Cube Whitepaper

Friday, June 15, 2018
Download document
Save for later
Add to list is a project of OpenBook

CUB cube


cube Remark We can not open the detailed technology to protect the intellectual property rights of CUBE security technology. CONNECTED CAR/AUTONOMOUS VEHICLE SECURITY AND INFORMATION PLATFORM WITH BLOCKCHAIN TECHNOLOGY CUBE is a network security company based on Blockchain At present, the automobile market of 1.7 trillion USD, has reached an important turning point. The rapid expansion of car connec vity has brought a en on to the significance of network security. Newly manufactured cars from 2020 will be produced with a built-in CDMA. Connected cars are expected to be under constant cyber-a acks, similar to the malicious a acks towards network connected PCs. CUBE is preparing to embrace a pivotal role in the rapidly changing vehicle market. CUBE’S THREE MAIN SECURITY BLOCKS 010101010101 010101010101 010101010101 A acks Cloud Endpoint Blockchain CUBE Mul -layer Protec on Connected cars and autonomous vehicles require a myriad of communica on hubs: vehicle to vehicle (V2V), vehicle to network(V2N), vehicle to Infrastructure(V2I), vehicle to device(V2D) and et cetera. The increased connec vity increases the user convenience, yet also increases the a ack surface for possible malicious malware a acks. Thus, network security is pivotal in stabilizing the connec vity in vehicles. CUBE u lizes Blockchain, Deep Learning (in-depth learning) based AI-intelligence, Quantum hash encryp on, Endpoint protec on and Cloud-based Intelligence to ensure secure automo ve network. CUBE Intelligence Page 3

cube Check Box Sandbox Cube Auto Blockchain Server Firewall Cube Box I. BLOCKCHAIN SECURITY The conven onal blockchain technology grants security whilst maintaining simplicity, but has its limita on on transmi ng large files. CUBE Blockchain security resolves the drawbacks of conven onal blockchain, through u lizing peer-to-peer hypermedia protocol and asymmetric encryp on. CUBE Blockchain technology will provide next genera on security for the automobile industry, taking advantage of the impossibleto-hack nature of blockchain, whilst providing rapid data transfer. 2. CUBEBOX ENDPOINT SECURITY TECHNOLOGY Connected cars engage in mul lateral communica on from various external sources: vehicle to network(V2N), vehicle to vehicle(V2V), vehicle to Infrastructure(V2I), and vehicle to device(V2D). CUBE Endpoint Security Technology generates endpoints at all external connec on points for protec on. CUBE’s unique technology uses hash to scan and dis nguish over 300 million ‘known’ a acks with only 10MBs of disk space. In fact, even with our light-weight so ware, it holds the fastest endpoint processing speed amongst the exis ng endpoint security product. CUBE Intelligence Page 4

cube 3. CLOUD CHECK BOX TECHNOLOGY CUBE Cloud Technology can dis nguish ‘unknown’ a acks through transmi ng suspicious files to the cloud malware database, the Check Box. Our Cloud Check box is partnered with more than 50,000 corpora ons, that uploads and shares newly encountered malware on a transna onal database. ‘Unknown’ a acks that are le uniden fied are sent to Sandbox, where files run in a virtual vehicle environment. Based on the verdict from the Sandbox analysis, the results are uploaded to the database and will be shared across other vehicles. TECHNOLOGY APPLICATION AUTOMOTIVE MARKET TREND CUBE incorporates blockchain, endpoint and Cloud-based CheckBox to implement following business models. 1. DATA BUSINESS According to the Mckinsey Report, the value of data collected from the car exceeds that of the car itself. CUBEBOX enables anyone to become data producer through simply plugging our device in a car. CUBE will establish an ecosystem of automobile data sales and aim to become the world’s largest Big Data company. CUBE Intelligence Page 5

