 Okay, hello everyone. Thank you for coming to my presentation. Today we are going to talk about our new research hunting the Isarin smart contract color-inspired insertion of protection attack and which based on our another research project at the name R2D2. Okay, my name is Tong Tong, Tang Tang, it's okay. My major research includes deep learning, enjoying malware analysis, and the type 2 phyrological and the ontology application. With deep learning, I have organized a deep learning 101 meetup in Taiwan and every month we share the impression of deep learning book. Now I'm working for Labour Mobile in Taiwan, and also I'm also a PhD candidate at an intelligent knowledge management lab at National Chenggong University in Taiwan. Okay, this is my outline. First of all, I will introduce who we are and how to land in our core deep learning research to the blockchain safety. Second, I will talk about the state of the blockchain included in Isarin and the smart contract, and I also show you the related work for its name for blockchain deep learning and the problem with the extent approach. Next, I will make a brief discussion of our deep learning solution for smart contract analysis. Also, I have prepared more demos for everyone. If you're interested in blockchain, our cryptocurrency, and the deep learning, hope you will like our demo. Okay, I'm from Taiwan. I think anyone has visited Taiwan. Oh, good, so you know Taipei 101? Yes, my office is located on Taipei 101 and our 18th tree flower. So anyone, if you come to Taiwan, come to Taiwan, please let me know. I can take you to 18th tree flower to see the sunset or anyone. We are the number one publisher of mobile tool apps in Google Play. Our corporates have reached 600 million global monthly active users in more than 200 countries. Here are some of our corporates just like security master, clean master. I think clean master is everyone, some someone know. And our blockchain corporates is coin master, rating token, and step wallet. So at least I will show you some demo. Okay, as we know, there are two popular blockchain technology. One is Bitcoin and another one is Ethereum. Bitcoin, I think everybody know what is Bitcoin, just like peer-to-peer electronic cash system. And Ethereum is the next generation, next generation encryption platform and which were launched in 2009 and 2015. Also, there are many related reports for example, Bumble Book, Business Work and the New York Times. The blockchain behind Bitcoin and Ethereum received huge attention from both industry and the academic. However, the security of the Ethereum smart contract has not received the same attention. Why? Let me show you some sheet. Let me make a short discussion of recently the Ethereum smart contract status this month. In this slide, we can find before July 14, the average daily growth of almost $1,000. Also, here are the usual growth trend from July 14 to July 23. And July 27, the average daily growth of almost $10,000. So why? Based on our research, here are some status reports. On July 17 and 18, we found there are 20,000 smart contracts have been generated. Among these, there are 10,000 smart contracts are the same. There are only generated $20,000, but $10,000 smart contracts are the same. And on July 23, we found there are some new smart contracts named Phone 3D and Morpho 3D, which based on punch schema. I will present our research which how to reserve the problem about the safety of the Ethereum smart contract. I thank everybody in this slide. We also know why deep learning and why machine learning. So I thank this slide just to show you as we know deep learning gets a huge amount of attention because of deep mind alpha goal. And the difference is that with machine learning and deep learning, the difference is that it didn't need virtual extractor. And when the amount of data is bigger, the performance is better. Machine learning and deep learning already has a wide range of applications, especially in security problems such as spam filter, binary detection and malware classification. However, most of the insulin security research still focus on analyze the control flow graph, our symbolic execution of the smart contract. Our proposed system is designed particularly for the security accumulation of contract with the minimum labor cost. Here are two related difference. One is finding the gritty and okay, a very recent study has analyzed nearly one million contracts. Among them, 34,000 and 200 contracts are guaranteed and nearly 4,000 contracts are particularly exportable. And here are another related worker is making smart contracts smart. Also, okay. Here are another related worker which are developed by IBM, Microsoft, Microsoft Research and the area. In our humble opinion, to arch secure contracts, the key step is to have through security information before deployment. Our proposed system is designed particularly for the security accumulation of contract with the minimum labor cost. Anyway, let's hear. We have proposed our related deep learning research with Android security. We have proposed our another research on OWASP EPSTECH 2017 and look, ransomware is there, larger scale ransomware detection with neck eye on LOXCOM 2017. Based on this research, we extend the core technology to the Israel smart contract. Before we start make this question of our research, why we are focused on the security of the Israel smart contract? Because you may, you may know emission coin over ICO. And based on crypto research report, 1,800 of ICOs were scammed. 1,800 were scammed. And 6,504 and 5% had gone dead. And 8% went on to trade on exchange. And the resources is from, from this EYL. Okay, so our data set collected from Israel scale, which provide the online and the bug smart contract. In this slide, we can find this smart contract has many, many bugs. At the same time, we can collect the by call of the smart contract. And here are this, this smart contract's by call. Okay, so our, our proposed methodology is we, here's the example by call of smart contract. We have developed a color representation for translate the by call of Israel smart contract development language into RGB color code and transfer them to a fixed size encoder image. For example, okay, 606060 is red, 96, green, 96, blue, 96. For example, and we can get this image. Here are some images. We can reach a color image and the image are fit to convolutional neural network training a model for the detection of malicious behavior flow of smart contract. Here are the benches of our methodology. Translate into RGB color image. And translation from its query code to image within 0.2 second. And the company and analysis an image within one second. Okay, such a translation is also featured by the more complex