- SaTC: CORE: Small: Adversarial Network Reconnaissance in Software Defined Networking (NSF, $500,000, 3/15/2020-2/28/2023)
Develop reconnaissance techniques to infer internal policies and states of a software defined network from compromised/malicious hosts and switches, and demonstrate the impacts of leaking such information via the design of intelligent attacks against target components (e.g., flow table, control application).
- CPS: Medium: Coupled cAscade Modeling, Prevention, and Recovery (CAMPR): When Graph Theory meets Trajectory Sensitivity (NSF, $999,000, 9/1/2018-8/31/2021)
Develop mechanisms to model cascading failures in interdependent power-communication networks, to minimize damage during cascading by preventive control, and to maximize restored services during recovery under limited resources and uncertainty.
Related publications: ISGT-Europe’20, SmartGridComm’20
- CIF: Small: Adversarial Network Tomography: Inferring Network State from Manipulated End-to-end Measurements (NSF, $170,000, 10/1/2018-9/30/2020)
Quantify the vulnerability of existing network tomography algorithms in an adversarial setting, including (i) developing optimal attack strategies for representative tomography algorithms wrt typical performance metrics (e.g., additive, min, boolean), and (ii) analyzing their impact in terms of the maximum performance degradation and/or inference error.
- Agile Analytics Enabled by Decentralized Continuous Learning in Coalitions (ARL – DAIS ITA, $147,957, 1/15/2020-9/14/2021)
Develop algorithms with provable performance guarantees for approximate machine learning in resource-constrained edge devices, with focus on optimal design of data reduction techniques (coreset, quantization,random projection, etc) for edge-based learning.
- Network Topology Inference under Generalized Forwarding
Study the inference of network routing topology and link performance metrics from end-to-end measurements (a.k.a. network topology tomography) in new networking regimes, such as software-defined networking and network function virtualization, which allow non-tree-based generalized forwarding.
- Distributed Analytics and Information Science (DAIS) ITA, BPP18 (ARL, $1,600,000, 1/18/2018-1/17/2020)
Study challenges in enabling efficient and reliable distributed analytics in highly dynamic, resource-constrained, and adversarial network environments (e.g., tactical networks). Focus in BPP18 is resource allocation (e.g., service placement, request routing, scheduling) and “universally good coresets” that are much smaller in size than the original dataset but give provable approximation guarantees for a broad range of machine learning models.
- Network and Information Sciences (NIS) ITA, BPP13-IPP15 (ARL, $1,500,000, 2013-2016): Network Tomography in Multi-domain Networks
Developed theory and algorithms for unique identification of link/node states from states of end-to-end paths measured between monitors, including: (i) verifiable conditions and efficient network planning (monitor placement, path construction) algorithms to guarantee identifiability of additive link metrics, (ii) novel measure and efficient algorithms to quantify network capability in localizing failures from Boolean path states, (iii) efficient algorithms to design probe allocation for inferring link parameters from stochastic path performance metrics.
Related publications: Performance’17.a, Performance’17.b, TON’17.a, TON’17.c, TON’14, INFOCOM’17, ICDCS’16, INFOCOM’16, MILCOM’15.a, Performance’15.a, SIGMETRICS’15, IMC’14, ICDCS’14, INFOCOM’14, Globecom’13, IMC’13, ICDCS’13
- Fast Network Configuration in Software-Defined Networks
Study novel optimization problems arising in the selection of paths for new flows and/or adjustment of paths for existing flows that aim at: (i) balancing rule updates across switches, or (ii) minimizing disruption on existing flows.
Related publications: TNSM’18
- Distributed Machine Learning at the Edge
Study challenges in training machine learning models over a large and distributed dataset based on the architecture of mobile edge computing, with focus on computation/communication tradeoff in “federated learning”.
Related publication: JSAC’19, INFOCOM’18.b
- Optimal Service Provisioning in Edge Computing
Developed theory and algorithms for: (i) optimal content placement and retention for edge caching, (ii) joint service data/code placement and request scheduling, (iii) optimal service migration in response to user dynamics, (iv) user location privacy in the presence of cyber eavesdroppers, and (v) incentives and mechanism design.
- Online Learning in Large Dynamic Networks
Developed theory and algorithms for: (i) end-host-based learning of shortest path using coupled or decoupled probing, (ii) tracking Markovian time-varying link states using adaptive sampling, and (iii) timing-based tracking of transactions through a distributed transaction processing system.
- Workload Scheduling in Cloud Computing
Developed algorithms and performance analysis for workload scheduling in cloud computing networks with novel features, including (i) stochastic availability of computation resources, (ii) diversity of network scheduling mechanisms, and (iii) requirement of deadlines and commitments.
- Controlled Mobility in Delay-Tolerant Networks
Developed control policies for dynamically controlling mobile relays (aka data ferries) to provide delay-tolerant communications between partitioned mobile nodes based on partial observations.
- Quality of Information in Sensor Networks
Quantified the impact of imperfect communications (e.g., delays, losses) on the quality of information (QoI) for representative applications in wireless sensor networks, including (i) tracking moving targets and (ii) detecting transient signals.
- Timing-based Information Flow Detection
Developed detector and performance guarantee for detecting information flows (a sequence of relayed packet streams) among background traffic and noise based on transmission timestamps, with application to the detection of encrypted stepping-stone attacks.
- Anonymous Networking among Timing Eavesdroppers
Developed transmission scheduling strategies and fundamental limits for sending anonymous information flows that cannot be detected in both content (via encryption) and timing (via embedding) domains.
- Non-parametric Change Detection in High Dimensional Space
Developed detector and performance analysis for detecting and localizing changes in an unknown distribution in a high dimensional (>1D) space, with application to change detection in 2D random sensor fields.