cube PRODUCERS OTA OTA OTA OTA OTA OTA OTA OTA OTA DATA SERVICES Automakers Automo ve Mechanics Traffic Informa on Providers Gas Companies Insurance Companies Car Sharing Companies *Above companies’ logos are not related to actual contracts. It is a representa on of poten al partners. DATA CONSUMERS THERE ARE TWO KINDS OF PARTS. DATA PROVIDERS AND DATA CONSUMERS Data Providers The CUBE token links driving-related informa on from the car to data consumers who need this informa on. Drivers generate this informa on as they drive and receive CUBE tokens in return. CUBE then links it into a blockchain. Data providers can use this to tap into services such as “Over the Air”(OTA) autonomous car security, and mileage insurance services from data consumers. Data Consumers Data consumers who need this informa on will provide their services in exchange of CUBE tokens. Automobile generated informa on, is sent, u lized, and received by all nodes. Data consumers can also provide OTA services and autonomous car driving support informa on to data providers. This circula on is a key principle of the CUBE token. CUBE’s tech can be applied to different blockchains and improve data security and sharing in various real world situa ons. CUBE Intelligence Page 6

cube THESE DATA PROVIDERS AND DATA CONSUMERS EXCHANGE DATA AND SERVICES AS BELOW: Cube Token Usage Plan 1- CUBE 3- CUBE 2- SERVICE 4- INFORMATION Data Consumer Affiliate i.e. Gas Discount, Data Producer i.e. Big Data, (i.e. Automo ve Companies, (i.e. Gas Sta on, Car Repair Shop, (Date Producer, CUBE Owners) Traffic Informa on Providers, Insurance Companies) Car Purchase Discount, Driving Records, Insurance Companies, Government, Lab) Mileage Insurance Fuel Efficiency, Vehicle Condi on 1. CUBE token owners can receive services at a CUBE affiliates. 2. These affiliates can be gas sta ons, car dealers, repair shops, insurance companies, etc., and can provide services such as gasoline discount, vehicle maintenance, vehicle purchase discount, insurance discount, and cash back service. 3. Informa on consumers such as automobile companies, insurance companies, traffic informa on providers can pay cube to individuals and purchase driving related data. 4. CUBE token owners can become informa on producers by installing CUBE OTA. Driving informa on is big data which includes driving records, vehicle condi on, fuel consump on and so on. 2. OTA BUSINESS OTA OTA technology provides remote automo ve so ware upgrade. This will upgrade the func onality of in-vehicle so ware. CUBE’s blockchain based OTA technology will provide remote so ware diagnos c and the installa on of bug patches. 3. AUTOMOTIVE SECURITY BUSINESS OTA According to ‘Cybersecurity Ventures’, the cybersecurity market grew from 3.5 billion USD in 2004 to 120 billion dollars in 2017, more than 35-fold within a short me span. The switch to connected cars will result in burgeoning of the demand for network security. CUBE will provide robust security, that is vehicle- op mized and based on blockchain, endpoint and cloud checkbox technology CUBE Intelligence Page 7

cube CUBE’S BLOCKCHAIN AND SECURITY Fig.1. Most hackable points of autonomous car. INTER-COMMUNICATION NETWORK SECURITY An autonomous vehicle must acquire as much informa on as possible about its surroundings to operate the vehicle alone. Such informa on may be path informa on for naviga on, traffic informa on, or data for upda ng an old firmware of an autonomous vehicle. Receiving such a variety of data can be very helpful in opera ng autonomous vehicles, but at the same me increases the risk of malicious intrusions. V2V, which means Vehicle to Vehicle communica on, is an important func on to make an autonomous car safer. In a connected car, the vehicle receives various vehicle data points from other nearby vehicles. V2V helps to operate autonomous vehicles in various aspects. There are two types of V2V informa on, short range and long range. Representa ve informa on of the short range can be the distance between the vehicles and the behavior data of the driver. Long range informa on includes road condi ons, accident informa on on the road, and traffic informa on. CUBE Intelligence Page 8

