Is there anyone working on either a decentralized deep learning algorithm, or a consumer facing app that uses AI to help people diagnose themselves?
My wife was just diagnosed with CVID a couple of weeks ago, it's like AIDS except it's not Aquired, it's part genetic and part environmental - but it's a rare primary immunodeficiency disease.
She's had this her entire life. She's 33 years old, a wonderful mother to our 5 year old daughter, and beautiful singer, actor and writer. She was misdiagnosed 3 or 4 times, most recently she was eating gluten free for the last 8 years because she was diagnosed as celiac disease.
She's lost most of her hair over the last 6 months and has been in the hospital 3-4 times this year. It turns out, she never had celiac, she has always had CVID. Where Deep Learning fits in.
My wife should have been diagnosed in her childhood years, in fact, all it would have taken was a simple blood test to measure her antibody levels, and an Immunologist appointment.
With all of her symptoms, her medications and blood test results - a deep learning algorithm would have been able to suggest a proper diagnosis in a few minutes instead of the 30 years that it took for her to get properly diagnosed by just letting doctors do their thing. The problem with diagnosing rare diseases:
CVID affects 1 in 25,000 - 50,000 people, it's a rare disease where patients present with a myriad of symptoms and autoimmune problems. It's hard to get a correct diagnosis because a patient typically sees many different doctors to treat the different types of symptoms, and they don’t typically share information efficiently - nor do they have an incentive to properly diagnose her.
There should be a visually appealing, easily marketable app that combines machine learning and crowdsourced input from app users to give the "hot/cold" direction that will greatly improve time to diagnose these "zebra" cases.
The average lag time for CVID diagnosis is 6-7 years. This is common with most rare diseases. If my wife was diagnosed even 2 years ago, she would not have lost all of her hair.
The treatment for my wife’s condition is IVIG every 2-4 weeks, and it greatly improves quality of life and life expectancy. The earlier a rare disease is diagnosed, the better the quality of life. The problem with doctor-facing AI solutions:
I see that there are some machine learning startups, but they are mostly targeted towards health professionals. There’s resistance from doctors to adopt AI.
The problem is that this technology needs to be available for the patient, not just doctors, and not just specialists at John Hopkins or the Mayo Clinic.
Nobody is going to be as motivated and investing in someone's health as the person and their loved ones. Quite often, patients with diseases become more knowledgeable than the specialists treating them for the disease.
There’s a lot of knowledge to be tapped into there from the 'zebras' themselves. Potential barriers to a centralized organization providing this solution:
The FDA and drug companies are resistant to technologies that allow users to diagnose themselves. 23andme ran into issues with this. They just finally got FDA approval in October to start helping people agian (http://www.popsci.com/23andme-gets-fda-approval-for-direct-to-consumer-genetic-tests
) Some existing projects:
A friend who sold his company to Salesforce for 70 million dollars introduced me to this TED talk shortly after my wife was diagnosed, where Jeremy Howard explains how deep learning works, and it’s potential applications: https://www.ted.com/talks/jeremy_howard_the_wonderful_and_terrifying_implications_of_computers_that_can_learn?language=en Jeremy's
company Enlitic is using deep learning to help doctors come to a proper diagnosis faster, currently only focusing on radiology. Again, it's just doctor-facing. www.findzebra.com
is a search engine that you can input your symptoms, it uses something similar to deep learning to suggest possible health issues, but it doesn’t use AI. It crawls and indexes only curated medical sources.
Jeremy Gardner & the Augur guys introduced me to www.crowdmed.com
which is a great solution for diagnosing rare diseases, but costly.
CrowdMed is a site where you tell your story, and then it uses crowdsourced knowledge to come to diagnosis suggestions. You pay a monthly fee of $299 - $749 a month, and then “medical detectives” investigate for you, and their predictive market algorithm ranks what the detectives submit.
Vitalik Buterin of Ethereum told me about www.numer.ai
which is a competition that uses homomorphically encrypted machine learning to let people try to predict the stock market. It’s a way to anonymize data, to alleviate some concerns about people’s medical data being publicly available.
