November 2017 Summaries

The AIY Vision Kit, announced by Google's AIY Projects team, is a new do-it-yourself project that allows makers to add computer vision capabilities to their creations using a Raspberry Pi Zero W. This affordable and hackable intelligent camera kit includes a VisionBonnet featuring the Intel Movidius MA2450, which runs neural networks on-device, eliminating the need for a cloud connection. It comes with three pre-trained models for object recognition, face detection, and differentiating between cats, dogs, and people, and also allows users to integrate their own models with the included TensorFlow code. The kit is designed for customization, with a Python API for modifying camera reactions and GPIO pins for hardware extensions. The AIY Vision Kit, available for pre-sale at Micro Center, encourages users to solve real-world problems and share their innovative projects on social media.

The Poly API, launched by Google, provides developers with a powerful tool to enrich AR and VR applications by allowing dynamic searching and downloading of 3D assets. As a REST-based API, it supports cross-platform functionality, enabling developers to access and integrate 3D models into apps and games at both edit and run time. Google offers Poly Toolkits for popular game engines like Unity and Unreal, simplifying the process of importing and managing assets with plugins that handle API calls, asset conversion, and authentication. Additionally, sample codes and toolkits are available for Android, iOS, and web developers, facilitating seamless integration with platforms like ARCore and SceneKit, and supporting formats such as OBJ and GLTF. The API also features filtering options for assets based on keywords, categories, and complexity, making it easier for developers to find and utilize the right resources for their projects.

Google's Launchpad Accelerator has announced its fifth class, featuring an array of startups from around the globe, each aiming to address diverse challenges such as improving medical records in Africa and enhancing breast cancer screenings. This six-month program includes a two-week intensive bootcamp in San Francisco and offers mentorship from over 30 Google teams, alongside guidance from experts in top tech firms and VCs. Participants benefit from equity-free support, Google product credits, and media training, and they continue collaborating with Google after returning to their home countries. The upcoming session, starting on January 29, 2018, will host startups like Helium Health from Nigeria, NIRAMAI from India, and Nubank from Brazil, among others, providing them with all-expense-paid training as part of the program's comprehensive development efforts.

In November 2017, Google launched a YouTube video series aimed at helping developers understand best practices for creating applications on the Actions on Google platform. This initiative, led by Developer Advocate Ido Green, focuses on guiding developers through the process of building apps for the Google Assistant, which can be accessed via Google Home, Android, iPhones, and other future platforms. The series covers essential topics such as passing the review process, designing effective Voice UIs, and integrating server logic into Assistant apps, all intended to enhance user interaction and engagement. Developers are encouraged to join the ongoing community discussion to further their learning and development efforts.

Container-diff is an open-source tool developed by the Google Container Tools team to help developers identify changes between container images, thereby facilitating the development of containerized software. It addresses the challenge of understanding how changes in a Dockerfile affect the resulting container image, which can lead to bloated images and difficult-to-track issues. By computing semantic differences between images, container-diff provides insights into system and language-level package changes, as well as file modifications, in a readable format. The tool supports image analysis through various formats, including local Docker daemon, remote registry, and tarball files, and can operate independently of a Docker daemon. It proves useful in scenarios such as changelog generation and continuous integration and offers both human-readable and JSON outputs for automated processing. Container-diff also allows for single image analysis to provide detailed information about installed packages and their sizes, aiding in compatibility verification and space management. The team encourages community contributions to expand its capabilities, with potential future support for more languages and runtime environments.

The blog post delves into the concept of feature columns within TensorFlow, emphasizing their role in structuring input data for machine learning models. It explains that feature columns serve as a bridge between raw data and the model's required input, supporting both numerical and categorical data types. The post covers various types of feature columns, including numeric, bucketized, categorical identity, and vocabulary columns, as well as more complex types like feature crosses and embedding columns. Embedding columns, in particular, are highlighted for their ability to represent high-dimensional categorical data in a lower-dimensional space, enhancing a model's ability to learn relationships. The discussion also touches on the integration of feature columns into different TensorFlow Estimators, such as LinearClassifier and DNNClassifier, noting the specific requirements for each. Additional resources are provided for readers interested in exploring feature engineering and embeddings further, with links to tutorials, videos, and projects.