cube IOT SECURITY Many IoT devices are used to operate autonomous vehicles. The most representa ve is the “Guide Assist IoT,” which will be applied to smart roads. This IoT informs the autonomous car of its current posi on, receives speed and driving informa on from the autonomous vehicle, and sends this informa on to the clients who need it. While a variety of technologies are available to ensure safety when an autonomous vehicle is operated, the technology within the vehicle alone is not enough. The most obvious method is to install the IoT on the road where the autonomous vehicle is running. The autonomous vehicle is constantly in communica on with the IoT on the smart road. In this case, it is necessary to cer fy that the IoT of the smart road is the authorized IoT. The biggest problem here is the speed of authen ca on. An IoT should be manufactured at low cost because it must be distributed in large quan es. Therefore, we cannot expect a high level of processing power. In par cular, it is impossible for a car running in the public chain of this type to operate IoT. Therefore, a method of authen ca ng a chain closer to real- me than real- me authen ca on is needed. CUBE sees this IoT’s real- me authen ca on method as one of the important future development factors. INTRA-COMMUNICATION NETWORK SECURITY Automakers need to communicate with autonomous cars con nually. Most important is naviga on rou ng informa on, which includes traffic informa on. For completely autonomous self-driving cars, the car should receive the rou ng informa on from the automaker’s traffic management centre. Even though the car has the naviga on map data, it should receive the best route informa on, which comes with live traffic informa on. The poten al danger is a malicious a acker with fake traffic informa on, such as fake traffic or a fake road block. CUBE Intelligence Page 9

cube An autonomous car has a much more complicated Electronic Control Unit, which includes Break Control Unit (BCU), Transmission Control Unit (TCU), Wheel Control Unit (WCU), and many other units to control self-driving func ons. An ECU being penetrated by a malicious a acker could be seriously dangerous. A great problem arises in terms of security because of the network between automakers and the gateway of autonomous cars. An automo ve company should check each autonomous car's IoT devices to make sure that every autonomous car's ECU works without problem. At the same me, an automo ve company should upgrade their automo ve car's firmware remotely through the network. These checks and upgrades must be done regularly as all of these network connec ons will make the automo ve cars vulnerable to malicious a acks. 1. BLOCKCHAIN LAYER The key to blockchain is that the technology ensures trust. CUBE uses blockchain technology to ensure the security of autonomous mobile networks. But there are various difficul es in applying tradi onal blockchain to autonomous vehicle safety. The problem of blockchain is the slow speed and the low scalability. At the heart of blockchain is securing trust with technology. So far, no technology has been able to prevent hacking 100% in the network. But the blockchain has shown a remarkable ability to prevent hacking completely without any failure in the past decade. By solving the slow speed problem and the scalability issues that the blockchain has, we will be able to provide the highest level of autonomous car security CUBE Intelligence Page 10

cube 1.1 THE PROBLEM OF CONVENTIONAL SECURITY AND NEED FOR BLOCKCHAIN There are three major problems in tradi onal automo ve technology. The amount of data that the car must process is tens of mes larger than the data amount of other virtual currencies. CENTRALIZATION ISSUE The amount of data received from the outside while driving is es mated to be more than 4Terabite per day. Such a large amount of data causes a serious problem. Centralisa on reduces the opera on speed of the CPU and eventually stops the system if the number of cars increases. PRIVACY ISSUE The centralised approach eventually threatens the privacy of the driver. If someone can access the central server, you can access the personal iden fica on of all drivers as well as the driving record. SAFETY ISSUE A malicious a ack on a car's network cannot be blocked 100% by current security methods. Alterna ves must be created in order to reach maximum security. Considering the weakness of conven onal security methods, CUBE adopted blockchain as a key security pla orm for autonomous car. Blockchain is a distributed database that maintains a growing list of blocks that are chained to each other. Blockchain is managed by a distributed peer to peer network. There are various difficul es in applying tradi onal blockchain to autonomous vehicle safety. Blockchain instan a ons suffers from high overhead and low scalability. All transac ons and blocks are broadcast to the en re network which results in extremely large packets. CUBE Intelligence Page 11

