The last few years have seen a meteoric rise in artificial intelligence research and development, with nearly $23 billion in financing distributed to startup companies across the globe between 2012-2016. Over that same period, large companies like Google, IBM and Facebook have all injected large sums of capital to launch their own AI groups.
While AI is an incredibly exciting, emerging field, there is a major issue that could potentially hinder startup and academic research programs. No, we’re not talking about the possibility of Skynet. What we’re interested in is the extreme computing power required to develop these machine learning systems, and how much that computing power potentially costs startup firms.
According to the MIT Technology Review, Renting 800 GPUs from Amazon’s cloud computing service to train a deep learning system for just a week would cost around $120,000 at the listed prices. You’d be hard-pressed to find many startup firms that could sustain that type of burn rate. We are clearly dealing with a scaling problem, and unique solutions may be necessary.
When considering alternatives to expensive cloud computing platforms, one need look no further than the blockchain. Specifically, DeepBrain Chain (DBC), a decentralized neural network that leverages mining nodes across the world to supply computational power. DBC’s vision is to provide a low-cost, private, flexible, safe and decentralized artificial intelligence computing platform for AI products.
Interested in DeepBrain Chain? Here’s a quick rundown of the project:
Platform & Development
The DeepBrain Chain network is built around a collection of mining nodes. These nodes can range from large mining nodes, such as those of a large mining pool, to medium-sized mining nodes using Azure and Ali cloud for mining, or even a home high-performance computer.
Once the system is online in Q2 2018, it will be pretty straightforward for miners to get started. All they need to do is install the DeepBrain Chain mining software to set up their node and begin earning DBC, which is traded via smart contracts based on NEO. Network rewards occur every hour and the total number of DBC produced by mining will be 5 billion. Every 5 years, the number of DBC obtained from mining will be halved.
We’re not going to get lost in the weeds with the specifics of DBC’s system architecture. Instead, we’ll highlight the elements that make the DBC project unique, as detailed in the DBC white paper.
Low cost: The core problem is the high cost of hardware input by artificial intelligence factories. Through the unique model of the DeepBrain Chain, 70% of the income of each mining node comes from mining DBC while 30% comes from the GAS cost. Artificial intelligence factories only need to pay for such 30%, i.e. GAS.
Optimization of neural network computing performance: DeepBrain Chain focuses on the service of artificial intelligence factories, and the current artificial intelligence products are developed on the basis of deep neural network as the core algorithm. DeepBrain Chain is currently optimized on CUDA GPU and plans to dock the current mainstream deep learning framework, such as TensorFlow, Caffe, CNTK, and so on.
Highly concurrent: Users of artificial intelligence factories are massive in number. DeepBrain Chain needs to facilitate high-performance computing while supporting a massive number of users. Through a unique load balancing technology, each node container can cooperate with each other to share concurrent pressure.
Low latency: While it is possible that the training time of neural network can be very long, all online user requests must be responded in seconds. This requires that each module of DeepBrain Chain is able to respond quickly, taking up as little resources as possible.
Privacy protection: To protect the privacy of each participant in the ecosystem, participants can freely determine the degree of information developed. We need to ensure via the encryption algorithm and the separation mechanism.
Flexible supply: Artificial intelligence factories’ user requests are not temporally homogeneous, and they are likely to be, say, ten times more frequent at the peak times than at the non-peak times. This issue needs to be effectively dealt with the burst traffic, which requires flexible expansion technology, so that the docker container can be automatically deployed. Fast replication is deployed to multiple idle nodes at the peak of traffic.
Automatic operation and maintenance: When a node container fails, it should be able to alert in a timely manner, remove the faulty node, and add a normal node.
The DeepBrain Chain team hails from China and looks to be well rounded given that they’ve been in business since 2012, and have successfully launched a number of AI products. On the downside, they currently lack quality English translations on their website and white paper, making for a fairly difficult read. Ideally, we’d look for that to be ironed out so potential investors aren’t turned away.
DeepBrain Chain (DBC) currently has a market cap of ~$47 million with a circulating supply of 900,000,000 DBC and a total supply of 10,000,000,000 DBC. It is important to note that a supply of this size could turn away potential investors.
DeepBrain Chain has developed one of the most interesting use cases for decentralized blockchain technology and is tackling a massive market. To properly build long-term value, DBC will need to penetrate the English markets, which will require a more thorough marketing approach.
Follow us: Telegram | Twitter | Newsletter
Disclaimer: The author(s) of this article may have a position in one or more of the securities mentioned above. This article is for informational purposes only and should not be taken as investment advice. Always conduct your own due diligence before making investments.