Talks and Sessions

Impact


What is the impact on the development of the principal disciplines of the project?

  1. Our results have expanded the theory of secure compression by giving the limits on compression when only partial information is secured for inference applications. The results clearly show the cost of security by the extra terms introduced when compared to existing nonsecure compression theory.
  2. Our results have provided a new classification of possible attacks on IoT systems for inference while showing the impacts of these attacks and how to protect these systems. A study of the most dangerous attacks provides concerns for dangerous commercial systems, like self-driving cars, and underlines the importance of providing protection in these currently unregulated cases.
  3. We have developed theory and methods for distributed decision making and fusion approaches where underlying probability models are learned and approximated by empirical distributions. 
  4. We have developed spatial multiple hypotheses testing methods and fusion methods for performing inferences about the state of spatially varying fields and unsupervised learning through clustering. 
  5. We have developed methods for lightweight encryption and secure short packet communication of local inference results to the fusion center. 
  6. The developed methods are multi-disciplinary have very broad practical applications in monitoring radio spectrum, environment, pollution, agricultural fields, biomedical fields, smart buildings and traffic, and any massively connected IoT sensing system.

What is the impact on other disciplines?

Our results demonstrate that developers in all areas must build in security when new systems are first deployed. Such protection will require experts on all areas to participate in secure design. Thus all disciplines are impacted.


What is the impact on the development of human resources?

We are training postdocs and graduate students who can lead security-by-design teams to make sure future products are securely designed at their initial deployment. Researcher exchange will further enhance close research collaboration between USA and Finland in IT area. 


What is the impact on physical resources that form infrastructure?

The project is helping justify a new IoT living laboratory on the Lehigh Mountaintop Campus. No approval on this yet.


What is the impact on institutional resources that form infrastructure?

The project is helping justify a new center at Lehigh on IoT systems under the Cyberphysical Institue for Infrastructure and Energy (CPIE). No approval on this yet.


 

Participants

Participants associated with the project:

Name Role Affiliation
Rick S. Blum Project Director Lehigh University
Visa Koivunen Co-Investigator Aalto University
H. Vincent Poor Co-Investigator Princeton University
H. Seyedmohammad Postdoctoral scholar Lehigh University
M. Hosseinitoushmanlouei Postdoctoral scholar Lehigh University
Y. Shkel Postdoctoral scholar Princeton University
H. Henri Graduate Student Aalto University
Ananth Samudrala Graduate Student Lehigh University
Topi Halme Graduate Student Aalto University
Martin Gölz Graduate Student Aalto University
Hassan Naseri Graduate Student Aalto University

Organizations Involved as Partners:

Name Type of Organization Location
Lehigh University Academic Institution Bethlehem, PA
Princeton University Academic Institution Princeton, NJ
Aalto University Academic Institution Finland

 

Products

Journal and Conference Publications:

  1. Zhang, Jiangfan and Blum, Rick S. and Poor, H. Vincent, “Approaches to Secure Inference in the Internet of Things: Performance Bounds, Algorithms, and Effective Attacks on IoT Sensor Networks”, IEEE Signal Processing Magazine. 35 (5) pp. 50-63, 2018.

  2. Samudrala, Ananth Narayan and Blum, Rick S. “On the estimation and secrecy capabilities of stochastic encryption for parameter estimation in IoT”, Conference on Information Sciences and Systems 2018, pp. 1-6.

  3. Samudrala, Ananth Narayan and Blum, Rick S, “Asymptotic analysis of a new low complexity encryption approach for the Internet of Things, smart cities and smart grid”, 2017 IEEE International Conference on Smart Grid and Smart Cities (ICSGSC), pp. 200-204.

  4. Yanina Y. Shkel, Rick S. Blum, and H. Vincent Poor, “Lossless compression and secrecy by design”, submitted to IEEE Transactions on Information Theory, 2018.

  5. H. Naseri and V. Koivunen, “A Bayesian algorithm for distributed cooperative localization using distance and direction estimates,” in IEEE Transactions on Signal and Information Processing over Networks. (early access in IEEE Xplore).

