Machine Learning and Google APAC - Genrehow - A Technology Blog

Monday, July 2, 2018

Machine Learning and Google APAC

With the goal of spreading the good news of Google to the Asia-Pacific region, the tech giant’s annual press event highlighted the most vital products and services that will help both consumers and professionals i.e., Machine Learning.
What’s a complex set of software and algorithms without real-life usage? Google is pushing machine learning like never before, and the search engine specialist is placing significant investments in the project. Google’s push for machine learning actually began roughly three years ago with their speech recognition and ad targeting, and now they want to expand all the information they’ve gathered to more of the company’s vast array of consumer products.  One product at the forefront of Google’s machine learning push is inbox, which is Gmail’s smarter emailing sibling. Smart reply-reads your emails and suggests responses based on all your other replies. the more you use it, the better the software understands your style through machine learning, and the options become more precise. Gmail has been implementing machine learning a little more subtly, choosing which emails belong in
your inbox and which go to the spam folder. Google claims that Gmail can intercept 99.9% of all spam now.

Google Photos has a similar application, wherein you can search for pictures without the need to add your own labels and tags. As for Google translate, the conversations with the app are becoming more and more fluid, thanks to deeper algorithms and constant machine learning.

What is Machine Learning?

For the sci-fi nuts: No, this isn’t about creating hyper-intelligent cyborgs that’ll take over the world. As Google
themselves put it, humans won’t build bad robots; at the same time, the company also predicts that computers will rival human intelligence in 15 years time. Still, a scary thought, don’t you think? in a nutshell, machine learning is about making machines understand things using patterns and computational theories. the old way
involved using explicit rules to create smarter machines; the new way is about learning from examples, then adjusting to reduce errors. With that,  the learning process is gradual and would take billions of cycles to master
a function. Deep learning, which is a powerful class of machine learning, aims to advance the method using
simple, trainable math functions, but more research is needed before we hit human-level intelligence.
Deep learning is not new to Google, as they’ve been developing their own internal deep learning infrastructure (DistBelief) since before 2011. DistBelief has powered Google’s experiments in automated image captioning and their research into neural networks like Google’s own words, DistBelief had some limitations. it was narrowly targeted to neural networks, it was difficult to configure, and it was tightly coupled to Google’s internal infrastructure– making it nearly impossible to share research code externally. that’s where tensorFlow comes in.
Google’s second-generation machine learning system is specifically designed to be general, flexible, portable, easy-to-use, and completely open-source. that’s a key point because it means anyone from academic researchers to hobbyists to software engineers will be able to contribute to the development of tensorFlow and machine learning, accelerating the development process. Because tensorFlow is also built to be very general, any computation that can be expressed as a computational flow graph can be computed with tensorFlow, and any gradient-based machine learning algorithm will be able to tap into tensorFlow’s auto-differentiation and first-rate optimizers. Google says tensorFlow is available as a standalone library with the associated tools, tutorials, and examples with an Apache 2.0 license so you can jump right into development.

No comments: