced talk with Android rockstars Jake Wharton and Christina Lee.

With the day coming to an end, it was also time to try out one of the Code Labs. This is a large space setup with preconfigured OSX images full of samples and exercises. This was not personally a focus of mine, but it was a powerful demonstration of Google's due diligence and organisational skills. You may be interested to know that you can access it all, even as a non-attendee, by visiting this link.

After having recovered from information overload and jetlag, I will probably look into these exercises. With that, Google I/O is a wrap and I must take my hat off to Google for providing such a rich venue and experience - if I could I would certainly come back to this Google festival another time.