When the lower boundary of an individual currency is exceeded, a trade recommendation is promoted to the execution engine. The execution engine is an externally focused machine learning algorithm that continuously compares the values of current DEX with the same currencies across nearly 70 exchanges by way of API.
The execution engine rank orders the buy and sell prices of target currencies net of transaction fees and automatically executes transactions where there is a spread greater than .05%. Like the currency valuation engine, the trade execution engine is operational 24/7 in search of optimum execution pricing and possible arbitrage cases, which are still widely pervasive across various cryptocurrency platforms.
CoinDEX founders have extensive backgrounds in the fields of traditional inferential statistics and machine learning. Machine Learning is the foundation of autonomous or semi-autonomous decision making in modern computing. Through a combination of statistics, probability, and classification models, variables can be fed into a machine with predictable decisions resulting in far shorter time than human capability. When The core decision criteria that drives this engine along with any algorithmic changes or updates are posted on the CoinDEX homepage.
It makes sense then that our product destiny is a blockchain optimized deployment that intersects with an automated learning model. For reasons described above, we have no doubt that machine learning goes well beyond the standard distributions and parameter estimates of traditional statistics to simulate a logical thought process in a changing environment. We firmly believe that computer science has demonstrated certain advantages over human beings such as memory, processing speed, and consistent output delivery.
Our purpose therefore, is to set forth an algorithm that adequately models our currency basket in context of its environment where each iteration results in a minimized loss function and automated buy sell execution based on this information. This is achieved through the synchronization of our neural net and machine learning platforms with continuous training, measurement, and convergence criteria as the executable output.
Neural network and machine learning applications in DEX:
CoinDEX has two machine learning algorithms that are synchronized to optimize portfolio value. The first is an internally focused neural network that functions as our currency valuation engine. This neural network integrates parallel processing for speed together with knowledge acquired from the training set, which for our purposes is the weighted historical valuation of crypto currencies maintained in our basket. For instance, the weighted input variables for a given crypto currency at t₀ would be: opening price, closing price, high price and low price from a continuous set of structured numeric data. The network is then organized into a system of three neural layers with normalized inputs fed forward in such a way that a single output of a ‘neuron’ in layer one becomes an input for all neurons in layer two. Layer two is also called the ‘hidden layer’ in the following diagram:
Each neuron is a formula that determines if the signal strength of the input will or won’t be transformed into an ‘activation,’ which adds to the weight of the overall output. Layer two feeds forward to the output layer, which is calibrated to minimize the average error function. The error is then distributed back across the network layers through a process called back propagation and the process is repeated until a threshold is reached where there is no further reduction in error. This is the DEX money manager which continuously evaluates individual basket holdings as described as well as the basket itself using individual holdings as inputs.