cube 1.2 CUBE AS A HYBRID BLOCKCHAIN CUBE's autonomous vehicle security is based on decentralised blockchain technology. Each en ty is a node, making it basically the same as the security systems running on Ethereum. However, in the case of an autonomous car, there is a limit to the use of exis ng blockchain methods. Due to the blockchain’s slow speed and scalability problems, the blockchain technique cannot be used “as-is.” For this reason, CUBE has been preparing a hybrid chain. Ali Dorri et al., ‘BlockChain: A Distributed Solu on to Automo ve Security and Privacy’, IEEE Communica ons Magazine, 2017, pp2-3 CUBE is composed of hybrid blockchains that use public and private blockchains together. When datafying a vehicle’s drive, tons of data is generated, though the data varies depending on how the data was accumulated. Even when coun ng only the driving informa on and the peripherally recognised informa on used in datafica on, enormous amounts of data (up to 4TB) are generated every day. This takes a lot of me to process or share. By contrast, vehicles require fast data processing, transmission, and receipt to prevent accidents. The security of current autonomous vehicles is inefficient, in terms of velocity and volume, when it comes to handling by the public blockchain. To date, we have seen Hyperledger, the Linux Founda on’s Umbrella, and R3, in addi on to more than 75 banking consor ums around the world, as examples of private blockchains. CUBE cons tutes the private blockchain of autonomous cars, opera ng as a private blockchain to solve the problems of velocity related to data processing and volume. This, however, requires a very high level of trust in cri cally important elements, such as firmware upgrades by automakers. Those elements that require such a high level of trust will rely on public blockchains, such as Ethereum. In summary, CUBE uses its own private blockchain, and also uses a public blockchain for cri cally important elements; thus, CUBE is a hybrid blockchain that can be used for all aspects of autonomous cars. CUBE Intelligence Page 12

cube 1.3 CUBE’S AUTOMOTIVE BLOCKCHAIN MAINNET CUBE's ul mate goal is to build an automo ve blockchain mainnet, which will require two steps. The first step is Over-the-Air (OTA). OTA is a technology that remotely updates a car's firmware. OTA can easily and remotely update many bugs in the car. These updates do not require high chain speeds since as they only take about an hour; fast blockchain hashes are not seen. Therefore, blockchain speed is not a major issue. CUBE will complete these technologies in 2018. The second step is the autonomous car's security pla orm. The security of an autonomous car requires high-speed confirma on of the incoming informa on. It does not have to be too quick to receive simple naviga on path informa on; however, obstacle informa on and road informa on in front of the car require very fast valida on. Therefore, CUBE plans to apply one method of deep learning for this rapid valida on. The applica on of the stochas c gradient descent (SGD) method, which is quicker than gradient descent in deep running, is applied. SGD is a bit slower than slow perfec on and faster processing. Facebook's Mark Zuckerberg says, "Done is be er than perfect". Gradient descent uses the full-batch method to check all the cases and to make decisions; however, it takes too much me. SGD uses the mini-batch method for trial and error, but at a much faster rate. Deep learning can provide a variety of applicable technologies for making automo ve blockchain. Momentum can be applied for faster blockchain speed processing. The incoming informa on is considered validated once it is the same. Nesterov's accelerated gradient (NAG) first predicts the incoming informa on, and then calculates and checks the incoming informa on. AdaGrad can be applied when you have not heard anything. AdaGrad is a way to search more closely because it has not visited, yet. In this case, you can apply AdaDelta to prevent it from stopping because it will slow down if you validate it too precisely. Considering all these factors, Adam can be applied to a blockchain. CUBE Intelligence Page 13

cube 1.4 CUBE GOVERNANCE ISSUE CUBE’s governance will come from the par cipa ng automakers and owners of autonomous vehicles. Each par cipant becomes a node and has suffrage in the main decision-making process. The cost of this transac on will be paid to the carmakers as Proof of Share (POS). Automakers will have exclusive rights to data upgrades that require high reliability, such as super nodes firmware for each autonomous vehicle. 1.5 DAAP BASED ON CUBE PLATFORM CUBE is a hybrid blockchain pla orm that allows vehicles to easily share reliable data peer to peer (p2p). On this pla orm, a variety of future Internet of Things (IoTs) (e.g., guidance assist IoT, traffic informa on assist IoT, collision avoidance IoT, etc.) can be easily shared with trust. 1.6 CUBE’S BLOCKCHAIN SECURITY OPERATION CUBE solves these limita ons of tradi onal BC technology with two concepts of segmenta on and permission. In the opera on of autonomous vehicles, many IoTs provide informa on to autonomous vehicles. This informa on may be transmi ed directly between the IoT and the vehicle or may be transmi ed through a center that controls the IoT. The invader may try to interfere between the traffic centre or IoT and autonomous vehicle by taking access to the main network. Eventually, the invader can take control of the so ware binary with an inten on of introducing malware into a huge number of vehicles. If an a acker enters a new informa on hash value is not associated with a chain of exis ng blocks, it is considered a malicious a ack. In case of important update informa on, the CUBE pla orm and the automaker require mul ple signatures and only perform remote update if the two private keys match. CUBE Intelligence Page 14