"Build Out" is a new YouTube series created by Google Developer Advocates Reto Meier and Colt McAnlis, aimed at demonstrating how to integrate various Google developer products to address complex user problems by building fictional products. Each episode features the presenters facing off with different architectures, employing a range of technologies such as Google Cloud, Android, Firebase, and TensorFlow, and going beyond the basic solutions to explore advanced possibilities. Viewers are encouraged to engage by commenting on the videos to decide which solution is more effective or by sharing their own ideas and solutions. The series begins with "The Smart Garden," where the challenge is to design autonomous, scalable gardens capable of learning and adapting, with technical details and alternative approaches available in accompanying Medium articles and podcasts.

Google has launched the Launchpad Africa program, an extension of its global Launchpad Accelerator, to nurture tech startups across Africa with a three-month accelerator, offering more than $3 million in equity-free support, workspace, travel, PR assistance, and access to expert advisors. This initiative follows a commitment made by Google's CEO Sundar Pichai to support African entrepreneurs in building successful tech companies. The program is initially open to startups based in Ghana, Kenya, Nigeria, South Africa, Tanzania, or Uganda, with future expansions anticipated. Eligible startups must be technology-based, have already secured seed funding, and address significant challenges in their local or broader African context. The selection process emphasizes the potential value creation for users and the willingness of startups to share their learnings for the benefit of their local ecosystems. Google is enthusiastic about contributing to the vibrant innovations emerging from Africa and fostering further technological advancement on the continent.

Google Assistant has introduced a series of updates aimed at enhancing user interaction with apps through improved discovery, engagement, and re-engagement features. These updates include an enhanced app directory with trending sections and autocomplete functionality, new subcategories for more precise app discovery, and the implementation of implicit discovery, allowing users to find apps via contextual queries. The introduction of a "For Families" badge will help users identify family-friendly apps, and international support has been expanded to include several additional languages. To create a more engaging user experience, features such as speaker-to-phone transfer, personalized app experiences, and advanced SSML audio design have been added. Developers can now utilize a cancel event for improved conversation closure and account linking at appropriate times during user interactions. For re-engagement, features like daily updates and a new push notification API are available, alongside enhanced directory analytics to help developers understand user interactions better. These updates are designed to make Google Assistant a more dynamic and versatile platform for both developers and users.

In November 2017, the TensorFlow team announced the developer preview of TensorFlow Lite, a streamlined version of TensorFlow designed for mobile and embedded devices. This new tool allows for low-latency inference of machine learning models on various platforms, including Android and iOS, and is optimized for fast initialization and small binary size. TensorFlow Lite supports hardware acceleration through the Android Neural Networks API, enabling efficient processing of machine learning workloads on custom hardware, while also providing a fallback to optimized CPU execution. The architecture includes components such as the TensorFlow Lite Converter and Interpreter, which reduce the model size significantly compared to TensorFlow Mobile. It initially supports models like MobileNet and Inception v3, with plans to expand functionality based on user needs. TensorFlow Lite is considered the evolution of TensorFlow Mobile, and though still in development, it aims to simplify the developer experience and broaden model deployment across devices.

The Grow with Google scholarship, in partnership with Udacity, offers 50,000 opportunities for individuals to enhance their skills as Android or mobile web developers, with applications closing on November 30th, 2017. Specifically designed for both novices and experienced programmers seeking a career change, the program provides free training tailored to career goals and prepares participants for Google's Associate Android Developer and Mobile Web Specialist certifications. The top 5,000 students from the scholarship challenge will receive a full Nanodegree scholarship to one of four specialized programs in Android or web development. This initiative utilizes a curriculum developed by experts from Google and Udacity, offering a valuable chance to advance or launch a career in mobile or web development.

Universal App campaigns (UAC) are becoming essential for app marketers as existing Google app promotion campaigns on Search, Display, and YouTube will cease operations on November 15th. UAC enables advertisers to reach target audiences across Google’s major platforms like Google Search, Google Play, YouTube, and the Google Display Network through a single campaign. Marketers using UAC have reported significant improvements, such as a 140% increase in conversions per dollar compared to other Google app promotion tools. For instance, the car-sharing service Maven from General Motors has achieved a 51% increase in monthly Android registrations through UAC. To maximize UAC effectiveness, marketers should create separate campaigns for distinct user types, optimize for valuable in-app actions, and utilize a diverse mix of creative assets. By adhering to these strategies, advertisers can enhance their app's reach and engagement in a competitive digital landscape.

