How to use Google to help solve the world’s food crisis
- by admin
Posted July 03, 2018 12:42:33Google and other search engines have long been a boon to food research, as they have been able to search the web for a variety of foods, from berries to mushrooms.
But, until recently, the search engines had been unable to answer the questions about what makes a good fruit or veggie or what makes it good for human consumption.
Now, researchers have used machine learning to answer some of these questions.
In a paper published today in the Proceedings of the National Academy of Sciences, researchers from Princeton University and Google demonstrate that using a Google machine learning system, they can correctly identify which types of fruits and vegetables are most nutritious, and thus how much energy they need to digest to produce those nutrients.
The researchers used Google’s machine learning software called Bigtable to analyze more than 1,000 different food images.
They then combined the data with a database of more than 2.5 million nutritional information to create an analysis that showed fruit and vegetables with higher nutrient content were more likely to be eaten.
“This is really a groundbreaking achievement,” says researcher James Risley, a professor of computer science and engineering at Princeton and a co-author of the paper.
“It’s very exciting that we can actually do this on a large scale.”
The researchers also found that using the machine learning approach, fruit and vegetable producers could identify specific types of fruit that would be more nutritious and thus more economically profitable.
They also discovered that it would take less time to prepare a specific type of fruit than to prepare the same fruit with different ingredients, like a different type of sugar.
In other words, using a machine learning technique to understand how different types of food are prepared and eaten could improve food safety, nutrition, and production.
“It’s not just about how to make the best fruit or vegetables.
It’s about what’s in the right place in the supermarket and what’s the right time to buy it,” Risler says.
“We can look at food, and then the system can look into food.
That’s really the magic.”
For example, fruit may look tasty, but if you put it in the refrigerator overnight, it will spoil.
Fruit can also become moldy, so you can’t put it on your grocery list, even if you know the name of the fruit.
And when it’s eaten, it may taste sour, so it won’t attract the taste of your favorite fruits.
Risley and his colleagues were able to develop a system to automatically identify what types of foods were most nutritious.
They took this information, which was then fed to Google’s own automated system, and fed it to the data set, allowing it to determine which fruits and veggies had the most nutrients and which ones had the least.
The results showed that certain types of berries and vegetables had higher nutrient values than others.
For example: red berries, such as strawberries, were more nutritious than white, blueberry, and greenberry berries, and red and yellow varieties of potatoes were more nutrient dense than other varieties.
“We’re finding that if you have good data, you can actually get good answers from this system,” Rhasley says.
For the past several years, researchers at Princeton have been working on an alternative to traditional food labelling, called food label inference.
This method relies on machines to identify which foods are safe to eat, based on data from thousands of foods.
The researchers had previously used this method to determine if apples are safe for humans to eat and had previously developed a system that could identify the type of apples, how they were produced, and how much they cost.
The new work is the first to use machine learning and Google’s food labeling system to identify the types of nutrients a food is most likely to have.
“A lot of food labels today are based on the type and the nutritional content of foods,” Risalley says, “but this is really about figuring out how to build this information into a way that it can be used in a machine-readable way.”
The research was supported by the National Institutes of Health and Princeton’s Center for Health Analytics, the National Science Foundation, the Google AI Lab, and the University of Pennsylvania.
Posted July 03, 2018 12:42:33Google and other search engines have long been a boon to food research, as they have…
- How Fresno Community College’s staff has been targeted by cyberbullying
- How ‘Nashville’ will make you hate the ‘Community’ again
- Warhammer: Warhammer 2039 – A Tale of Two Cities?
- ‘We’ve never had this kind of fear before’: Bayside College students set to attend university
- The Dakota Community Bank is a good deal for the Dakota family