Abstract
Topic: DTN for vehicular applications with diversified demands
Challenges: distributed collaborative computing faces challenges in trusted resource cooperation and accurate decision making
Solution/Goal: blockchain-based vehicular digital twin architecture for distributed collaborative computing where blokchain facilitates the resource sharing and digital twin enables decision making

Structure of the paper:

Section II: Proposes the blockchain-based vehicular digital twin architecture.
Section III: Presents the resource allocation model.
Section IV: Describes several open issues.
Section V: Provides numerical results.
Section VI: Concludes and suggests further work.

Blockchain in 3 very brief facts:

"Blockchain utilizes consensus algorithm to manage ledger data, cryptography to ensure data access and transmission security, and smart contract to operate data automatically."

Architecture:

• cloud-edge-end collaboration architecture empowered by digital twin and blockchain (section II)
• based on the architecture, a resource allocation model (to provide accurate task scheduling and resource management decision) (section III)
  
  
  

⭐
Why chat thinks this paper has relevant information for my work:

Given your specific goals and the focus on creating a comprehensive Personal Digital Twin encompassing various physical objects, along with the emphasis on individual data control, the papers you've been reviewing may provide valuable insights. Here's why:

• Blockchain for Data Security and Control: The integration of blockchain in the vehicular digital twin architecture emphasizes security and privacy. Applying similar principles could enhance the security and control aspects of your proposed Personal Digital Twin system.

  
  

• Collaborative Computing and Resource Allocation: Understanding how tasks are offloaded, shared, and collaboratively processed in a vehicular network may offer insights into designing collaborative mechanisms for managing data across multiple digital twins in your context.

  
  

• Digital Twin Integration: The concept of digital twins for vehicles and RSUs, as discussed in the papers, aligns with your objective of creating digital twins for individual objects (watch, phone). The section on digital twin functionality and its interaction with blockchain might provide inspiration for your implementation.

  
  
  

What is Collaborative Computing?

Collaborative computing involves multiple computing entities working together to achieve a common goal. In the context of vehicular networks and digital twins:

Example Scenario: Vehicles, Road Side Units (RSUs), and possibly cloud servers collaborate to perform computational tasks or share resources within a vehicular network.

1) What is Computation Offloading?

Computation offloading refers to the process of transferring computational tasks or workloads from a local device to another computing resource. This is often done to optimize resource usage, improve performance, or achieve energy efficiency. In the context of vehicular networks and digital twins:

Example Scenario: A vehicle has a task that requires significant computational resources. Instead of processing the task locally on the vehicle's onboard computer, the computation is offloaded to a more powerful computing resource, such as a cloud server or an edge computing node.

Blockchain-based vehicular DT architecture

• vertical level vs horizontal level: in the vertical, vehicle can offload task to the cloud or edge, depending on the time-sensitiveness of it and in the horizontal, vehicles in the end can process their tasks locally or through surrounding vehicles

  
  

• DT builds the replica of each physical enitity in the virtual space, real time data is synchronized to the virtual space

  
  

• Trust guarantee: DT is dependent of frequent data interactions so data security and authenticity must be effectively ensured, hence blockchain - trustworthy because of its decentralized and tamper-resistent nature ⭐

  
  

• Communication Interaction: intra-twin vs inter-twin interaction - i think intra-twin is when a task vehicle sends its data and service demand to its DT, which then implements an allocation algorithm for obtaining the optimal decision. And inter-twin (even though in the text it seems they give this definition to intra-twin instead of inter but i think they made a mistake) is when data sharing can be implemented between different DT's ⭐

  
  

• Intelligence deployment: AI - the intelligence is mainly deployed in the cloud and the edge due to resource limitation in the vehicles ⭐

  
  
  
  
  

Computation Offloading in Blockchain Based vehicular DTN

• The blockcain-enables vehicular network can be mapped to the virtual network, which includes vehicle DT, blockchain DT and RSU - road site unit - DT
  

• Trust Evaluation (1- Identity trust, 2- Behaviour trust, 3- Ability trust)
  

• DT and Blockchain-empowered resource scheduling:

  
  
  
  • Blockchain-enabled resource scheduling
    

  • Digital Twin-enabled service guarantee: given the large scale and high dynamic of vehicular networks, they use multi-level DTN, which includes edge DT's and cloud DT's, deployed respectively in the edge and the cloud layer
    ⭐IMPORTANT - if i understood correctly, there is one cloud DT which aggregates information from multiple edge DT's, thus we have a hierarchy, could there be an extrapolation from the edge DT's to the device DT's and thus from the cloud DT to a PDT?
    

