Google I/O 2019 keynotes highlights

Google I/O event was held May 7-9 at California. For those who don't know, I/O is an annual developer conference by Google that showcases technical in-depth sessions on various google services, gives insight on what's new to look for in the following year. Here are some note worthy points from Google I/O 2019.

The key point they really focused on was user trust and privacy in providing helpful google services consisting knowledge, success, health and happiness for everyone.

Google Search:

The search for which google is known for since the start now provides full coverage of different types of stories, different aspects of searched story with catchy timeline of events happened related to the story.

Search will also include Podcasts, and you can find them based on their content not just title. All of this is possible by combining Machine Learning with the classic search.

Camera Capability:

The assistant, photos, camera is used more than a billion times according to their stats. This year they've enhanced camera based utilities for variety of use cases.

Camera and AR to deliver visual information in google search: You can search for a 3D model and place it in your space and experience the VR which is available on your fingertips. For example, search for a shoe and see if they're matching with your clothes or not. They also showed a cool live demo of great white shark on stage.

Google lens and Maps : Now we are able to find popular dishes of any restaurant using the lens. Look at the delicious looking pictures without entering the place itself. Food menu is connected with information on google maps. Functionalities like split the total and scan the bill can be done with the lens. Recipes are converted to videos; just point the lens on the text. So much delight for the always hungry foodie! :P

Google go : It is designed to enable search on entry level devices. The app is meant to be really light weight and fast. You can translate and listen text to speech just by capturing the text. The booming tech like computer vision, natural language translation and search come together to achieve such class. It works on more than a dozen languages.

Truly google is for ALL!

Google Assistant:

Assistant is based on AI and is under constant development for 3 years now. Last year Google Duplex actually behaving as assistant booking your appointments was the best moment from I/O event. After receiving great feedback this year they introduced duplex on the web. Available in 40 states of U.S, it can handle from restaurants reservation to booking movie tickets to renting car on your behalf.

New pixel phones now come with next gen assistant which can operate multi-task and compose mails for you. Multiple apps continuous commands is finally here. No more 'ok google' every time. It is also made efficient by reducing the size of the model from 100 GB to 0.5 GB.

Some little pleasures like ability to edit your preferences, saying 'hey google lets drive' to use assistance with maps in driving mode, assistant attending call while giving you call transcript and reply recommendation on screen, ability to *snooze/stop alarms* and reminders is just cherry on the cake.

AI for Everyone:

Users and developers both can take a sigh of relief after knowing about google's dedication for the data privacy and their innovation to implement such system that support the AI for everyone.

For privacy - Google takeout, incognito mode in browsers, auto delete feature for your data, one tap access to google account, incognito mode in maps, safe browsing, phishing prevention in gmail and two step verification are introduced and improvised.

For security - Federated learning a method to train AI models without collecting data from device. The model is trained on device and only the updates are transferred back to the global model on cloud. This is what every AI service needs to employ and this is what data scientists like me look forward to research about.

Accessibility:

Google has always come forward to include everyone despite of their special abilities to be able to use its services. They introduced Live transcribe, live caption on video, Live relay on call for the same. Read more news.

Their project euphonia is doing research and development for providing access to ALS and similar disease affected people to voice based interfaces.

Android 10:

Android receiving 26/30 security rating from Gartner and 2.5B active android devices, the news and details about android 10 is already out there. I will just point down some features which they announced.

1. Focus on security and privacy(multiple settings options).
2. Android for foldable smartphones supporting seeming transition.
3. 5G connectivity.
4. On-device model learning.
5. Live caption - available offline.
6. Smart reply in all messaging apps along with maps.
7. Dark theme.
8. Location sharing notified for each application.
9. Regular security updates.
10. Digital Wellbeing improvised with focus mode, parental controls with little features like bonus time for kids.

Google Home:

Is now Nest - feels more real. All the google home services and devices comes under this brand name now. They have made it easy, personal, allowing multiple users at the same time respecting your privacy.

Hub max is available with features like face match for individual profiles, smart display, smart home controller, kitchen tv, digital photo frame, indoor camera and video calls. That little 'pause' hand gesture brought smile to everybody's face.

Pixel:

Google is making pixel available for everyone by overcoming the challenge of providing efficient AI and software on the affordable hardware. They introduced pixel 3a and pixel 3axl with premium features with performance at as low as 399$. Comes with AR in Google map, 30 hours on a single charge, 7 hours battery time in 15 min charge, 3 years warranty, high built in security with the useful smart reply and undo sent mail feature in gmail.

The last one is a bonus highlight for AI practitioners and Data scientists:

TCAV - to detect features responsible for prediction bias.

The interpretation of deep learning model is a challenge due to their size, complexity, and often opaque internal state. To understand the bias and how some features affect the prediction google research team has introduced Concept Activation Vectors(CAVs). Read more in the paper authored by Been Kim et al.

Thank you reader. Until next year's google I/O, keep enjoying google services and their new updates. I have covered the event in past years as well, if you wish to compare google's progress follow below links:

Google I/O 2018

Google I/O 2017