Synapses, 2

Source: Deep Learning on Medium


How Deep Learning is used to reduce wildlife poaching, help poor farmers improve yields, detect illegal deforestation and save endangered languages.

Chart of the Week!

Dragon-Leap! China is surging in AI patents, will it change her IP compliance history? On the other hand, the AI landscape in US is made up by just ~5 tech companies, with IBM being wolf-pack leader.

Wild Neurals

OpenCV to classify elephants with 98% accuracy

Today almost all reserved forests has camera traps which generates hundreds of hours footage and thousands of images each day. What happens to all that corpus? Some human (mostly unpaid or underpaid) manually analyses it for some basic pattern , which is always if the endangered animal present or not and if a unexpected humans are seen. The sheer volume creates a lag in the system and inefficiencies. Most wildlife reserves still uses manual estimation to do animal census; which generates bad data and then that results in poor decision making. When I saw how Anthony’s naive CNN code can detect elephants (killed heavily for ivory) & can help prevent poaching; it struck me that AI is more about imagination than algorithm. It can also potentially help track movements & habitat changes and avoid human conflict.I hope to expand it to Tigers and deploy as one of Wild Neurals project in tiger reserves soon. Still, lots of wildlife species out there which needs little attention ! Git

Lost in Translation

There are ~7000 languages in the world, but most are dying. And with it dies the culture, the stories and a link to human evolution. For example; Madia, the language of the land where famous Jungle Book is based out of, is critically endangered.The major cause of extinction is lack of scripts as most native languages are only verbal.Technical revolution makes it even more marginalized . There is a serious dilemma as to how to educate natives in their own language when all education is only available in foreign language & no one from outside can easily learn their language? Forcing natives/tribal to learn modern languages may seem noble but is also seen as colonialism by them. How can we make a language accessible? @TeHiku used Deep Learning to build the first speech recognition system for the Māori language of the indigenous NZ people. The goal is to train machines to transcribe thousands of hours of native language speaker audio recordings to make native te reo Māori more accessible to language learners as native speaker population is in decline. The project for Luxembergish is also underway.Github

Green Revolution 2.0

India in 60’s was a chronically starving nation, always begging US to feed its hungry masses. This is despite agriculture being largest industry then. The problem? Poor yields due to age old methods! Late 60’s saw India adopt modern technology with the help from US & Mexico , that increased its food production so much that it could export even after feeding a billion. This agri-tech transformation is known as “Green Revolution” in India. But unfortunately nothing much has changed since then. Now its facing a new crisis led by crop failures that results in ~50,000 farmer suicides each year and still worsening. A German company uses crop pictures, later processed by Deep Learning model to predict plant diseases, which coupled with other features can provide further recommendations on epidemics of crop worms & yields. Model is based on vanilla CNN algorithm. While world is focused on high risk human diagnosis, this low risk plant diagnosis is having a larger human impact. Full Story (5 mins )

When AI is not enough ….

Keras based DL used to detect illegal mining in over 70,000 acres

Illegal Mining in Africa and Latin America , is a huge problem . It not just causes deadly pollution but loss of habitat for natives, revenue plunder, child labor and human trafficking. One company is helping detect this Leprosy of the Land by doing image segmentation with Keras in Ukraine. Rainforest Foundation is providing drones in Amazon to arm native Peruvians protect their habitat. But does detection really lead to a cure? As below video story unfolds we see that, while AI & drones can detect daylight robbery; it is still helpless in a society where justice is a luxury. The deforestation continues even after police sees it nakedly. Does AI work when their is no rule of law or human rights? Unfortunately, the hype of AI seems to suggest it as a panacea but forgets to add that star — not applicable everywhere! It can work effectively if society has foundations built on equality, liberty & justice. This is not being pessimistic but reality check that AI may have all the solutions but its not an replacement for human struggle! Rainforest VICE STORY (7 mins video) Keras Project