Questions of the last AMA [2019-10-24 Youtube AMA] :
Q1: When do you expect the first real retail order to run over MAN network for AI services?
For Matrix AI Network and our mining network, we are developing our decentralised cloud computing platform. As we know, this platform will be finished next year. I think we can expect the first order to happen in the second quarter of 2020.
Q2: How do you intend to attract customers both on the corporate and the public side? In which market do you see the most potential for AI services and why?
We are a public blockchain network, but our AI timing is similar to the industry basic solution. Our focus right now is covering our customers. We do a lot of things to attract customers. One of the more important ones is we collaborate with public cloud services, like Amazon and Tsinghua Unigroup. We already assigned a strategic agreement with one of these, and we plan to sign more in the future. Because this cloud service providers areserving many customers. It’s partially because they have a strong momentum to attract customers. Of course, they will have a need for data method solutions. So this cloud service provider will naturally introduce customers for us. That’s one way to attract customers. On the public side, we are investing, but you can see the decentralised computing platform is an important way to attract customers.
When people are using our platform, naturally they will seek a solution from us. I think this is the way to attract customers. For now, our focus is on predicted maintenance. The basic idea is to use AI methods to forecast the future status of machines, this is because the machines we use to do our jobs will fail from time to time. But, if we gather data from this machine, we can predict and forecast the future of this machine. So that can take accurate actions to prevent accidents and reduce maintenance costs, and so on. This is very important work because every year a lot of money is spent on machine fitness, over 600 billion dollars. This is a big deal, and it’s even much better if we can lower this. It’s a lot. I think it’s an issue we can focus on.
Q3: What will be the real-world application of AI mining?
In the first iteration of our AI mining, we just offer picture stitching and picture analysis algorithms in our mining system. If the industry needs bigger picture stitching or analysis. They will need our AI mining service, then, we can offer service to our AI traffic [network]. The traffic department can use our AI mining to develop their applications to potentially analyse car data and tell people how crowded the area is. I think that in the first stage, we can offer services to these industries.
Q4: What aspect of decentralised AI do you think people underestimate or undervalue the most?
Currently, we know how to use different models to solve real problems. The problem however is that we need experts to tune the models and train the parameters. That’s not something everybody can do. It requires a lot of expertise. The problem with this is we do not have so many technical people to do this, but everybody needs his own model to solve his problem.Currently, machine learning is coming to our rescue. The idea is we use reinforced learning to design a new model to solve the problem. Basically, we can do this on language learning on a large scale, and it will be less expensive. Everyone will be able to use this method to solve their own problem.
Q5: In a general sense, could you give us your views on how FPGA miners are coming along?
FPGA is a very useful device for computing solutions. But for mining, probably it is not that proper. The reason is for mining algorithms like bitcoin, FPGA cannot provide high enough mining power. They are not comparable to basic solutions. For other mining algorithms that are mining intensive,there isn’t much bandwidth available, so computing is also less featured. For other algorithms, the problem is high-end algorithms are very expensive.They [FPGAs] are actually more expensive than GPUs. This makes FPGA the least favorite solution for the industry.
Q6: What area in AI (relating to Matrix) have you seen the most unexpected progress?
As far as I know, it’s audio-based machine learning. And by audio I mean sound signals of the natural world, sounds that are not human language sounds. Currently, most solutions are targeting video signals and English analysis. We have to remember that audio signals are at least as important as video signals to convey useful information.
Currently, we found that deep learning technology can be very useful for audio analysis. I know of a few such projects that are using audio signal analysis to detect failures in machines. When a machine is about to fail, there would be an unusual noise. We can use this mechanism to detect failures in machines and predict future problems.Another interesting project is that we can use the chirps of birds to analyse the distribution of birds in the sky above an airport. You know, birds’ impact on airplanes are a big concern. But with sound signals, we can figure out what height the birds are at in the sky. With this information, we can figure out a way to reduce bird impact. Audio analysis could be a big thing in the future, it’s a field of great potential and real-world use.
Q7: What are the biggest challenges facing Matrix AI in both the technological sense and the regulatory sense? How do you plan to overcome these?
As for the development part, the biggest problem for us is that the product we are developing is very new to the world. This means we don’t have too much previous experience from other people to learn from. In this situation, every time we start a new project, we need to do a lot of research, read a lot of papers and do a lot of discussions. After we finish a project, we need to do a lot of testing to make sure it really works and is safe.As for the regulatory part, we are trying our best to get cooperation with different industries and companies. We want more people to use our technology to upgrade their technology or their services. By now, I think more and more countries are warming up to blockchain technology. I think it’s a very good focus.
Q8: What will be the real-world application of object detection on the Matrix Wallet?
The application on the Wallet is a personal application. One such application is you can use the object detecter as a calorie calculator. When you’re about to have a meal, you can take a picture of the dish and the calorie detection application can figure out which food is in the meal. The picture/information is then loaded onto a database. You can then figure out how many calories you’ve consumed in the meal.
Q9: What is your strategy to raise Masternodes’ ROI?
For this question, I think the key point is our decentralized cloud computing platform. When the platform launches, the miners can rent their computing power to users who need the AI computing power when they haven’t been selected as Masternodes. That decides their mining revenue, and they also get a fee from renting their computing power. In this situation, the more users there are using our platform, the more ROI Masternodes will get.
Q10: How many nodes/how much computational power do you estimate will be needed to attain stability of the AI network of Matrix’s platform?
Actually, we don’t have a predefined number for this. We are currently on the language learning application, I guess for this to work, we need 200 CPUs (GPUs?) at least so that users will have an acceptable user experience. Of course, the more the better.
Q11: How many people are there in the Matrix team. What is Matrix team’s development progress?
There are about 70 people in our team. Also, we are making great progress on two sides. One is the launching itself about CPU we have a team working on this. You can know the announcement of the progress so far.Another team we have is the AI team, we are currently working on a few very interesting and supposedly high-value projects like predictive maintenance. For this ,we use AI technology to the operation of machines and try to figure out the best way to maintain these machines. Another is wind power prediction and other predictions also related to the industry but less powerful.
Q12: How does Matrix AI help to diagnose lung cancer?
Previously we have developed something like a lung cancer diagnosis procedure. Basically, it’s image-based AI models. We built an image classification model that can detect early signs of lung cancer. I don’t remember the exact number, but the accuracy is about 90%, so it’s better than human diagnosis.
Q13: When will the mining machine be for sale, and how much will it be worth?
Some of our cooperation companies are designing machines that are just for Matrix AI Network. I think next month they will release some information about how to sell the machines and how much.
Q14: Do we have an estimation of when GPU mining will be available for testing by the community?
The last release didn’t have compiled binaries, and I don’t believe anyone in the community was able to test. Maybe in a week or two, we’re going to do some tests ourselves. This is a major update of the mining algorithm, so in a week or two, we’ll release something new and run tests.
Q15: Where do you see Matrix in five years’ time?
We’re really trying to build a distributed cloud platform, so it’s really hard to say. We wish for our decentralised cloud computing system to be big and we want to offer the cheapest computing power to the public.We will try to implement some auto-deep learning typological search, and we we wish all this can happen in maybe one to two years. Five years is a really long time, so it’s hard to say. We have very high goals. We have the confidence to make those happen in the next five years.