  • Artificial intelligence-based decision making: algorithm - Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm
    

Gaps:

• Connectivity:
  
  • Issue: High mobility of nodes leads to intermittent connectivity, disrupting data transmission between nodes. This affects consensus-building in the blockchain system and data interaction between physical entities and digital twins.
  • Consideration: Mobility of nodes should be considered in the blockchain-based vehicular digital twin system to ensure connectivity for successful data interaction.
    

• Resource Measurement: (only matters if my architecture is cloud-edge-end)
  
  • Issue: Vehicular computation resources are distributed across cloud, edge, and end, with various types (CPU, GPU, FPGA) and measurement units. Aggregating and mapping them to a unified measurement unit is crucial for efficient task scheduling and resource management.
  • Consideration: Developing a method to aggregate diverse computation resources and unify measurement units to facilitate efficient resource management.
    

• High Fidelity:
  
  • Issue: Achieving high fidelity in digital twins, reflecting the state and behavior of physical entities accurately, is challenging. Multi-physics multi-scale integrated modeling is proposed as a promising solution for accurate management decisions.
  • Consideration: Implementing multi-physics multi-scale integrated modeling to enhance the fidelity of digital twins and improve the accuracy of management decisions.
    

• Real-time:
  
  • Issue: Real-time performance of digital twins is crucial, dependent on efficient data interaction between physical entities and digital twins, and high computing performance. Reducing data transmission latency and optimizing the computation platform, data structure, and algorithm structure contribute to real-time capabilities.
  • Consideration: Enhancing data interaction efficiency and computing performance to achieve real-time capabilities in digital twin systems.
    

• Incentive Scheme:
  
  • Issue: Resource sharing and collaboration are vital for distributed computing, digital twin, and blockchain systems. Given the selfishness of nodes, an incentive scheme is necessary to ensure nodes share benefits. Evaluating node contributions is essential for effective incentive scheme design.
  • Consideration: Designing an incentive scheme to encourage resource sharing by evaluating and rewarding the contributions of nodes in distributed systems.

The requirements:

• A connectivity scheme between a virtual twin and its physical counterpart in this context (hdt) requires privacy, security, accuracy and the overall connectivity cost (the connectivity cost must be reduced to ensure timely synchronization between them)
  
  
  

Fact #1:

Traditional blockchain-enabled systems are characterized by high latency ⭐

The challenges:

• Ensuring reliable connectivity (mapping) between physical and virtual environments
• Insufficient possibilities for the physical-virtual environment synchronization to establish closed loops
• Lack of high fifelity and quantification models
• Difficulties in obtaining accurate predictions of complex physical systems
  

Federated multi-task learning:

Yes, federated multi-task learning is a machine learning technique. Let me break down the key components:

• Federated Learning: It is a machine learning approach where a model is trained across multiple decentralized edge devices (or servers) holding local data samples, without exchanging them. The model is trained collaboratively while keeping the data localized.
  Multi-Task Learning (MTL):
  

• Objective: In traditional machine learning, models are trained for a specific task. Multi-task learning, on the other hand, involves training a model to perform multiple tasks simultaneously.
  Benefits: MTL aims to leverage shared knowledge across tasks, potentially improving the model's performance on each task.
  Federated Multi-Task Learning:
  

• Integration: This technique combines federated learning and multi-task learning. It means training a multi-task model across decentralized devices or servers without centralizing the data.

  
  

  
  

• In this context: this technique is used to capture the impact of heterogeneous environments in the Human Digital Twin (HDT) system. It provides a personalized learning approach capable of delivering higher accuracy by considering multiple tasks simultaneously and across different entities. ( the term "higher accuracy" in the context of federated multi-task learning within an HDT system likely refers to the system's ability to improve the precision and reliability of predictions, analyses, or insights related to the health and well-being of the physical counterpart.)