Xprize even has a $5 million dollar prize up with IBM Watson for AI http://www.xprize.org/ai
There's also openai.com, with names like Elon Musk & Sam Altman attached, it's a non-for-profit with a billion dollars committed, but they haven't yet released what their focus is. Potential solution to being blocked by FDA etc:
Decentralized deep learning on a blockchain, where users are rewarded tokens for providing the hot/cold and running the network. Think ethereum, bitcoin, etc.
In my limited understanding of machine learning, it seems that for a deep learning algorithm to learn, humans need to give it hot/cold inputs to the correlations it comes up with as it compares datasets (Jeremy’s ted talk video explains that)
My theory is that a decentralized deep learning algorithm on a blockchain could be built where the people giving hot/cold inputs are awarded with a token for doing the mechanical turk style work. When consensus is achieved, the people who were correct get rewarded. Similar to how Augur’s reporting system works, or bitcoin’s proof of work.
If users are rewarded for giving correct hot/cold inputs to help the deep learning AI learn about subjects, there’s a financial incentive to keep the network running.
Companies, individuals, universities, etc could tap into the algorithm to use it for whatever purpose they want - and they would pay to use it.
IE I want to build an application that uses deep learning to help diagnose rare diseases so people like my wife don’t have to suffer going undiagnosed and untreated their entire lives. I would pay to have the algorithm learn about the human body, how it works, diseases, treatments, etc. The users of the network get paid to "train it” with hot/cold inputs.
Is there anyone working on anything like this, whether it’s centralized or decentralized?
This is very valuable - Homomorphic Encryption & Blockchain - We write about @48coins #blockchain #CryptoNewsnet #Encryption #Homomorphic You are here: Home / Crypto Blog / Bitcoin / Homomorphic Encryption & What it Means for Blockchain. Homomorphic Encryption & What it Means for Blockchain January 14, 2019 / in Bitcoin / by Crypto News. For those not familiar with cryptography, encryption, in its most basic sense, is a cryptographic equivalent of a lock and key. In a way similar to locking your valuables in a safe, encryption ... Even when the technology matures, homomorphic encryption is likely to find applications largely in niche fields, such as stock trading, where the need for privacy outweighs the tremendous computational costs. Nevertheless, computer scientists have shown time and again that the science fiction of today can very well be the reality of tomorrow. Homomorphic encryption has some immediate practical applications. Consider the Paillier scheme that’s used in several electronic voting protocols. Paillier is homomorphic with respect to addition. Now imagine: each voter encrypts their their ballot as a number (0 or 1) and publishes it to the world. So! Homomorphic encryption allows this to happen where the patient data and the algorithm are both protected. The 'fully' homomorphic encryption (i.e. inducing a ring homomorphism onto the encrypted data) allows for a much more efficient and robust set of computations to operate on the data.
This talk was given at a local TEDx event, produced independently of the TED Conferences. Tech entrepreneur and mathematician Charles Hoskinson says Bitcoin-related technology is about to ... Homomorphic Encryption Part 2 - Duration: 3 ... Amazing Stock Recommended for you. 15:14. What is Bitcoin? Bitcoin Explained Simply for Dummies - Duration: 12:49. 99Bitcoins Recommended for you ... Pitch of the Team "Homomorphic Blockchain: Quantum computing safe and GDPR conform encryption for DLT / Blockchain.” at Blockchained Mobility Hackathon 2018 From July 20th to 22nd 2018 the ... Fully Homomorphic Encryption from the Ground Up ... Asymmetric encryption - Simply explained - Duration: 4:40. Simply Explained - Savjee 453,096 views. 4:40. Math Behind Bitcoin and Elliptic Curve ... Homomorphic Encryption -- Nick Gonella - Duration: 34:48. White Hat Cal Poly 4,127 views. 34:48 . e-Security Week 1 (Introduction and Crypto Fundamentals) - Duration: 1:54:36. Bill Buchanan OBE ...