Matrix AI Network : An analysis

In today’s world you have to sell before you create , that way if it fails you already pocketed the cost — A Guy in a bar

This is a message to those who will call this FUD

This report isn’t a personal attack against the product that is researched or its team this isn’t FUD as you may think if you have a bag of this product it’s your responsibility to own it , we don’t care about your bags let alone your opinion .

We would put a disclaimer here to say that we’re not investors and blablabla … point being this isn’t financial advice this is a research of the technological aspects of the product .

We will skip the part where we discuss the team but let it be a task for the reader , to verify the credentials of an academic all you have to do is search for his name in https://scholar.google.com or https://arxiv.org where most of the academic research is published .

The Innovations

In page 5 and 6 respectively the paper lists a number of innovations which according to the definition of the word “innovation” means they’re the first to bring it to its existence .

1 — Base Protocol

The first innovation in base protocol section is “random clustering based voting for a network of delegates” clustering means “grouping” in machine learning clustering is a statistical method to group samples of a dataset into groups that share the same features , if I were to cluster fruits for example I can group them based on shape or color for example :

• Red apples,blood oranges , strawberries , cranberries belong to the cluster of color “Red”
• Avocado, Green Apples,Kiwi belong to the cluster of color “Green”

Here I clustered a basket of fruits based on a feature “color” .

• Random Clustering based voting … doesn’t have an actual meaning the only thing one can understand is that they randomly select delegates based on votes but even then the process isn’t RANDOM since the selection is based on votes . Nothing innovative here (we’ll come back later)
• “Introduction of a separate control chain” , they mean side-chains nothing new here again .
• “Evolutionary parameter optimization” , evolution here is a method of optimization , optimization is the field of study of how to select the best solution to a problem evolution optimization is a technique that inspires from the “evolution process of species” we simulate possible solutions as a set of individual species and using technique such mutation,reproduction and selection we try to create better solutions by combining previous ones these solutions are measured with a fitness function . Here’s a good example of an evolutionary algorithm to understand it better . Evolutionary parameter optimization means that the blockchain will self adapt itself using the same techniques for example if you remember the Bitcoin Block debate , in MAN’s blockchain apparently there won’t be any debate the blockchain itself will choose to change its parameters to sustain the usage of itself . Is this possible , we don’t think so , in fact let’s suppose its possible then somehow the blockchain will change its block size to 8MB this means a change in the software code which means the blockchain is somehow sentient and is able to change its own source code and recompile it ... this technology itself is worth more than market capitalization of “Crypto” .

Smart Contracts automatic code generation is called program synthesis , now this is clearly an innovation but you have to understand that the research is still in its early phases to quote Neel Kant it is still far from being solved or even competitive with most existing methods .

Security :

• Formal Verification : This is similar to what Quantstamp does , formal verification means that programs can be mathematically proven to be correct or not . Deep Learning based auditing means that they trained a neural network to spot mistakes in code this is what a company does , one company is specializing in this yet Matrix Network can do all of the above .
• Binary Code Checking : This is what TrailOfBits does nothing innovative in fact Binary Code auditing is how software vulnerabilities in closed-source products work , you pick a binary file and you reverse engineer it trying to find mistakes .
• Credit-score based … bla bla : Apparently it’s a reputation system .

Transactions :

• Pattern Match : Pattern Matching is the process of finding whether a sequence of tokens (numbers,words,strings) match a pattern for example Regular Expressions do pattern matching using a backtracking algorithm for example your banks website uses a regular expression to make sure your SSN is valid , “Allowing data exchange and identification…” this sequence of words only exists in this white-paper , we don’t quite understand what this sentence means and welcome your explanation .
• Transaction search engine .. privacy based tracking system .. : They apparently want to have the ability to track transaction in other words they want to make it non private , this is a violation of user privacy ,similar to this . This isn’t an innovation this is against the idea of a “TRUSTLESS SYSTEM” this is against Satoshi’s true vision .

We have no clue how people who believe in crypto and blockchain can support such kind of things , apparently if it sells it’s all good .

Mining :

• Bayesian and Deep Learning as PoW : Proof Of Work will be in the form of neural networks training in other words the possibility for miners to train neural networks at the expense of the researchers or the users without a privacy framework to protect the data , the intellectual property NO RESEARCHER in his sane mind will do this , check https://www.openmined.org/ a project that want to do the same using Homomorphic Encryption , an encryption scheme that supports mathematical operations in other words using a Homomorphic Scheme such as Paillier I can encrypt two numbers and give them to you , you can add them and send me back the result , encrypted, and I can decrypt it and the result will be correct .

The not so technical description

In their whitepaper, the team describes a hybrid PoS+PoW scheme based on stochastic network contraction , “stochastic network contraction ” .

A stochastic process is a random probabilistic process , we think they reefer to what they called “random clustering” ,apparently one buzz word that means nothing wasn’t enough .

Let’s have a look at a lousy made diagram and explain it .

lousy made diagram using Microsoft word 2003 ©

The node delegation is said to use PoS , meaning that stake and coin maturity is what qualifies you to be a miner which contradicts the word “random” ,an algorithm apparently will pick a group of delegate node , this network of delegate node constitutes a new network that will do the Proof Of Work .

Then the transaction is only broadcasted to the delegated nodes — Page 7

The PoW is allocated to the delegate node which can then send it to the nodes that voted for it .

A Markov Chain procedure is a probabilistic state machine , imagine a washing machine , in the real world a washing machine is deterministic meaning that it knows exactly what’s it going to do when it starts , a probabilistic machine is the contrary when it starts at IDLE it has 25% to go to the DRY state and 75% to the WASH state , this process of moving from one state to another is described using a Markov Chain in mathematics .

So suppose nodes jointly generate a random seed , how can a third party verify the “random seed” since it’s actually random ? — can anyone explain .

The algorithm described is also proprietary meaning that it won’t be open-source and in no way a user can verify if it’s true or it does exist …

Deep Learning Based Automatic Code Generation

For a treat about the subject of deep learning you’re welcome to read this book http://www.deeplearningbook.org/ if you’re not mathematically inclined you can watch Siraj Raval does a great job explaining the subject in easier terms .

Matrix says that a user can describe a smart contract using a Domain Specific Language and a Neural Network will convert it to actual code , a 1-dimensional Convolutional Neural Network (Conv1D) network can indeed extract text features from a sequence and a Recurrent Neural Network (RNN) can indeed generate a text sequence , this part of the white paper is the only one that employs actual vocabulary anyone can understand . +1

Deep Learning based verification of the smart-contracts

finally a good drawn diagram from the Internet

A GAN network is a Generative Adversarial Network , it’s a neural network capable of generating new inputs (capable of creating images for example) after learning what images are can it generate or hack ? We don’t know and this whole particular approach will seem useless if you used formal verification it means you mathematically proved that a program is correct (it always produces the same output given the same input), so why do you need to simulate a hacking against it ? Well because the formal verification is done on the static code( the program implementation/ definition) we also need to protect against dynamic attacks the ones that happen against the virtual machine that runs the code .

After this part of the white paper is sane from a technical perspective .

Natural Evolution of Blockchain Parameters

Well until this part , here they try to explain how the blockchain is actually going to optimize itself .

Remember that the blockchain is nothing but a database except that it’s distributed and verified , now they don’t use the term “evolutionary” but they introduce the term Reinforcement Learning which is a different thing ,Reinforcement Learning is another area of Machine Learning that inspires from behaviorial psyhology .An agent (in this case an algorithm) are ought to take actions so as to maximize a reward function ( AlphaGo for example uses reinforcement learning in addition to other methods) ,they will apparently put a neural network inside the blockchain this neural network will learn a reward function so it can make better choices they don’t discuss how the changes will actually happen but a lousy diagram is nonetheless present .

Conclusion

We think that from a technical perspective Matrix AI Network is where ambitions turned into delusions the technical paper isn’t technical in itself it’s more of a “put the most buzzwords and invent your own if you need to” paper , we expected more from such a “technical” team .

A lot of caffeine was consumed into one setting if you would like consider putting a tip in the jar , if not we are glad you read the whole report and thank you for your undivided attention , remember to always #DYOR.

Ethereum : 0x7EAd1FF67B888E59c7ca3C8b006Ab804601b025a

Bitcoin : 1N8hYmGcHQi5xcevcTTGtNU7Si1w4BMw3U

P.S : We weren’t paid by Quantstamp we just refeer to it’s existence .