1 of 28

Fragile & Vulnerable�Prone to Performance Issues�Easier to Attack & Compromise�Can be Compromised by Botnets & DDoS Attacks, etc.�Hence Research Projects such as IoTAC to Achieve�Better Security + Performance ���������

IoT Networks

1

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

2 of 28

- A Combination of IP Nodes & Other Communication Modalities�- Contain Many Low Cost Nodes – Low Computational Power – Factory Initialization�- Some Nodes May be Battery Operated or Rechargeable�- Channel Sharing Among Some of the Nodes (e.g. UWB, MAC, ..)�- Large Networks that are Difficult to Coordinate and Self-Regulate�- Some Nodes May Transmit Periodically�- Special Gateways May Be Used & Overall a Greater Variety of Nodes���������

IoT Networks

2

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

3 of 28

Internet of Things Addressed by IoTAC Research�

  • The Massive Access Problem & the QDTP Algorithm

E. Gelenbe & K. Sigman (Theory – ICC ‘22)

E. Gelenbe, T. Czachorski, D. Marek (Trace Driven Simulation & Diffusions – MASCOTS’21)

​

  • Using the G-Network to Detect Different Types of Attacks

E. Gelenbe & M. Nakip (MASCOTS’21,

IEEE Access Dec. ’22)

​

​

3

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

4 of 28

What is the IoT Massive Access Problem?�IoT Traffic Characteristics

4

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

5 of 28

�IoT Massive Access��« Disastrous » Consequence on Queue Length at IoT Gateway�

5

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

6 of 28

What is the Massive Access Problem?�

  • The Problem: IoT devices transmit to a Gateway at relatively synchronized instants, causing massive congestion, packet loss and missed deadlines

​

  • Scheduling IoT devices is very costly: two-way communication and computation overhead

​

  • My QDTP invention forwards packets in a stable manner, resulting in tiny queuing at the Gateway and tiny queues at the Many IoT devices

6

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

7 of 28

�Massive Access Problem��Probability of Missed Deadlines vs Load in Number of IoT Devices �for Different Number of « Slots (Deadline Length) »�

7

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

8 of 28

Solving the Massive Access Problem �by Traffic Shaping with QDTP�QDTP: Quasi-Deterministic Transmission Policy�tn+1= an+1 if an+1>tn+ D, = tn+D if an+1< tn+D�Wn+1QDTP-IoTD = tn+1-an+1= [WnQDTP-IoTD -(an+1-an)+D]+ ��

8

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

9 of 28

Solve Massive Access: �Traffic Shaping with QDTP�tn+1= an+1 if an+1>tn+D� = tn+D if an+1< tn+D�Wn+1QDTP-IoTD= tn+1-an+1=[WnQDTP-IoTD -(an+1-an)+D]+

Key Theorem If D < Sn then

​

Wn QDTP-IoTD + WnQDTP-IoTG < Wn.FIFO

9

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

10 of 28

Solving Massive Access via QDTP

10

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

11 of 28

Solve Massive Access via QDTP�Congestion & Deadline Missing Disappear !!

11

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

12 of 28

Solve Massive Access via QDTP�Congestion & Deadline Missing Disappear !!

12

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

13 of 28

​

​

​

  • G-Networks (1990 .. onwards) are a strong generalization of Open Jackson Queueing Networks
  • Jackson Queueing Networks: N-Service Stations, Poisson External Arrivals of Customers, Exponential Service Times, Markovian Movement Between Queues, Probabilistic Departures (including to outside the network) from Service Stations after Service
    • Chapman Kolmogorov Differential-Difference Equations
    • Product Form Stationary Solution Obtained from Linear Traffic Equations

- G-Networks: Negative Customers, Triggers, Batch Removal, Resets -- Initially Inspired by Neural Networks

    • Product Form Stationary Solution from Non-Linear Traffic Equations
    • Universal Approximators for Continuous and Bounded Functions
    • Gradient Descent computation of Time Complexity O(n3)

​

​

13

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

14 of 28

​

​

  • G-Networks: Major Generalization of Open Jackson Queueing Networks
  • Simplest Case « Positive and Negative Customers »
    • N-Service Stations, Poisson External Arrivals of Pos & Neg Customers
    • For Positive Customers: Exponential Service Times & Markovian Movement Between Queues, Probabilistic Departures (including to outside the network) from Service Stations after Service, Probability of Pos. Customer Changing into Negative Customer when Moving to another Queue
    • Arriving Negative Customer Mutually Cancels One Positive Customer
    • Negative Customer Vanishes if it Arrives to an Empty Queue .. No Negative Queue (this is easily extended using Multiple Classes)
    • Chapman Kolmogorov Differential-Difference Equations
    • Product Form Stationary Solution with Non-Linear Traffic Equations

