For almost 40 years, Peraton Labs and its precursor organizations have made significant contributions to the advancement of military communications. At the 39th MILCOM conference on Enabling All-Domain C2 and Communications, taking place November 29 – December 2, 2021, we will be presenting six technical papers in the unclassified conference track. We organized and will moderate a panel on software defined radio (SDR) and Qinqing (Christine) Zhang, Ph.D., chief research scientist at Peraton Labs, is a technical committee chair on the conference organizing committee.
Our MILCOM 2021 panel is on heterogeneous processing for SDR and features a cross-section of speakers from government, industry, and academia. The panel focuses on the state-of-the-art in heterogenous SDR implementations, which have the native capability to utilize diverse processing elements (CPU, GPU, FPGA, TPU, etc.). Heterogeneous SDR is particularly beneficial on platforms constrained in size, weight, and power (SWAP), such as aircraft, UAVs, and handhelds.
Three of our technical papers are on adversarial machine learning (ML). ML models are vulnerable during training and use. Hackers can disrupt ML algorithms to force them to make erroneous decisions, such as misidentifying road signs or failing to detect cyber attacks. Adversarial ML identifies, assesses, and mitigates these threats to secure and defend ML solutions.
Our MILCOM 2021 panel is on heterogeneous processing for SDR and features a cross-section of speakers from government, industry, and academia. The panel focuses on the state-of-the-art in heterogenous SDR implementations, which have the native capability to utilize diverse processing elements (CPU, GPU, FPGA, TPU, etc.). Heterogeneous SDR is particularly beneficial on platforms constrained in size, weight, and power (SWAP), such as aircraft, UAVs, and handhelds.
Three of our technical papers are on adversarial machine learning (ML). ML models are vulnerable during training and use. Hackers can disrupt ML algorithms to force them to make erroneous decisions, such as misidentifying road signs or failing to detect cyber attacks. Adversarial ML identifies, assesses, and mitigates these threats to secure and defend ML solutions.
- “Poisoning Attacks and Data Sanitization Mitigations for Machine Learning Models in Network Intrusion Detection Systems” considers poisoning attacks in which training data sets are corrupted to produce mis-performing algorithms; it develops scalable methods to identify and remove the corrupt data.
- “Combinatorial Boosting of Ensembles of Diversified Classifiers for Defense Against Evasion Attacks” considers evasion attacks in which inputs are designed to produce incorrect results; it develops efficient methods to create and fuse multiple, diverse ML classifiers to block these attacks.
- “Privacy Leakage Avoidance with Switching Ensembles” considers membership inference attacks in which hackers can obtain private information about the contents of training data; it develops a novel approach to protect against these attacks without significant loss in accuracy.
- “Reservoir-Based Distributed Machine Learning for Edge Operation of Emitter Identification” uses ML to authenticate mobile users at the network edge. Our approach achieves high accuracy and high reliability with low power requirements and is specifically designed for efficient operation at the tactical edge and for Internet of Things (IoT).
- ”Tactical Jupiter: Dynamic Scheduling of Dispersed Computations in Tactical MANETs (Mobile Ad-hoc Networks)” adapts the Jupiter framework for scheduling computations over heterogeneous resources. Our approach improves network performance and enables faster decision-making by leveraging the distributed processing capability available at the tactical edge.
- “BBR (Bottleneck Bandwidth and Roundtrip propagation time)-Inspired Congestion Control for Data Fetching Over NDN (Named Data Networking)” develops congestion control solutions for networks running the NDN protocol. Our approach outperforms the original BBR method and works particularly well in dynamic and contested tactical environments.