information. We found, we also found to more approach might be use escape our detection. So we address these two issue we apply in session B3 pre-trained model to train our benign and bugs smart contract model. We found the, okay, so everybody knows the in session B3 model just from this reference network in network. Now this is the concept, his concept. Okay, brief description of our system flow chart is show us this slide. Step one to five are offline phase for our internal development and the step six to seven are online phase where the user can interact with the system. More specifically after upload the by call and the invoke our list for API, the user can obtain the analysis result. This is our public level API screenshot, but I think this is project is too okay. And anyway, we at least we have more demo not just in slide. And here are our another screenshot is the example we will make more demo. Okay, this is our research and the developer environment. Our hardware is sitting is the media Titan be Titan SP and the GTS is one thousand eighteen GPU and the software sitting we use a media darker tensor floor and the media CUDA for example and the the research result and the data we be found on our website is R2D2.twman.org. We will show our research and the research data and the result. Okay, this is our experiment result. We before we start this research, we try to use LSnet, Google net and then such B sun in this slide you can find in such B sun have good result. Oh, also the scene is believing. I will show you some demo. We have we okay. If you and okay. Okay, okay. This this website is ESOS game or any AV smart country or the blockchain address you can find in this website and we can see this smart country address has his by call. Oh, okay. So we only to we have the the public website for scan smart country just like we can select any other smart country's address and you can visit our website rating token net and you only pass the smart country address and push scan now wait a minute the name work I check okay. Our database we are scanning and the last we will provide you some scan results. For example this smart country is 4.152 and we also have to provide detail result. We also can try to select another anyone anyone smart country address and then which past the address only past the address and the scan now. This is our public website and I will show you another prototype with our research. Our prototype research you need past the by call to my website I can find okay okay. This is our prototype not it's private website in our internal name work but you need to pass the by call and then send it to the smart country by call only three I think I check the website okay and we will when we finish the scan we will show you the benign our little function our acr receiver many many other scan results but if you interested in easterly smart country scan you can visit our public website okay okay back to our slide okay this is our public easterly smart country and the website rating token done okay we provide we provide auto called auditor offset distance test and transition monitor. If you interested in crypto currency we also have another product just like this coin master coin master we will provide three functions like smart inference and the profile management and the intelligent support so if you interested in you can scan this QR code okay this is another blockchain our related blockchain application its name step wallet okay uh okay now we have make a brief description of our research and the implement as a public assistant which apply convolutional neural network to the easterly smart country safety however there are too many smart country have been generated every day if we can find the master attention crypto token and the currency and then provide the detailed information to the user it will be better just like lending our core deep learning research to real user situation so we also have another this is our another ongoing research with deep learning we have collected many social network comment and the design new nlp model with lstm and convolutional neural network also we also have a private website based on our easterly smart country's safety research public opinion mind is the interesting issue if you want to discover ico social network attention our any other new trade status we will show you this website okay so we have collected more than four four thousand ico's information we collect their comment from their facebooker twitter and the telegram so the right side is positive with the social discussion and another one is negative with the social issue so if you want to find the interesting information you can find the detail page and we provide three days seven days ten days and 14 days the ico's trend with our desired nlp model and the convolutional neural network but this is our private website future more we will public this website and if you're interested in this project you can be the redding token.net to find more information just like this one you can find if i select her any day we can find this one good comment excellent interview everyone through and another bad comment we can any choose any any day this is our collector their information from their facebooker twitter comment and if you find seven days or the network okay just like this one you also can find the good comment and the bad comment okay the last one is because we've seen this model with lstm and the commercial neural neural network but everybody knows if we want to develop a deep learning model the data's label is the important thing so this project we also provide information to help us just like this one we can we will show you some comment from our nlp model and this is our petition score if you also agree you can put this one to this and send okay submit oh sorry this is a chinese version okay the next one okay oh i like this project uh based on our petition is 0.4 and maybe i think i like this project is good so i can select this part and then select and submit it's okay so this website will be public soon okay back to our our slide okay okay based on uh so if you are interested in our ongoing research before we launch the project website please help us to make the better analysis result okay so this is our conclusion uh according to our data there are one uh there are 15 000 smart countries produced on ethereum mainnet per day almost less than 30 percent is verified by ethers again in 2018 there has been 11.75 billion uses dollars in by ico project uh our goal is to optimize the amount of parameter network structure and release auto automatic verification tool and public resource for api so if you're interested in our project you can contact me by this email my personal website