Technology that can mimic and improve on the cognitive abilities of human brain has been the stuff of dystopian movie storylines for decades. But for large companies and research labs, such artificial intelligence has been a longstanding pursuit. Now, a specific breakthrough in AI — deep learning — is allowing business to use data to teach computers how to learn.
Deep learning uses layers of algorithms known as neural networks, which are designed to loosely represent the layers of the human brain. By feeding a computer tons of sample data and identifying what that data represents, the computer can learn patterns and begin to make inferences.
Deep learning requires vast amounts of data and a huge amount of processing power. But, since industries across the spectrum are generating tons of “Big Data,” i.e. digital data now being produced at an unprecedented rate and massive volumes, there is a clear opportunity for deep learning-powered applications.
Here are 13 industries that are drawing on innovations pioneered by deep learning to make major advances.
Diagnosis and treatment of diseases have become much more precise as doctors have begun to look at an individual’s gene sequence and molecular makeup. They use this information to gain a better understanding of what the markers on an individual’s DNA mean and tailor treatments accordingly. In other words, genetics allows doctors to treat the individual rather than the generalized disease. But there is far more molecular information than doctors can analyze, and a lack of clarity in understanding how an individual may react at a molecular level to external factors, from the environment to drug interactions. This is where deep learning comes in.
Startup Deep Genomics, which is backed by Bloomberg Beta and True Ventures among others, has fed deep learning machines tons of existing cellular information in order to teach machines to predict outcomes from alterations to the genome, whether naturally occurring or through medical treatment. The technology could provide the most precise understanding of an individual’s specific disease or abnormality and how that person’s wellbeing can best be advanced.
The pharmaceutical industry spends billions on R&D aimed at discovering the right set of compounds to treat specific diseases. Only a fraction of that research ultimately translates into commercial drugs. If deep learning techniques could look at the wide variety of molecular compounds that have already proven themselves effective and use this information to develop drugs to attack new or established diseases — or identify alternative uses for drugs that are already FDA-approved, it could help fast-track treatments.
A number of different startups are in the business of “virtual drug discovery,” including deep learning company Atomwise, which is backed by investors including Draper Fisher Jurvetson and Khosla Ventures. The company has worked with companies like Merck and been involved with Ebola treatment research.
A more devices become internet-enabled, hackers have an increasing number of entry points to infiltrate systems and cloud infrastructure. The best cybersecurity practices not only create more secure systems but can predict where the next attack will come from. This is critical since hackers are always on the hunt for the next vulnerable endpoint, so protecting against cyber attack requires “thinking” like a hacker.
Companies like Israel-based and Blumberg Capital-backed Deep Instinct aim to use deep learning in order to recognize new threats that have never been detected before and thus keep organizations one step ahead of cyber criminals.
Among the oldest and most unpredictable industries, agriculture is becoming increasingly data- and tech-driven. More predictable crop outputs — based on weather forecasting and data-based estimates — could potentially even take some of the uncertainty out of commodities markets.
New Mexico-based Descartes Labs aims to make “a living map of all the world’s agriculture” using satellite imagery fed into deep learning machines and lots of computing power, rather than a more traditional focus on surveys and site visits. Then, based on patterns over time, the technology is designed to predict crop yields and production.
Pinterest — where people often share aspirational images of products — is arguably the leading site for social commerce. But often, these images show a collection of items and don’t necessarily identify where something similar might be purchased. Now, using deep learning, the company is turning that repository of images into potential sales by allowing users to zoom in on specific items within an image and be served visually similar pins, some of which could be purchased from retailers’ pins. Pinterest is using the same deep learning-based discovery technology to surface more relevant videos.
Other internet publishers might eventually be able to turn every image or video on their site into a kind of high-gloss ad referral. The French company Deepomatic has already used deep learning to develop this kind of tech aimed at web publishers, allowing them to earn revenue off of the visuals within their content.
6. Auto/self-driving cars
There are already plenty of cars on the road with driver-assistance capabilities, but these cars still rely on users to take over when an unforeseen event occurs that the car isn’t programmed to respond to. As Sameep Tandon of startup Drive.ai notes, the challenge with self-driving cars is handling the “edge cases,” such as weather. This is why, using deep learning, Drive.ai plans to help the car build up experience through simulations of many kinds of driving conditions.
Nvidia is also working on self-driving car technology. Nvidia says it has used deep learning to train a car to drive on marked and unmarked roads and along the highway in various weather conditions, without the need to program every possible “if, then, else” statement.
7. Climate study
Predicting weather patterns has become increasingly important as governments and globe-spanning corporations prepare for broad-scale climate change and extreme weather events.
The National Energy Research Scientific Computing Center, for example, used NEON, an open-source library from deep learning company Nervana (acquired for up to $408M in August 2016 by Intel), to train a system to recognize extreme weather events based on visual pattern recognition. So far the results have been promising, with the company claiming near-perfect accuracy in identifying tropical cyclone patterns.
