Issue #10 of the
Infinite Waves
Newsletter
If you’re interested in Game AI and not currently subscribed, do so below. I’ve just conducted an interview with a Game AI developer (for a very interesting and currently unreleased game) and the insights will be published in the next newsletter issue. Subscribe if you want to be emailed with the issue as soon as it’s released:
Picture the scene. It’s a Friday night. I’m having a wild time by ordering a pizza and sitting on a beanbag at my coffee table doing research. It sounds pretty extreme so I’ll end the story there before we get heart palpitations.
Last Friday I decided to look into Tensorflow. Something I felt would be an issue of the newsletter sooner or later.
I had a look and within a few minutes I was pretty impressed at what the disembodied voice in the introduction video was telling me.
It also has a pretty cool name, just say out loud, see how it feels ‘Tensorflow’. Yesh, it’s sexy.
What is Tensorflow?
According to the friendly disembodied voice in the video:
“Tensorflow is an open-source, end-to-end platform for machine learning. It provides a comprehensive ecosystem of tools for developers, enterprises, and researchers who want to push the state-of-the-art of Machine Learning (ML) and build scalable Machine Learning powered applications. Tensorflow was designed to help you learn to build models easily. It simple for you to learn and implement Machine Learning, Deep Learning, and Scientific computing. It provides you with a rich collection of tools for building models.”
So then, it seems to have a lot of bells and a lot whistles. But that’s a lot to take in and unpack. So I’m going to spend the rest of this issue trying to contextualise that description with some Tensorflow examples in the real world.
airbnb 🏩
When picking a place to stay photos of the accommodation can sometimes make or break a decision. Images are crucial to Airbnb’s marketplace platform. They host millions of properties, each associated with multiple images. Until recently they were unable able to do anything meaningful with all of these photos.
“When a guest interacted with listing photos of a home, we had no way to help guests find the most informative images, ensure the information conveyed in the photos was accurate or advise hosts about how to improve the appeal of their images in a scalable way.”
Shijing Yao - Staff Machine Learning Scientist at Airbnb
The team set out to classify Bedrooms, Bathrooms, Living Rooms, Kitchens, Swimming Pools and Views.
They used Deep Learning with Tensorflow as their method. It enabled the company to optimise the user experience. For potential guests the system allowed for the re-ranking and re-layout of the photos, surfacing the ones that people were most interested in first.
On the host side, the classification solution helped to automatically review listings and therefore increase accuracy and maintain adherence to quality standards.
Twitter used Tensorflow to build their ‘Ranked Timeline’ feature. It helps users by making sure they don’t miss import Tweets, even if they’re not following a large number of accounts.
So when you open Twitter and you see those (sometimes) juicy, relevant tweets, yeah, you have Tensorflow to thank for that.
When ranking, each candidate Tweet is scored by a relevance model in order to predict how relevant it is to each user. “Relevance” is defined by multiple factors including how likely a user is to engage with the Tweet, and how likely it is to encourage healthy public conversation. This model uses thousands of features from three entities: the Tweet, Author, and viewing User. Some example features include:
Tweet: its recency, presence of media (image or video), total interactions (e.g. number of Retweets or likes), real-time interactions on the Tweet.
Author: viewing user’s past interactions with this author, the strength of user’s connection to them, the origin of the relationship.
User: Tweets the user found relevant in the past, how often and how heavily they use Twitter.
Yi Zhuang, Arvind Thiagarajan, and Tim Sweeney from Twitter Cortex and Home Timeline Teams
Tensorflow is helping doctors detect respiratory diseases 🩺
I thought this was a particularly cool application.
Using a combination of Tensorflow and innovative hand-held medical devices doctors are able to better diagnose respiratory diseases.
“The traditional stethoscope hasn’t changed for close to 200 years. The doctor is limited by the human ear, and they can not hear specific frequencies, this method is very inaccurate, and causes a lot of misdiagnoses. Our mission is to use Machine Learning and Tensorflow to revolutionise the diagnoses and treatment of respiratory diseases in low-resources areas such as Sub-Saharan Africa.”
Lewis Wanjohi - Co-Founder and CEO of Tambua Health
Online demos using Tensorflow.js 💻
https://www.tensorflow.org/js/demo
My flow with Tensor
Tensorflow state that they offer a scalable and highly portable solution that can run on a range of devices and platforms. They also have a ton of great resources to learn from, and you can start from the super basics.
Currently I’m following an AI learning path with Kaggle, but along side this Tensorflow looks like a pretty good way to get up and running with some projects.
So over the coming weeks I think I’ll be attempting to put something together as a maiden AI project 🤞.
Keep up-dated with the progress:
Tensorflow Resources
The Tensorflow website is full of educational resources and guides, here are a few to get started with:
If you liked this content please share it out and subscribe:
Sources:
Application Programming Interface - Wikipedia
Central processing unit - Wikipedia