Machine Learning Blueprint Newsletter, Edition 5, 10/15/17

 View Only

Machine Learning Blueprint Newsletter, Edition 5, 10/15/17 

Wed June 19, 2019 02:46 PM

B
This newsletter is written and curated by Mike Tamir and Mike Mansour. 

October 15, 2017

Hi all,
It’s Mike Tamir - I’ve teamed up with Blueprint News in order to launch the Machine Learning Blueprint, an ML Newsletter. We’ll be publishing weekly (fifth beta edition below), covering general news, tutorials, research, etc. and how it applies to the realm of Machine Learning and Data Science.
The stories are meant to be quick, with commentary when necessary. We're sending this week’s edition to you because we’re looking for feedback on this issue. This can mean the story is too short, not relevant, you already saw it, etc.
Please reply to the email address sending the newsletter (ml@theblprint.com) with any thoughts you have. If you’d like to unsubscribe from further emails, please reply “unsubscribe” to this email. Thanks to everyone in advance!
Links to stories are in the titles.
Best,
Mike

Spotlight Machine Learning Articles
Facebook, Google Spread Misinformation About Las Vegas Shooting. What Went Wrong?
A critical review of how crowd-sourced information wrongly implicated an innocent man, with an exploration of how algorithms on Google, Facebook, and Twitter should manage sources of false signals like this though automated methods.
Machine Learning Blueprint's Take
This is a serious issue, especially when the topic has as much gravity as what happened in Las Vegas. Because Google, Facebook, and Twitter function as mainlines for how many get their information, and because the information cannot possibly be vetted by humans in real time, they must grapple with treading the fine line between censorship vs not enabling the irresponsible propagation of information. This topic promises to become more punctuated over the next twelve months as US midterm elections gear up and no suitable AI Machine Learning solutions are available yet.
Can AI Be Used to Eliminate Bias And Improve HR Operations?
A review conducted by the Human Resources Council (HRC) on eight areas where Machine Learning can be applied to help bring objectivity to HR activities.
Machine Learning Blueprint's Take
Last week we featured an article on comments recently made by Google’s AI chief, John Giannandrea, concerning the risk of increased bias that automating processes with Machine Learning might bring. Giannandrea’s warning might seem at odds with the recommendations of the HRC. However, this tension brings to light both what we should take away from Giannandrea’s comments, and how the HRC’s recommendations should be implemented responsibly. Many areas can benefit from Machine Learning, including HR. But we also must be circumspect, looking for hidden bias in how our algorithms are trained.
Google's Deep Learning AI is No Smarter Than a First Grader, Study Says
A study published last week evaluated the IQ of various AI systems. Google's AI, while the best, scored low at 47.28. Almost twice as well as Apple’s Siri, which scored 23.94.
Machine Learning Blueprint's Take
These results are not surprising. Machine Learning driven assistants are still designed to execute specific suites of tasks. They are not fully integrated intelligent agents. This may be more of a reflection of the abuse of the term “AI” in recent years. Products that simply employ some sort of applied Machine Learning are not intelligent agents ready to take an IQ test, Ok Google, Alexa and Siri included.
Machine Learning in Healthcare
This week we saw a number of articles focusing on the intersection of Machine Learning and healthcare:
An explanation of why healthcare data makes a good target for hackers because personally identifying data in these systems tends not to change.
Machine Learning technologies like Watson are being used to better inform doctors, providing relevant information for them to review and synthesize on the job.
A review of Machine Learning efforts on the diagnostic side of the equation, this article supports providers like radiologists in better evaluation and risk identification.
Learning Machine Learning
Train a Deep Learning Model Without Code
Google released an interactive Machine Learning demo that allows users to train and see the architecture of Deep Learning models on images from their webcam without any code.
Set Up Your Mac With TensorFlow
A how-to guide for setting up your Mac for Deep Learning image detection, using python and Keras (a “front end” deep learning wrapper with a TensorFlow backend) and GPU acceleration.
Interesting Research
Generating Sentences by Editing Prototypes
In the past year we have seen a lot of advances in text summarization using RNNs and attention mechanisms. In a way, results in this article solve the opposite problem: instead of summarizing text abstracting out only the essentials, their algorithm is designed to enrich text with semantically equivalent reformulations of higher perplexity.
Machine Learning News Links
Evidence online dating is changing society Online dating results in one third of new marriages. Researchers analyzed how this leads to increased inter-cultural marriage rates. Further, they report that their “model also predicts that marriages created in a society with online dating tend to be stronger”.
Google releases TensorFlow Lattice: flexibility empowered by prior knowledge Google’s new TensorFlow tool provides fresh functionality, enabling models to adapt to context changes in the data. This is achieved by leveraging lattice based interpolators capable of adapting to local monotonic behavior, like the impact of drive time on coffee shop recommenders in rural vs urban contexts.
Falling Q3 seed funding could stunt AI deep learning hyped innovation cycle
Machine Learning Is easy to fool — Why that needs to change
Learning diverse skills via maximum entropy deep reinforcement learning
Should the west stop worrying about China’s AI revolution?
An AI deep learning algorithm for the blind spot in your car
Phone-powered AI spots sick plants with remarkable accuracy
Summaries and reviews from RecSys 2017
Machine Learning Books

#GlobalAIandDataScience
#GlobalDataScience

Statistics

0 Favorited
5 Views
0 Files
0 Shares
0 Downloads