investor bank extending blockchains for collaborative

17
Tara Salman DSS18 Extending Blockchains for Collaborative Decision Making and Risk Assessment Applications Tara Salman Supervised by Prof. Raj Jain Investor Bank Bank Investor Mining Node Cloud Provider Scan this to download the slides

Upload: others

Post on 29-Dec-2021

5 views

Category:

Documents


0 download

TRANSCRIPT

Tara SalmanDSS18

Extending Blockchains for Collaborative Decision Making and Risk Assessment

ApplicationsTara Salman

Supervised by Prof. Raj Jain

Investor

Bank

BankInvestor

Mining Node

Cloud Provider

Scan this to download the slides

Tara SalmanDSS18

Proposal, Publications and Talks 1. Proposal: “Extending Blockchain Technology- a Novel Paradigm and its Applications to

Cybersecurity and Fintech,” QNRF accepted proposal, November 2018, Co-PI: Raj Jain.

2. Raj Jain, Tara Salman, “Probabilistic Blockchains for Decision-Making Applications,”submitted WashU patent, October 2018.

3. Tara Salman, Raj Jain, and Lav Gupta, "Probabilistic Blockchains: A BlockchainParadigm for Collaborative Decision-Making," 9th IEEE Annual Ubiquitous Computing,Electronics & Mobile Communication Conference (UEMCON 2018), New York, NY,November 8-10, 2018.

4. Raj Jain, "Extending Blockchains for Risk Management and Decision Making," invitedtalk at Innovation and Breakthrough Forum 2018, Hong Kong, Nov. 9, 2018.

2

Tara SalmanDSS18

Big Pictureq Blockchains have been used as a distributed database

Ø Records, contracts, tracking objectsq Problem: Current blockchains have no way to process these vast

amount of data Ø Process only validityØ i.e. Blockchains are not intelligent

q Our work: Extend blockchains to knowledge-based blockchain Ø Process blockchain data to achieve useful knowledge, Ø Make decisions using the stored data (intelligent)

3

Tara SalmanDSS18

This Presentationq Focus: How to extend the technology to process data within the blockchain

q Introduce probabilistic blockchain Ø A paradigm to make probabilistic, collaborative, and consensus decisions

q Risk assessment applications: Ø Networking: intrusion detection, malware detection, Ø Financial: stock market prediction, ..

4

Tara SalmanDSS18

Outlineq Centralized vs Distributed Ledger

q Blockchain Technology

q Can the Blockchains be Enhanced?

q Probabilistic Blockchain

q Empirical Results: Blockchain-based IDS

5

Tara SalmanDSS18

Centralized vs Distributed Ledger

6

Centralized Bank • Centralized registry• Single point of failure• Easier to hack

Distributed Cryptocurrency• Decentralized registry• No single point of failure • Very difficult to hack

Current trend is distributed. Blockchain technology is one way to distribute.

Tara SalmanDSS18

Blockchain Technologyq Peer to peer, distributed network

q Users interact by transactions

Ø Example: Alice send money to Bob

q Miners validate transactions and create blocks

q Blocks are linked by previous block ID

q Blockchain nodes validate blocks and add to their

chain

q Chain is hold by many nodes à globally distributed Blockchain

Network

7

Transaction (tx)

Alice

Miner

Bob

Blockchain nodeMiner

Blockchain node

Tara SalmanDSS18

Advantage of Blockchains

1. Distributed nature: Not easy to hack

2. Decentralized consensus: No single decision point

3. Cryptographically secure: Transactions and blocks signed, can’t deny

interactions in the system

q Use cases: banking, asset exchange, land registry, IoT, healthcare, …

8

Tara SalmanDSS18

Can the Blockchains be Enhanced? q Blockchain is currently used as a distributed database

Ø Data storage of valid transactions, e.g. money transactionsq Transactions are all deterministic

Ø Not being used for probabilistic events Ø E.g. “google stock market will rise with 80%”

q However, decision making is probabilisticq Blockchains are unsuitable for decision making applicationsq e.g. Collaborative prediction of google stock market tomorrow