  
  
  

"To address these issues associated with differentially private FL, especially in HDT, where data from PTs are arbitrarily heterogeneous with fundamental statistical heterogeneity issues, while also noting the presence of many external conditions that can influence the behaviour of each PT such as environment and genetic information, we propose DPFML framework following the privacy-aware multi-task learning approach, first discussed in [15]. This ensures a federated optimization of heterogeneous tasks while protecting the local model gradient information using DP"

Differential privacy:

Differential privacy is a privacy-preserving concept and mathematical framework in the field of data privacy and machine learning.
Here it is introduced as a key technique to ensure privacy in the communication and interaction between the physical (human) and virtual (digital twin) component, thus protecting sensitive personal or health data.

"Recently, differential privacy (DP) is being explored in FL to reduce the possibility of information leakages by hiding the contribution of each client during training thereby ensuring privacy guarantees."

PoMQ (proof of model quality)

To ensure that the final model from each LA is accurate and has not been modified through malicious activities, we propose a PoMQ consensus mechanism which offers the validation process in terms of the quality of the trained model during FML rather than solving computational-inefficient hashing puzzles, as in the proof of work. With the PoMQ, the validation process can be carried out before model updating and evolution in the virtual environment. Multiple validators are necessary to eliminate the possibility of malicious validation. These validators are responsible for validating the training quality of each model. After validation, each validator broadcasts its validation decision to other validators to reach a consensus. A virtual model is updated using any learned model only if the majority of V consent.

Intelligent IoT (IIoT), Digital Twins (DT) and the advances in mobile networks are now paving the path towards decentralized self-managed CPS (cyber-physical systems) in the industry. (DT permits mobile networks to provide adaptive and dynamic configurations for cooperative CPS. Moreover, trustworthy cooperation may be realized with blockchain.)

HFL - hierachical federated learning solves heterogeinity problem and enables multitask learning (not sure it solves the challenges in the previous branch)
This solution brings the following benefits:

• reduces overall data exchanged
• achieves bandwith efficiency
• enables intrusion detection
• increases the automation level between various factories
• supports digital twin in the smart industry
  
  

HFL allows for the splitting of FL tasks over two or more stages providing better management of the available resources (...) However, it also suffers from, possible attacks on the global model

1) A two-stage hierarchical FL solution (that uses DTs of industrial devices)

2) The adaptation of device FL coordination techniques (through an algorithm)

3) The use of blockchain to verify and validate uploaded trained models

4) The adaptation of model validation techniques achieved through a validator node selection algorithm.

could this be interesting?
"In [26], the authors propose a blockchain-assisted authentication method for IoT devices that belong to different industrial domains. The solution enables communication among the devices to complete manufacturing tasks securely. Given that blockchain is not efficient for scalable systems, the solution uses an off-blockchain storage technique to reduce the size of a block."

In a nutshell, how it goes down:

In essence, to realize Industry 4.0, DTs of the industrial IoT environment and devices are created. A DT model is created for the industrial devices that maps the physical status of the devices into a virtual space. This mapping is synchronized through consistent updates in real-time. A device’s physical state, in addition to the current and historical behaviours are mapped into its DT using manufactory installed sensors and actuators. The data collected is transmitted to the service edge node. The edge node will then create or update the DT of the industrial device accordingly and add this information to the blockchain. Any changes to the status of the device is added to the blockchain in the form of a block. This will ensure that integrity of the DT is maintained and cannot be manipulated by unauthorized entities.

interesting fact:
"Lightweight consensus algorithms are preferable to maintain mining simplicity and optimize the participant’s computing resources"

Problems in current blockchain-enabled FL systems when applying in DTN:

• a) DTs usually have limited computing resources, which can hardly support both consensus and FL training processes;
• b) since the local DT parameters are broadcast to all other DTs, the privacy issues are severe, while the falsified local model parameters can hardly be detected;
• c) a DTN is usually deployed in cloud servers with a virtual network interface card, which gives the DTN the potential to mitigate current low-efficiency aggregation issues.

Novel consensus algorithm

Proof of Federalism (PoF):

-> enables the simultaneous completion of consensus and FL training tasks and boosts the performance of DTNs.

asynchronous global aggregation algorithm and rollback mechanism of DTNs,

Interesting/Important quotes:
"In several industry areas, DT technology has been linked with blockchain technology to connect multiple DTs using distributed ledger technology (DLT)"