Healthcare is a very tough industry. I learned this by talking to several managers, executives and practitioners around the world in innovation challenges and hackathons. In Switzerland, privacy is a top priority before you can talk about anything else. Given my knowledge as a researcher in privacy and security, I personally had no idea how to solve that problem until hitting upon a concept called Federated Learning.
It was started by a random meetup in Bangkok. I was not very serious, just curious. I showed up at an event as a true backpacker with elephant pants and a pair of sandals bought in a wet market in Vietnam. Unexpectedly It was the first local meetup of a new community about AI and Privacy called OpenMined.
I went to Thailand for a fun trip. It was aimed at cultural exploration. I love spicy Thai Food. The best thing about Bangkok was street food everywhere. I could not kick my ass out of a city and stayed around in co-working space and coffee shops to … code and write. In 2015, it was a life changing moment when I was in a Thai Restaurant. Something popped up that changed the entire direction of my research and career into a completely different domain.
My first reaction was hold on, I was a researcher in this emergent field of privacy and information theory. If there was something like that I probably would have been able to detect it at a very early stage. That year, in addition to generative adversarial networks, I used to have a quick tour into quantum encryption, a mysterious field only people with a strong math background in number theory only able to dive deep into a literature. I had lots of doubts.
Federated learning is a concept combining ideas of authentication and data process together. It was started by a team from Google based on their own engineering framework. Then adopted by a research community in 2017, especially a group in Oxford University and DeepMind. On top of that is differential privacy - an idea that combines homogenous cryptography to encrypt data end-to-end seen as a holy grail in the privacy and security community.
When I asked about the reliability of this framework, most people said something very serious. “Yes, we have a math proof. This equation tells that information can not be revealed.” To be honest, this is not the first time somebody has told me something similar. Theoretically, deep learning “ancestors” were rejected by the research community also by a math proof.
I got an invitation to the Philippines. It was a one day event supposed to be quick and sweet except for one thing. Less than 24 hours before my flight, there was a super typhoon (or a hurricane) coming to the country. My second ticket was canceled along with more than a hundred flights.
After a few searches, I found out that there would be another flight but in another city, so I would take another train from the South to the North. That sounds like a feasible plan, so I bought one more ticket. Next day, I woke up super early in the morning to arrive at the airport and learnt that I needed a VISA to pass a border at the flight counter. As a Vietnamese citizen, I did not have to apply for a VISA application to Thailand and the Philippines, but still have for Taiwan. I was supposed to leave the airport and wrote an email that I could not have come, but then quickly I went to the internet and found out I could apply for an e-VISA by using my Schengen VISA. I did go through a similar process to obtain one to travel to Turkey but then had to cancel that trip to stay focused on BlueSkin. Less than an hour later, I came back with a solution and an amazing run to arrive on time after a few more similar incidents along the way.
This is my confession. I personally feel super awkward in all pitch competitions. When looking at all examples of all winners from every competition, I could not see myself in those people. Compared to all, my presentation was super simple with an unpolished script. However, I am pretty damn sure what I am talking about, especially about AI. However, the main reason why I came to that event was not to win but to see how much I can push myself out of my comfort zone to build a company. And it worked.
Two days later after a return trip, I made another flight to Canada for Neurips. It was not surprising that I could not buy a ticket. After ten minutes, everything was sold out, but there was an expo. I could not join all the sessions, but enough for a lot of interesting things to happen. On the first day, when in line for two hours in two buildings, two floors and one basement, I found out the guy behind me was from working for a Big Tech in Switzerland and a reviewer of Federated Learning, so it was not a hippy group in Bangkok. Plus, I got a mug with a new brand logo for the conference - the first and the last of its kind till now. Then, walking around I just discovered the whole expo was about applying finance and health. And the workshop of climate in the following year would be in EPFL where I completed and tested the scrappy version of BlueSkin. There were a bunch of autonomous driving car startups around. I searched for them for months in the US. Now they just showed up. Honestly, due to a constant traveling schedule, I did not know about an agenda until the arrival.
At the moment, I was close to the point of running out of budget. The Swiss Insurance would like to offer me a job, but I rejected it. However, this was neither the first nor the last time. This time it was a company. Different from several AI startups, I did not try to raise capital and have a fancy pitch with a big promise. Probably the only thing I tried most was using my own savings, credits and loans from the bank based on my previous income, a few hundred million Vietnamese Dong to travel to the West. It was supposed to be in only two weeks for one city at the end of everything which I could foresee. Then that trip was extended to 18 months in several countries to do research, build networks and collect data. To me it was like several random stumbles with Kanban, Agile and Scrum.
When coming to the verge of bankruptcy, I was lucky to have support from others. Not very much but it was enough for me to run a server and make a domain online. This fragile hope that made me win a prize from an accelerator and a free ticket back to San Francisco.
Interested more? Here is the outline.