9

Tara SalmanDSS18

Our Work: Probabilistic Blockchainq Objectives

Ø Deterministic event recording à probabilistic event recording Ø Blockchain dataà Knowledge Ø Decision making applications suitability

q HowØ Changing the architecture of transactions to store decisionsØ Changing the architecture of blocks to summarize decisionsØ Introducing summary function to make knowledge of the block’s

transactions

10

Tara SalmanDSS18

Probabilistic Blockchain

11

Timestamp

Block 0

TransactionTransactio

Timestamp

Block n-1

Timestamp

Block n

TransactionTransactioRecordTransactionTransactioRecordRecord

Timestamp

Block 0

TransactionTransaction 1

Timestamp

Block n-1

Timestamp

Block n

TransactionTransactioEvent i { DecisionProbability }

TransactionTransaction 3Event i { DecisionProbability }

Event i { DecisionProbability }

Transactions Summary { event i,Summary function}

Transactions Summary { event i,Summary function}

Transactions Summary { event i,Summary function}

Blockchain nodes validate the block and construct the chain

Block generators validate transactions and generate blocks

Users broadcast transactions or contracts (Records)

Traditional chain

Probabilistic chain

Blockchain nodes validate the block and construct the chain

Block generators validate transactions, create a knowledgeable summary and generate blocks

Users broadcast transactions (Decisions/opinions)

Tara SalmanDSS18

Malware Detection Exampleq Does file j contain a malware?

1. A user ask for a certain file providing its URL

2. Different distributed malware detectors inspect the file and reach there decisions

3. The block has the summary for file j

12

Timestamp

Block 0

TransactionTransaction 1

Timestamp

Block n-1

Timestamp

Block n

TransactionTransactioTransactionTransactionTransaction 3

Transaction 4Transaction 2

Transaction Summary { file j,Summary function of 1 and 2}

Transaction Summary { file j,Summary function}

Transaction Summary { file j,Summary function of 1,2,3,4 }

user

Malware inspectors

Malware inspectors

Malware inspectors

Malware inspectors

user

Transaction 1AgentID=1, File=j, P(File)=10

Transaction 3AgentID=3, File=j, P(File)= 40

Transaction 4AgentID=1, File=j, P(File)=50Inspection of file j

(URL)

Transaction 2AgentID=2, File=j, P(File)=20

Tara SalmanDSS18

Summary Functionq Application Dependentq Any reasonable function that summarizes multiple decisions can

be used q Examples

Ø Mean, Median, ModeØ 2nd Moment Ø A vector of multiple Ø Any statistical algorithm, data mining algorithm, machine

learning algorithm Ø …

13

Tara SalmanDSS18

Empirical Results: Blockchain-based IDSq Machine learning trained intrusion detectors make decisions about a

certain flow

q The summary function used is the mean

! (#$%&' () *+$(,(%-)) = 01∑!3 (#$%&' () *+$(,(%-))

P consensus decision, !3 individual decision

q Settings: Same dataset, different machine learning algorithms

q Accuracy is the evaluation metric used14

Tara SalmanDSS18

Different Machine Learning Algorithms

Attack type DoS attack taken fromUNSW-NB15 dataset

Number of algorithms Random Forest (RF) Decision Tree (DT) Linear Regression (LR)

Number of detection agents

1000

15

Collaborative decision making is better than any individual decision

99.05 98.89

93.92

99.06

919293949596979899100

RandomForest

DecisionTree

LinearRegression

Probabilisticblockchains

Perc

ent A

ccur

acy

Algorithms

Tara SalmanDSS18

Conclusionq Blockchain technology has many features that made it so popular

q Currently, mostly used as a distributed database

q Current blockchains is not suitable for decision making

q We propose probabilistic blockchains, a blockchain extension to process data

and make collaborative decisions

q A case study of blockchain-based IDS showed the feasibility of the proposed

approach

16

Tara SalmanDSS18 17

Tara Salman [email protected]://sites.wustl.edu/tarasalman/

Scan QR code to download the slides