The AI Hardware Bottleneck
Blockchain and Artificial Intelligence (AI) have been described as a match made in heaven. The idea of the blockchain hosting trusted data and AI marketplaces has been suggested by a number of projects. The concept behind may seem solid but projects still have to deliver. Enter DeepBrain Chain. This morning, the company announced the launch of its blockchain based AI Training Network.
In recent years, AI has finally seemingly begun to live up to its potential, at least developmentally. We can thank various improvements in algorithms, such as the adoption of deep learning. Deep learning involves the implementation of neural networks, which are organized in several layers to mimic more closely the functioning of the human brain. The output of each layer serves as an input to the next layer, refining the model. The results often even surprise the experts.
Even more, one of the most important reasons for AI’s uptake is undoubtedly our improved capability to crunch vast amounts of training data. AI works best when specialized hardware, such as clusters of Graphics Processing Units (GPU) are able to process exceptionally large data sets. It is the necessity for this specialized hardware that keeps AI out of reach for many users and acts as a bottleneck for many applications.
Cloud-based resource sharing could be one of the most promising solutions to this problem. Blockchain technology has the potential to provide for secure hardware and data sharing and provides the necessary incentive layer to make such a model work in practice.
DeepBrain Chain aims to provide a blockchain powered worldwide AI cloud computing platform. Providing access to the AI Training Network serves as the first step to this goal. Users will be able to rent GPU resources in order to train their model. The purpose-built DBC will form the basis for an AI resource ecosystem with the following criteria:
- The DeepBrain Chain training infrastructure consists of GPU’s implementing the Nvidia CUDA API, allowing users to employ deep learning frameworks, such as TensorFlow, Caffe, and the Microsoft Cognitive Toolkit.
- Optimized load balancing provides high concurrency for a potentially large user base.
- Latency is optimized by reducing the number of resources that need to be coordinated to a minimum for each task.
- DeepBrain Chain uses encryption to protect the privacy of each participant.
- Container technology ensures flexible scaling and container re-deployment during peak times.
DeepBrain Chain today launches the first iteration of its service in the form of the AI training network. CBC tokens will be distributed to universities and research institutes working on AI applications, in order to facilitate the system’s uptake.