Recurrent Neural Networks in Crypto Trading- A Case Study of Agate Ecosystem
With the advent of the Recurrent Neural Networks (RNN), the crypto trade has been given a completely new dimension in terms of cryptocurrency value predictions as well as the stability of crypto assets given the volatile market fluctuations.
With millions of online stores globally, not many accept cryptocurrencies for trade transactions. A very small percentage of merchants do accept cryptocurrencies but only in the form of Bitcoin or Ethereum but not any other. The reasons for this usage reluctance is due to:
- Fluctuations in the market, thus preventing merchants from taking the risk.
- Volatility in price tag as tracking and maintaining various cryptocurrencies across multiple platforms is impractical and extremely difficult.
- The conventional online stores are designed to run on the Fiat economy. Hence integrating cryptocurrency payment gateways with these stores is extremely hard.
Agate– a new decentralised blockchain is based on the Recurrent Neural Network (RNN) model and offers a platform that integrates users, merchants and developers for cryptocurrency usage in daily payment and merchant transactions. Agate achieves this using an unique Agate Payment Gateway API that offers seamless integration to merchant stores using the state-of-the-art POS system that can be installed at the merchant end. The user and merchant apps add to the flexibility and dynamism to the ecosystem.
The E-commerce payment gateway app and the plug-ins are extremely easy to install and offers the merchants & online stores the capacity to accept cryptocurrencies as payment option.
The Agate payment gateway API allows anyone to create their own payment gateway. With easy plug-in installation options and to support the payment gateway API, Agate has planned to develop similar app and plug-in for 10 online shopping platforms and release them for free. The shopping sites include Woocommerce, Squarespace Online Stores, Magneto, WixStores, MonsterCommerce, PrestaShop, Weebly eCommerce, OpenCart, and SAP Hybris. These would account for over 76% of the world’s online stores thus giving a massive boost to the acceptance of cryptocurrencies for everyday trade and transactions.
Agate has 4 important components that are based on the Recurrent Neural Network (RNN) model thereby providing an ecosystem that offers solutions to maximise the gain for any users:
- Agate AI Engine: Assists and guides users as to the best transaction time to maximise gains.
- Trading Bot: Lets users create conditions for transactions with the bot simply executing the transactions when the conditions are met. The bot also guides users to stabilise profits despite any external market fluctuations.
- iBucket: A decentralised wallet to transact iFiat (Agate’s internal cryptocurrency) and also acts as a repository for cryptocurrencies and Fiats. The iBucket lets users convert iFiats to Fiats and stores them for future trade executions as per the user’s discretion by analysing the market fluctuation and offering recommended suggestions to enhance profit. iBucket would be available on all of Agate’s apps.
- Agate Payment Gateway API & Plug-ins: The payment gateway lets users create their default payment gateway and with easy to install plug-ins, merchants world-over can accept cryptocurrencies without any hassles by simply connecting to the Agate Payment Gateway API.
Agate AI Engine Case Study
Agate’s AI team is influenced by Andrej Karpathy, Tal Perry, and Christopher Olah’s methodologies in creating a dimensional database (db) consisting of 4000 columns and 300 rows. On vectorising it against the vector, a new vector is generated of a size which is only 300. The numbers in the matrix are set at random and going with the ‘deep learning’ methodology, the excel spreadsheet updates automatically whenever the numbers are changed.
The RNN deep learning algorithm operates on sequences. Since RNN has a form of internal memory, it remember what it has seen previously and uses that memory to decide what would be the next operational input. This way, the RNN remembers that it is nested within an intended scope and thus maximises the result output.
This kind of RNN memory is called as Long Short-term Memory (LSTM) as was demonstrated by Christopher Olah. This cleverly designed memory allows it to choose selectively from what it remembers, decide on the ones to forget, and lets it decide the quantity of remembered memory to be given out as output.
By combing the db with the LSTM and altering the Karpathy code, Agate has simulated trading data for over 8 months with 3 different coins (Bitcoin, Ethereum, Ripple) and has found that for over 200 withdrawals using the Agate AI Engine, users were always able to save anywhere between 5.2% to 7.4% more than the same frequency random withdrawals when used without the engine.
Combining trade data of Bitcoin and Ethereum and simulating predictions for over 100 trades, it was found that a mere variance of 0.172787 when compared with the actual values was noted. This yet again proves the accuracy and integrity of the Agate AI engine.
Agate is not just a new decentralised blockchain platform but a revolution in itself that visions of integrating the crypto trade with the mainstream economy and is well on its way to making that reality manifest in the days to come.