cube CUBE CHAIN Figure 2 The data sent from the IoT to the car includes informa on such as loca on informa on and road condi ons. Conversely, the informa on sent from the car to the IoT includes speed informa on and vehicle status informa on. For the safety of autonomous vehicles, CUBE uses blockchain to handle communica on between these vehicles and the IoT. Each of the IoTs becomes a node and the autonomous vehicle becomes a node, as well. The data transfer between these cars and the IoT is considered a transac on. CUBE Intelligence Page 15

cube SUPER NODE Figure 3 The essen al informa on used in another autonomous vehicle is self-driving data provided by the automaker, the traffic management center, and so on. The automaker must remotely monitor the status of the vehicle from autonomous vehicles and remotely upgrade the firmware. The autonomous car should receive only the authen cated data from the authen cated centre. The authen cated centre should always be connected with the blockchain and should be updated con nuously. Most malicious en es will disguise themselves using these centres, therefore this node is very important and needs to be trusted. Blockchain was originally characterised as permission-free, but since communica on with autonomous vehicles requires the highest security, only autonomous vehicles should be allowed to communicate with permission. Therefore, unlike the case with normal blockchain, the data should be processed based on permission. To allow only authorised en es to access a node, CUBE uses the concept of a super node. A super node plays an important role, such as providing informa on to an autonomous vehicle and upgrading it, unlike a general autonomous vehicle. These super nodes can be specified by the automaker or the government. CUBE Intelligence Page 16

cube SUB BLOCK Figure 4 As seen in Figure 4, each sub-block works just like the current blockchain. It updates every nodes’ transac on, such as traffic informa on, automakers’ hardware upgrade and naviga on route informa on. The size of the sub-block is rela vely small, making the me to add a new block much faster than the current method. 1.7 HAND SHAKING If the autonomous vehicle's driving informa on needs to be handed over from one base sta on to another, the data may be interrupted in the mean me. In this case, the two base sta ons should nego ate so that the data can be transmi ed well so that the data is not disconnected. These two protocols are called hand shaking. The handshaking process generally occurs to set up guidelines for communica on when a system tries to communicate with another device. When a computer communicates with another device like a modem, printer, or network server, it needs to handshake with it to establish a connec on. Handshaking can adjust factors that are suitable for systems and equipment at all ends including coding alphabet, informa on transfer rate, interrupt procedure, parity and some other hardware features. CUBE Intelligence Page 17

cube Handshaking is a technique of communica on between two en es. However, within TCP/IP RFCs, the term "handshake" is most commonly used to reference the TCP three-way handshake. A common handshaking protocol may include the receiver delivering the message meaning –“I got your previous message and you can send me the next one” " A more complex handshaking protocol might allow the sender to ask the receiver if it is ready to receive or for the receiver to reply with a nega ve acknowledgment meaning “I did not receive your previous message properly, kindly send it again” 1.8 SINGLE-SIGNATURE AND MULTI-SIGNATURE The way in which informa on comes into your car depends on how important the informa on is. If the informa on is related to entertainment, it can be accepted without any confirma on. However, the same level of traffic informa on is secured with a single signature. If it is important informa on, such as upgrading the car's firmware, a single signature is not enough. In this case, the automo ve manufacturer and the security pla orm of the CUBE must be approved at the same me; a mul signature method is necessary for security approval. Ali Dorri et al has made the desirable validity checking method as follows: “All transac ons are broadcast to all OBMs. An OBM checks the validity of the received transac on by verifying the affixed signature(s). If the transac on is valid, then it is stored in a pool of valid transac ons which will be collated to form a block with a pre-defined block size, i.e., the total number of transac ons stored in the block.” “A mul sig transac on that arrives at the OBM may yet need to be signed by the recipient, par cularly when the recipient belongs to the cluster of that OBM. Each OBM retains a catalog of PK pairs that produce nodes which are permi ed to communicate with each other. The cluster members (i.e., overlay nodes) upload key pairs to the key list of their OBM to allow other overlay nodes to access them In case the OBM discovers a PK pair in its catalog that goes with the PKs in the opera on then it processes the transac on to the matching node which has uploaded the main pair. Or else, the opera on is transmi ed to further OBMs. Ali Dorri et al., ‘BlockChain: A Distributed Solu on to Automo ve Security and Privacy’, IEEE Communica ons Magazine, 2017, p.3 CUBE Intelligence Page 18