14

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

15 of 28

​

​

  • G-Networks: Major Generalization of Open Jackson Queueing Networks
  • Several Extensions
    • N-Service Stations, Poisson External Arrivals of Pos, Neg & Other Customers
    • Multiple Classes with Interaction Rules – Certain Negative Classes Eliminate Certain Positive Classes (but cannot affect other positive classes)
    • Batch Removal (1982): Negative Customers Probabilistically Eliminate a Batch of Positives, also works with Multiple Classes
    • Triggered Customer Movement (1983): Triggers are a 3rd Type of Customer that Instantaneously Moves Positive Customers to Other Queues, possibly in a Sequence of Moves with Synchronized Interactions, also Multiple Class relations
    • Chapman Kolmogorov Differential-Difference Equations
    • Product Form Stationary Solution with Non-Linear Traffic Equations

15

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

16 of 28

​

​

  • G-Networks: Major Generalization of Open Jackson Queueing Networks
  • Mathematical Properties
    • Product Form Stationary Solution with Non-Linear Traffic Equations
    • Existence and Uniqueness of Solutions of Non-Linear Traffic Equations (Gelenbe-Schassberger)
    • O(N3) Gradient Descent Algorithm for Optimization and Learning
    • Universal Approximation Capability of the Random Neural Network (Gelenbe & Mao & Li) – Stationary Solution of G-Networks can Approximate any Continuous and Bounded Function where
      • Inputs are the Arrival Rates of Negative and Positive Customers
      • Outputs are the Probabiities that the Queues are Busy .. i.e. the Probabilities that the « Neurons are Excited » – Multiplied by their Output Rates

​

16

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

17 of 28

​

  • Numerous Applications (* Industry Licensing or Beta Testing):
    • Image Compression using Deep Learning
    • Video Compression using Deep Learning
    • Travelling Salesman Optimization & Other Graph Problems
    • MRI Tumor Detection *
    • Adaptive Network Routing with Reinforcement Learning *
    • Digital Coding and Decoding
    • Vehicle Recognition from Infrared Images *
    • Perceived Video Quality Assessment
    • Network Attack Detection *
    • Wireless Channel Allocation

17

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

18 of 28

IoTAC �Example of Progress�

Auto-Associative Dense RNN for IoT Botnet Attack Detection

  • M. Nakip and E. Gelenbe, MIRAI Botnet Attack Detection with Auto-Associative Dense Random Neural Network, 2021 IEEE Global Communications Conference (GLOBECOM), 2021, pp. 01-06, doi: 10.1109/GLOBECOM46510.2021.9685306.
  • Filus K., Domańska J., Gelenbe E. (2021) Random Neural Network for Lightweight Attack Detection in the IoT. In: MASCOTS 2020. LNCS 12527. Springer, https://doi.org/10.1007/978-3-030-68110-4_5
  • E. Gelenbe and M. Nakip, IEEE Access Dec. 2022

18

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

19 of 28

​

​

Detection of Hitherto Unknown Attacks

Challenging: No Prior Supervised Learning from Known Attacks

Large Numbers of Distinct Attack Types

Detect Quickly to Avoid Dire Consequences

Novelty: Learn from « Normal Traffic » NOT from Attacks

Method

  • « Dense Random Neural Nework » (Yin-Gelenbe 2016)
  • Uses G-Network with Triggerred Customer Movement
  • Inspired by Clusters of Neuronal Soma to Soma Interactions
  • « Recurrent » Neural Clusters in a Feedforward Architecture

​

​

19

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

20 of 28

​

​

Unknown Attacks

Feedforward Architecture of Dense RNN

« Recurrent » Neuronal « Clusters »

​

​

20

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

21 of 28

​

n-neuron Cluster Eqn: Soma to Soma Interaction Probabiility p

​

​

21

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

22 of 28

For large n we have:

22

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

23 of 28

Learning is Achieved by Selecting

the Weight Matrices so that

​

||QL-Q0||

​

is Minimized

23

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

24 of 28

Testing with KDD’99 Data Set (Calibrated DoD Real Dataset)

The prediction accuracy is above 98% for 21 out of 37 attack types.

24

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

25 of 28

Overall AADRNN Evaluation with KDD’99 Dataset �41-20-41 vs 41-41-41 Architecture

25

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

26 of 28

Comparison (41-20-41) with SVM: Complete KDD’99 Dataset

26

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

27 of 28

- A G-Network Composed of Layered Feedforward Structure with Clusters, Each Composed of a Recurrent ”Triggered Customer” Network is Designed��- It has the Desired Properties with Regard to Function Approximation and Existence and Uniqueness of G-Networks in General��- After very efficient (as compared to SVM) Auto-Associative Learning with just Normal Traffic, we show that it very efficiently and accurately detects a large number of Different Attack Types��- Opens further research venues for ML with G-Networks

27

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.

28 of 28

Research supported by the EU H2020 Program under the Research & Innovation Project IoTAC, Grant Agreement No. 952684 to IITIS-PAN.

​

Merci de Votre Attention

​

Erol Gelenbe1�Institute of Theoretical & Applied Informatics (IITIS-PAN)�Polish Academy of Sciences

seg@iitis.pl

28

IoTAC - Security By Design IoT Development and Certificate Framework with Front-end Access Control

https://iotac.eu

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 952684.