  6. T. Halme and V. Koivunen, “DISTRIBUTED NONPARAMETRIC INFERENCE USING A ONE-SAMPLE BOOTSTRAPPED ANDERSON-DARLING TEST AND P-VALUE FUSION,” 2018 IEEE Data Science Workshop (DSW), Lausanne, 2018, pp. 1-5.

  7. T. Halme, V. Koivunen, and H. V. Poor, “Nonparametric distributed detection using bootstrapping and Fisher’s method,” 2018 52nd Annual Conference on Information Sciences and Systems (CISS), Princeton, NJ, 2018, pp. 1-6.

  8. P. B. Gohain, S. Chaudhari and V. Koivunen, “Cooperative Energy Detection With Heterogeneous Sensors Under Noise Uncertainty: SNR Wall and Use of Evidence Theory,” in IEEE Transactions on Cognitive Communications and Networking, vol. 4, no. 3, pp. 473-485, Sept. 2018.

  9. Martin Gölz, Visa Koivunen, Abdelhak Zoubir, “Nonparametric detection using empirical distributions and bootstrapping”, 25th European Signal Processing Conference (EUSIPCO), pp. 1450-1454, 2017.

  10. H. Hentila, J. Oksanen and V. Koivunen, “Distributed scheduling in multi-hop multi-band cognitive radio networks utilizing potential fields,” 2017 IEEE 18th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), Sapporo, 2017, pp. 1-6.

  11. Henri Hentilä, Visa Koivunen, Vincent Poor, Rick Blum, “Secure Key Generation for Distributed Inference in IoT”, CISS 2019, Baltimore.

  12. Topi Halme, Martin Gölz, Visa Koivunen, “Bayesian Multiple Hypothesis Testing for Distributed Detection in Sensor Networks “, submitted to 2019 IEEE Data Science Workshop.

  13. Martin Gölz, Michael Muma, Topi Halme, Abdelhak Zoubir, and Visa Koivunen, “Spatial Inference in Sensor Networks using Multiple Hypothesis Testing and Bayesian Clustering”, submitted to 2019 EUSIPCO (European Signal Processing Conference).

Accomplishments


What was accomplished under these goals? 

  1. We have been developing fundamental theory showing how much information can be securely transferred while also achieving compression. The theory is lacking in this area. There is a need to extend the theory of compression to cases where security is desired. Is it possible that significant compression can be achieved while maintaining secure communication? 
  2. We have are also studying the theory of security for IoT inference applications. We want to understand the types of possible attacks and how to protect against them.

Significant Results:

  1. We have made significant progress in developing some fundamental theory showing how much information can be securely transferred while also achieving compression. We have very important new results that build the theory in this area. The results show how to extend the theory of compression to cases where security is desired. The results show that if we secure only certain parts of the data, which is a very practical assumption, significant compression can still be achieved.
  2. We have also made significant progress on the theory of security for IoT inference applications. We have classified all possible attacks, those of greatest impact and we have shown how to protect against these attacks.
  3. We have developed theory and methods for distributed decision making and fusion approaches where underlying probability models are learned and approximated by empirical distributions. 
  4. We have developed spatial multiple hypotheses testing methods and fusion methods for performing inferences about the state of spatially varying fields and unsupervised learning through clustering. We have analyzed their performance using both analytical methods and simulations.
  5. We have developed methods for lightweight encryption and secure short packet communication of local inference results to the fusion center. Key generation is based on common randomness between each sensor and the fusion center.

What opportunities for training and professional development has the project provided?

We are training several postdocs and students who we expect will take faculty positions at US universities. We are teaching the students and postdocs how to be successful in these positions.


How have the results been disseminated to communities of interest?

We have published our results in the top journals and conferences and presented our work at the top universities in the world. The PIs (Blum and Poor) have both given plenary talks at major conferences, workshops, tutorials, and summer schools on our research. We have organized special sessions at the Conference on Information Sciences and Systems (CISS) for each of the two years the project has been running. Last year CISS was held at Princeton University. This year we have another session at CISS at John Hopkins University.