cube 2. ARTIFICIAL INTELLIGENCE (AI) DEEP LEARNING LAYER To enhance the security level, ar ficial intelligence (AI) is applied at this stage. Recently, a acks on hackers’ networks with malicious intent have rapidly been evolving. Un l now, however, cybersecurity technology has been a passive method of collec ng vaccines to defend against the a acks that have already been perpetrated. Cube has developed deep learning network security where a method is chosen based on predic ons of malicious a acks that will occur, rather than a passive approach that uses defensive (instead of ac ve) methods. CUBE will con nue to learn how malicious a ackers have a acked the network to prevent future a acks. CUBE’S SELF-TAUGHT LEARNING An invasion detec on system thinks of a common kind of a ach situa on, where affected data packets are inserted into the in-vehicle Controller Area Network (CAN) vehicle. CUBE Intelligence Page 19

cube Ar ficial neural networks have not worked well in the past to protect against these a acks, which is why new methods are needed. Three catalysts cause and maintain the explosions of malicious data—new mathema cal computa ons are the spark, big data represent the fuel, and massive computa on can be viewed as the horsepower. Cube uses previous cases of malicious a acks to recognise them. Cube allows the AI to train the Cube pla orm on cases of previous malicious a acks and predict hundreds of millions of new possible malicious a acks through this reinforcement learning. It then creates a defense system for each case. Cube employs TensorFlow, an open-source library built by Google. There are many other ways of implemen ng deep running, but at present, TensorFlow has the best posi on in the market. In addi on, there are a lot of data going on while watching the source code, which makes TensorFlow the most advantageous library. TensorFlow is an open-source so ware library for numerical computa on using dataflow graphs; one reason for its popularity is that it can be developed with Python. A graph is a connec on between one node and another, while a dataflow graph is an opera on. This edge is data to be easily said; this is called a data array. TensorFlow enables calcula ons through this process, and the TensorFlow run me is a cross-pla orm library. Google designed TensorFlow for large-scale distributed training and inference, but it is also flexible enough to support experimenta on with new machine learning models and system-level op misa on. CUBE Intelligence Page 20

cube Cube builds the first graph in TensorFlow. First, it can create a “placeholder” node, and each node becomes a placeholder. When a placeholder is created, it passes the value to the “feed data” as the graph runs through the session. This graph is then executed and updated as needed, or it returns an output indica ng whether a malicious a ack has occurred. A large quan ty of data must be input regarding previous malicious a acks in order to make the output more accurate. CUBE’s neural network is ini ally trained by being fed large amounts of data. Training consists of providing input and asking the network what the output should be. For example, to build a network for iden fying malicious a acks, the ini al training may include a series of past malicious a acks. Each input is accompanied by the matching iden fica on. Providing the answers allows the model to adjust its internal weigh ngs to learn how to do its job be er. For example, if nodes A1, B1, and C1 tell node D1 that the current input data represent a malicious a ack, while node E1 says they are normal data, and the training program confirms a malicious a ack, CUBE will decrease the weight it assigns to E1's input and increase the weight given to A1, B1, and C1. CUBE Intelligence Page 21

