In our latest monthly round-up, we include a look at analysis revealing medieval recipes with surprising antibacterial properties, what you can get for $Trillions, and how to make a good impression using data science.
Show Me The Money – $Trillions of it
Using data from the World Bank, New York Times, Bloomberg, The Guardian, Washington Post and CNN, (all detailed on this spreadsheet) this fabulous data visualization was created by Information is Beautiful to show where $Trillians are earned, lost, owed, hoarded or spent, as well as where they need to be funnelled in order to hit certain development targets.
See the full, big graphic here.
Machine Learning Made Easy – Salesforce Open Sources TransmogrifAI
In mid August 2018, San Francisco cloud computing company Salesforce published TransmogrifAI, an automated machine learning library for structured data, on GitHub.
Machine learning models are based on artificial intelligence that identifies relationships among many – sometimes millions – of data points. These are usually very difficult to architect and data scientists might spend months preprocessing and extracting useful features from the data, in order to narrow down algorithms and build a system that performs well in the real world.
Some of the code released to the open source community is used to power Salesforce’s AI platform, Einstein. In explaining the move, Director of Product Management for Salesforce Einstein, Mayukh Bhaowal, commented:
“The goal of democratizing machine learning can only be achieved through an open exchange of ideas and code, and diverse perspectives from the community will make the technology better for everyone.”
The full VB article on the open sourcing of TransmogrifAI can be read here.
Antibacterial properties found in medieval recipes, using data mining
Researchers used data mining techniques to analyse medieval texts, and found that apothecaries of the time used receipes with significant antibacterial properties.
The norm is to look back at medical practice from that age and to chuckle at the hocus pocus approach. However, research by teams at the University of Pennsylvania in the US and Warwick University in the UK shows that a great deal of medicine was backed up by science and not just superstition.
Analysis of the legendary medical book, The Lylye of Medicynes, was tricky as it detailed over 3,000 ingredients and 360 recipes used to treat 113 conditions.
Some of the practices, such as the use of honey for treating infections, are still in use today and the researchers believe that perhaps history has since painted a somewhat unfairly comical picture of the era’s medicine.
What is most exciting is that, partly due to the sillier antics of medieval physicians, subsequent eras have lost some helpful recipes that remain unknown to medicine. Their may be some ‘breakthroughs’ to come through further data mining of medical history books.
Read the full MIT article here.
How Coca-Cola use TensorFlow for Digital Marketing
In this video, TensorFlow talk with Patrick Brandt of Coca-Cola about how they use the open source dataflow programming software as part of digital marketing campaigns:
Use Your Data Visualizations Wisely in eLearning
An interesting article on how data visualizations within elearning content can unintentionally (or sometimes intentionally…) confuse, overwhelm or deceive learners.
It cites the four main issues that mislead learners as:
- Hiding relevant data
- Presenting too much data
- Distorting the presentation of data
- Describing the data inaccurately in annotations, titles, or within the visualization itself
The sound advice given in Pam Hogle’s article is not to avoid using data visualizations in elearning if you want to enhance user engagement and support learning outcomes, but to think very carefully about the format used and work closely with any designers involved to ensure, data, design and context are all aligned.
Making a Good First Impression – Data Analysis of Video Clips
Are interested in how to make a good first impression? Mattie Terzolo, a data scientist at Metis, describes a model that predicts how good a job you are doing based on a video clip submission.
Various studies suggest that we form initial impressions about people within anywhere from 0.1 seconds to 1 minute. There are good reasons for doing so that are hard-wired into our DNA, and the principle of first impressions having long-lasting effects is a pretty well established concept.
To understand the components of this principle, the ChaLearn LAP team of researchers analysed over 10,000 video clips, each 15 seconds long, and broke down data points within the three key areas of:
- Images – physical appearance, posture, eye contact
- Audio – tone of voice, energy in speaking
- Text – words you chosen, topics spoken about
Scores are given in each area and the analysis gets very granular, as described in this article by Mattie Terzolo, a data scientist at Metis. What will be exciting for many is that if you want to improve how you make a first impression, there are some constructive action points to take away and work on.