A lot has happened recently in the AI and machine learning realm, most notably a new academic boycott: the highly acclaimed journal Nature, which has one of the highest impact factors for academic journals, is aiming to start a closed-access machine learning journal, and some big names in the space, like Ian Goodfellow, Jeff Dean, and Yoshua Bengio, don’t like it. By signing this petition, they voiced their opinion that academic research in machine learning and artificial intelligence should be open to everyone and not cost prohibitive.

Now on to the fun stuff.

Can you perturb a Hydra?

There’s a simple fresh-water dwelling organism called a Hydra that’s easy for scientists to observe because its transparent body clearly displays its entire nervous system. This allows scientists to easily study how its movements are connected to its nervous system. Since this organism has a small repertoire of movements, scientists wanted to see if they had identified all the Hydra’s gyrations, and then once they had, they wanted to see if environmental factors would change the little organism’s behaviors.

Using machine learning, scientists were able to map the entirety of Hydra behavior. They had their tool review hours upon hours of its activities, identifying every action it could possibly make, and scientists confirmed that the organism has ten movements. Then they tested how it reacted to environmental changes, like dark versus light, and food versus no food. It turns out, environmental factors didn’t make one lick of difference for the Hydra. The organism just kept doing its same ol’ thing with very few, if any, changes.

The microscope that points out cancer

Google’s new microscope is receiving a lot of attention. It uses machine learning and augmented reality to point out breast cancer and prostate cancer cells to doctors. As a doctor puts biopsies under the microscope, the technology compares the image against its database of images. It then uses augmented reality to show where it has identified cancer by virtually overlaying a shape that delineates the cancer cell’s location. Get a much more fascinating view into how it works with this video:

The Vatican Secret Archives

The Pope has archives that date back 12 centuries, but those archives haven’t been easy to publicly access. Very little of what the archives contain is online, and even less of what is online is searchable. Part of the problem is that most of these archives were handwritten, and handwriting is really, really hard for a computer to parse and translate into typeface, especially the medieval, calligraphy-like cursive contained in this particular archive.

However, four researchers working to move the archive online may have come up with a solution. Using convolutional neural networks, a type of machine learning, along with other tech, they’ve trained their algorithms to identify letters from ink strokes and then predict which letters the ink strokes form. It’s 96 percent accurate in its translation, but that 4 percent can still account for a lot of mistranslated letters. With this good start, though, the idea is the tech will get better as it’s continually fed new material and properly trained and will eventually allow unprecedented access to a treasure trove of information.

More Stories

Pinterest uses machine learning to make their pictures searchable by skin tone
Physicists put tool to work identifying stages of star development
Scientists develop tech to better identify which bacteria can produce proteins which act as gates to our cells
Machine learning tool show promise in categorizing breast cancer risk
Student team develops machine learning refugee healthcare app that takes third place in competition

Extras

For any podcast fans, check out this podcast episode that talks about how one company decides when to use the latest AI tech—and when not to.

For anyone doing a lot of data preparation work in Excel, you should try out a great new tool that is easy to use and makes life so much easier. Trifacta’s Wrangler tool is free, and here’s a Udemy tutorial to learn how to use it, if you need one.