In November 2017, Google Play Services introduced significant changes to its APIs by deprecating the GoogleApiClient class and implementing a more streamlined, feature-specific API client model. This transition aims to reduce boilerplate code, enhance thread safety, and improve memory efficiency by utilizing the Task model, which was positively received in Firebase. The new API approach shifts authentication to the Google Sign-In client, providing clearer control over the authentication process and distinguishing between Google Sign-In and Play Games Services identities. Developers are encouraged to adapt their projects by converting Games category static methods to corresponding API client methods and using the Task class instead of PendingResult, facilitating simpler code with less multi-threaded complexity. The changes also include new handling methods for multi-player invitations and the introduction of the GamesClient class for additional support functions. Resources such as the Android Basic Samples project and Client Server Skeleton project are available to assist developers in transitioning smoothly to the new API structure.

Voice interaction is gaining traction as a way to simplify technology use, and tools like Dialogflow, Actions on Google, and the Speech Synthesis API are making it more accessible for developers to create voice-based experiences. Voice Experiments, introduced by Amit Pitaru from Creative Lab, highlights innovative uses of voice interaction in areas such as music, gaming, and storytelling. Examples include MixLab for music creation using voice commands, Mystery Animal for a new twist on classic games, and Story Speaker for interactive storytelling via Google Docs without needing to code. These experiments can be accessed through Google Assistant on various devices including phones and Google Home, or through web browsers like Chrome. As voice interaction is still in its early stages, developers are encouraged to explore these experiments or contribute their own at g.co/VoiceExperiments.

TensorFlow release 1.4 introduces significant updates, including the integration of Keras into the core TensorFlow package as tf.keras, enhancing its utility in production workflows with high-level APIs. This release also sees the Dataset API moving to the core package as tf.data, which now supports Python generators and is recommended for creating input pipelines due to its superior functionality and performance. Additionally, the new utility function tf.estimator.train_and_evaluate simplifies the training, evaluation, and exporting of Estimator models, supporting both distributed and local execution. Installation of TensorFlow 1.4 can be done using pip, and the release notes provide further enhancements and updates. The TensorFlow community encourages contributions on GitHub and participation through platforms like Stack Overflow.

Resonance Audio is a spatial audio software development kit (SDK) released by Google, designed to enhance immersive experiences in augmented reality (AR), virtual reality (VR), gaming, and 360-degree videos by providing rich, dynamic sound environments without compromising performance, even on mobile devices. Leveraging technology from Google's VR Audio SDK, it uses advanced digital signal processing algorithms and higher order Ambisonics to manage hundreds of simultaneous 3D sound sources efficiently. The SDK supports major platforms like Android, iOS, Windows, MacOS, and Linux, and integrates with popular game engines and digital audio workstations (DAWs), offering seamless workflows for developers and sound designers. It introduces features like precomputed realistic reverb effects, control over the directionality of acoustic waves, near-field effects for accurate distance perception, and sound source spread. Resonance Audio also includes a DAW plugin for monitoring spatial audio and an Ambisonic recording tool for capturing and deploying sound designs, making it a versatile tool for creating complex sound environments across various media.

Google's Launchpad Studio is a six-month mentorship program aimed at assisting machine learning (ML) startups in creating impactful healthcare solutions by leveraging Google's advanced tools and platforms. The program has selected four startups—Augmedix, BrainQ, Byteflies, and CytoVale—that use ML to address healthcare challenges such as doctor burnout, movement restoration in paralyzed limbs, acceleration of clinical trials, and early detection of sepsis. These startups will receive tailored, equity-free support and resources, including Google Cloud credits and training, as they work closely with Google engineers and industry experts to complete ML-focused projects. Launchpad Studio, based in San Francisco, seeks to foster the ML ecosystem and develop product methodologies by partnering with innovative startups globally, with a current focus on healthcare and biotech, and plans to expand into other industry verticals.