One benefit of blockchain’s decentralized technology is its application beyond cryptocurrencies. We are applying the techology to machine learning – to enable a smarter “thinking” blockchain process. Most machine learning models that have been used have an underlying assumption that some global knowledge is required by its algorithms to function (see fig 1). These are centralized distributed, where the entire data set is loaded into a cloud of distributed nodes so that the two main functions of machine learning are performed in one large library of functions ETL/Exploration and Model Training/Parameter Tuning. Spark, Watson, Azure all use this platform based approach and we are also biased towards methods that have been known to work.
gny.io’s DistributedDeep Learning breaks this pattern of one large platform library by creating two of the smallest configurable self-learning unsupervised neural net nodes – ETL node and ML node – and distributing these nodes into each block of the block chain to have them teach themselves the solution to each problem. The one conceptual problem though of machine learning is that error detection requires global knowledge that gets backpropagated to its constituents. This requirement though is fixed in gny.io’s DistributedDeep Learning systems.
Therefore we prefer the term ‘localized models’ and not ‘model free models’, even though both mean a measure of lack of intelligence of constituent parts. In general though, this is similar to the concept of ‘parsimony’, in that we seek the simplest of mechanisms that give rise to emergent properties of prediction. Gny.io is not a crushed up small machine learning library; it is one node of a huge distributed brain.
Gny.io has configurable ETL microservices and configurable machine learning microservices that can read the entire chain of data or it can read the current block data. Gny.io’s microservices uses deep learning which is a class of machine learning algorithms in the form of a neural network that uses a cascade of layers (tiers) of processing units to extract features from data and make predictive guesses about new data. The smallest ML node system varies the weights and biases to see if a better outcome is obtained using a neural network.