Similarly IBM is using deep learning to help solar and wind companies better predict weather and improve alternative energy production. So far, IBM has reported a 30% improvement in accuracy using its self-learning models, compared to conventional solar and wind prediction models.
The insurance industry is beginning to see deep learning as especially useful in managing claims. A machine that can recognize patterns in fraud could help these companies ferret out false claims and determine payouts for legitimate claims. Accenture, which provides consulting services to insurers, notes that it has already incorporated machine learning into claims dashboards to mimic the expertise and understanding of human agents. The company now sees deep learning as potentially playing a role in claims processing and customer service.
PwC similarly sees deep learning playing a role in standardized underwriting of common policies, such as auto, home, and commercial insurance, and in using deep learning image analysis to estimate repair costs after a claim has been filed.
The startup Tractable is specifically focused on building deep learning systems for auto insurance. The company aims to use images of car damages to teach machines how to estimate future repair costs, potentially bypassing the need to visit the auto body shop and rely on the garage’s assessment.
While the US government continues to heavily regulate and limit drone flights, the commercialization of drones is well underway, with UAVs used for delivery, search-and-rescue, photo journalism, and agriculture, among many other use cases. But much of this drone activity still takes place in areas where there are limited obstacles. Densely populated locations are more challenging, as the drone must react to circumstances that it has not necessarily been programmed to navigate, not unlike the challenges for self-driving cars. In addition, the skies lack the markers of the road that can help point a drone forward. Drones must figure out how to move to their destination without any obvious route to follow. As a result, companies investing in drones are looking to deep learning to help drones navigate around unmapped obstacles and without a preprogrammed flight path.
Swiss researchers at the Dalle Molle Institute for Artificial Intelligence, the University of Zurich, and NCCR Robotics used deep learning to teach a drone how to navigate a new hiking trail using lots of imagery of other hiking trails to understand the way forward. According to the researchers, the drone was able to determine the correct way forward with 85% accuracy, slightly better than the 82% accuracy scored by humans tasked with the same problem.
Similarly the drone startup Teal just incorporated deep learning company Neurala’s software into its “Follow Me” feature, allowing the drone to make “in-flight decisions” and follow a person or object through image recognition. Neurala, which is backed by NASA among other investors, claims it is the first truly autonomous drone.
10. Healthcare/medical diagnostics
To well-trained eyes, ultrasounds, MRIs, and other image-based health test results may be relatively simple to interpret. But AI experts still see room for taking human error out of medical diagnostics through deep learning. Last year, Australian radiology company Capitol Health invested in deep learning-based medical diagnostic startup Enlitic, which aims to make medical diagnosis more efficient and accurate. The radiology company hopes to use Enlitic to improve its radiologists’ speed, by using the technology to highlight key areas of study on a given test result, make measurements, and suggest similar patient cases for guidance.
Another startup, Bay Labs, is using deep learning to bring precise medical diagnostics to the developing word where radiologists and other trained professionals might not be on hand to analyze test results. The company is specifically focusing on the analysis of ultrasounds to treat heart disease.
Since deep learning has already seen widespread experimentation and refinement for textual analysis, it’s no surprise that Google, the leader in search, has made widespread deep learning-based updates to its search technology. Google’s deep learning-based RankBrain technology was added to how Google manages and fills search queries back in 2015. The technology helps handle queries that have not been seen before. As a reflection of how bullish Google is on AI for search, the new head of Google’s search team as of February 2016 is John Giannandrea, who formerly headed up AI at the company.
12. Marketing automation
Targeting the right clients is a key component of digital marketing. Now, a few different companies are using deep learning to help clients — especially enterprise-focused companies — reach out to the right prospects.
Deep learning company Metamind was recently acquired by Salesforce in April 2016. Salesforce said in a blog post that it would use Metamind to help “further automate and personalize customer support, marketing automation, and many other business processes.” Prior to the acquisition, Metamind’s specialty was parsing text for meaning and sentiment, which could be used for a broad array of marketing purposes, including analyzing signals from sources such as social profiles to determine the quality of a lead.
In a similar vein, the deep learning startup MarianaIQ, which partners with clients like Marketo, analyzes clients’ enormous database of potential customers and helps identify and reach out to the best prospects.
Robots are already playing a role in picking and packing at some major warehouses, like those operated by Amazon. Now it looks like those robots could get even better at their job with the addition of deep learning tech. A team from TU Delft Robotics Institute of the Netherlands and the company Delft Robotics won the annual Amazon’s Picking Challenge with a deep learning-based robot. Ahead of the competition, the robot studied 3D images of the stockroom to learn how to move items using the robot’s physical attributes. The strategy of creating a robot that can learn was extremely effective, with the deep learning robot picking items three times faster than the next-fastest robot competitor. The team’s robot also won the packing competition, moving items from a container into bins on a shelf.
That said, deep learning robots still haven’t bested humans at picking and packing, suggesting that it may be harder to emulate our nimble bodies than our overdeveloped brains. The robot picked 100 items an hour, compared to the 400 an hour typically picked by the average human.