Papers presented at CISS-2019 Session on Security and Inference on IoT Networks:

  1. Y. Shkel and H. V. Poor, “Parameter Estimation and Secrecy by Design”.

  2. H. Hentila, V. Koivunen, H. V. Poor, and R. S. Blum, “Secure Key Generation for Distributed Inference in IoT”.

  3. J. Perazzone, P. Yu, B. Sadler, and R. S. Blum, “Cryptographic Side Channel Signaling and Authentication via Fingerprint Embedding: Security Analysis”.

  4. A. Samudrala, H. Amini, S. Kar, and R. S. Blum, “Optimal Sensor Placement for Topology Identification of Smart Power Grids”.

  5. Z. Wang and R. S. Blum, “Topology Attack Detection in Natural Gas Delivery Networks”.


Papers presented at CISS-2018 Session on Inference processing for IoT:

  1. A. Samudrala, R. S. Blum, H. V. Poor, and Visa Koivunen, “On the Estimation and Secrecy Capabilities of Stochastic Encryption for Parameter Estimation in IoT”, CISS-2018.

  2. A. Chattopadhyay and U. Mitra, “Dynamic Sensor Selection for Time-Varying Stochastic Process Tracking”.

  3. Z. Yang and W. U. Bahwa, “Distributed Machine Learning in the age of cyber attacks”.

  4. J. Heydari, S, Sihag and A. Tajer, “Quickest search for Transient Changepoints under Composite Post-change Models”.


Papers presented at CISS-2018 Session on Theory and Bounds for IoT Security:

  1. O.Kosut and J. Kliewer, “Finite Blocklength bounds for arbitrarily varying channel”.

  2. C. Huang, P. Kairouz, X. Chen, L. Sankar, and R. Rajagopal, “Generative Adversarial Privacy: A Data-Driven Approach for Guaranteeing Privacy and Utility”.

  3. Y. Wei, K. Banawan and S. Ulukus, “Private information with partially known private side information”.

  4. I. Issa and A. Wagner, “Learning Maximal Leakage”.

  5. Y. Shkel, R. S. Blum and H. V. Poor, “Secure Lossless Compression”.

  6. A. Zewail and A. Yener, “Cache-aided combination networks with Secrecy Guarantees”.


 

WiFiUS: Secure Inference in the Internet of Things

Principal Investigators:

  1. Prof. Rick S. Blum, Lehigh University.

  2. Prof. Vincent Poor, Princeton University.

  3. Prof. Visa Koivunen, Aalto University.

Goals of the Project:

  1. To understand how much information can be transferred in secret for the resource constrained, short packet, low latency, inference applications most relevant to IoT.

  2. Develop approaches that approach the theoretical rates.

  3. Understand the fundamental limits on one’s ability to compress data in IoT environments while still maintaining secrecy, along with compression and reconstruction algorithms to approach these bounds.

  4. Understand the possible attacks on IoT inference systems, their impact and how to protect against them.

  5. Understanding the limits of secrecy, privacy, and security for IoT inference applications employing statistical inference and machine learning in distributed settings.  How does distributed processing impact inference, security, secrecy, and authentication?   

  6. Develop distributed inference and learning methods for smart IoT sensors that fit with the developed secure communication approaches. Light encryption of short messages containing local inference results using common randomness.

  7. Low-complexity and energy efficient estimation, multiple hypothesis decision making and clustering methods for distributed smart sensors in IoT will be derived. Instead of explicitly assuming probability models, the empirical models will be learned from data to develop estimates or decision statistics and quantitative reliability measures.

  8. Novel fusion methods combining local inference results, performing multiple testing and unsupervised clustering and providing quantitative information about their reliability will be developed.

Project Details:

  1. Accomplishments

  2. Products

  3. Participants and Organizations Involved in the Project

  4. Impact of the Project

  5. Talks and Sessions


This work is supported by the National Science Foundation under grants CNS-1702555, and CNS-1702808 and by the Academy of Finland WiFiUS grant ”Secure Inference in the Internet of Things”.