cube CUBE’S SELF-TAUGHT LEARNING Cube uses a neural network to determine the malicious a acks expected in the future. In fact, neural networks do not always provide good results. There are three problems with such networks, namely underfi ng, low speed, and overfi ng. Underfi ng refers to a problem whereby the learning is insufficient. Low speed is an issue where it takes too long to learn. Finally, overfi ng means that the network is inflexible, even when it has learned. These three problems make a neural network’s results unreliable. UNDERFITTING The first problem, underfi ng, means that learning is not effec ve enough. A neural networklearns via backpropaga on. It updates itself by repea ng the processes of differen a on, mul plica on, and adding, and all in vice versa. The problem, however, is that a sigmoid func on is used for ac va on. We have found a repeated issue with the vanishing gradient phenomenon: As the layer becomes deeper, the updates disappear. Therefore, underfi ng occurs, which results in poor fi ng. Instead of a vanishing sigmoid, Cube uses an ac va on func on that does not disappear. Specifically, rec fied linear units (ReLUs) are used as the ac va on func on. This solves the problems related to the vanishing gradient. LOW SPEED The second problem to be solved for neural networks is the low speed problem. Exis ng neural networks have used gradient descent (GD) to op mise the weigh ng parameters. The gradient is obtained from the current weight of the loss (or cost) func on, and it is updated to reduce the loss. To put it briefly, this is wrong (Slide 35). There are hundreds of millions of training data on hacking, and if we consider those hundreds of millions of hits each me we hit a mark, the network will slow down significantly. This raises the following ques on: Could there not be a faster op miser than GD? Here, we find the stochas c gradient descent (SGD) method. The concept of SGD is that it is a method that can be completed quickly, even if the results are not perfectly accurate. As Mark Zuckerberg has said, “Done is be er than perfect.” Gradient descent is a method that takes one step a er a full batch; in contrast, SGD uses mini-batches and moves forward step by step. CUBE Intelligence Page 22

cube Cube employs Adam as an improved version of SGD. Combining the advantages of RMSProp + Momentum, this solu on is suitable in terms of both the step size and direc on. In the future, we can consider a aching NAG to Adam instead of Momentum. OVERFITTING The final problem with neural networks is their inflexibility. To solve this problem, we can deliberately omit informa on or turn off the intermediate nodes when training a neural network. Dropout lets us learn what the important factors are without obsessing over specific parts. Thus, the problem of overfi ng relates to obtaining flexibility without dropout. Cube solves the three problems related to neural networks, namely underfi ng, low speed, and overfi ng. The problem that Cube is trying to solve is not snapshot data like images, but sequen al data that discriminate whether the signals are malicious a acks. Hence, Cube uses a recent neural network (RNN) or long short-term memory (LSTM). CUBE Intelligence Page 23

cube TRANSPARENCY POLICY CUBE considers transparency to be of the utmost importance. CUBE has established policy for transparent opera ons and will execute the following: 1.) CUBE shall publish monthly opera onal reports and financial reports to contributors to share the opera onal status of the company a er token distribu on. 2.) When hiring new staff members, such as new developers, CUBE shall build up a valida on process as well as thoroughly examining por olios from past and set the reward policies in accordance with their abili es. 3.) The company's budget shall be ghtly managed so that it would always be possible to operate and manage the company without addi onal funding for more than three years. CUBE Intelligence Page 24

cube CONTRIBUTORS COMMUNICATION 1.) CUBE shall send a monthly report every month to share important company’s update, a er the token distribu on. 2.) In case of an important occasion, CUBE must share the important ma ers by email immediately. 3.) Follow CUBE’s social media accounts for live company updates and to communicate directly with the team. CUBE Intelligence Page 25

cube TOKEN VALUATION OPERATION POLICY To provide its contributors with be er usage and u lity, Cube will be operated as follows. 1. The execu ve managers of CUBE are under the effect of Lock-Up System, which means they are not en tled to token sale for one year. The lock-up policy is intended to make sure the company focuses on the development of CUBE pla orm and company’s growth. 2. Cube shall strictly control the use of its budget for transparent opera on. At least 2/3 of the beginning budget must remain a er one year of funding. In case of any event which requires using more than 1/3 of the beginning budget, the approval of the board and the contributors par cipa ng in the ballot must be at least one-half. CUBE Intelligence Page 26

cube REFERENCE 1. G.V.Assche, ‘Quantum Cryptography and Secret-Key Dis lla on’, Cambridge University Press, 2006 2. C.Huang, Y.Shi, ‘Quantum hashing is maximally secure against classical leakage’, University of Michigan, 2017 3. Ali Dorri et al., ‘BlockChain: A Distributed Solu on to Automo ve Security and Privacy’, IEEE Communica ons Magazine, 2017, p.3 4. Mark V. Slusar, ‘Insurance system related to a vehicle-to-vehicle communica on system’, US 9390451 B1, Jul 12, 2016 CUBE Intelligence Page 27

cube h ps:// Social Pages © CUBE Intelligence, All Rights Reserved.

CUB Website: h ps:// Follow us on: h ps:// h ps://twi h ps:// h ps:// h ps:// h ps:// Join us on: h ps:// Contact: [email protected] cube