Defence Science and Technology Group

Certified defenses promise provable robustness guarantees. We study the malicious exploitation of probabilistic certification frameworks to better understand the limits of guarantee provisions. Now, the objective is to not only mislead a classifier, but also to manipulate the certification process to generate a robustness guarantee for an adversarial input—certificate spoofing. A recent study in ICLR demonstrated that crafting large perturbations can shift inputs far into regions capable of generating a certificate for an incorrect class. Our study investigates if perturbations are needed to cause a misclassification and yet coax a certified model into issuing a deceptive, large robustness radius for a target class can still be made small and imperceptible. We explore the idea of region-focused adversarial examples to craft imperceptible perturbations, spoof certificates and achieve certification radii larger than the source class—ghost certificates. Extensive evaluations with the ImageNet demonstrate the ability to effectively bypass state-of-the-art certified defenses such as Densepure. Our work underscores the need to better understand the limits of robustness certification methods.

Backscatter sounder systems co-located with Over-the-Horizon Radar (OTHR) can be leveraged to provide a near real-time characterisation of the ionospheric propagation conditions to aid the OTHR surveillance mission. Ionogram inversion can be performed using features derived from a backscatter ionogram to infer the optimum OTHR operational parameters and to enable the geolocation of the detected targets. In this paper, we demonstrate the application of a deep learning segmentation model on a backscatter ionogram dataset obtained from the Jindalee Operational Radar Network (JORN) frequency management system to extract the F2 leading edge feature. Our results show that deep learning is a viable method for leading edge feature identification and extraction in backscatter ionogram imagery.

The ongoing exponential growth of computational power, and the growth of the commercial High Performance Computing (HPC) industry, has led to a point where ten commercial systems currently exceed the performance of the highest-used HPC system in radio astronomy in Australia, and one of these exceeds the expected requirements of the Square Kilometre Array (SKA) Science Data Processors. In order to explore implications of this emerging change in the HPC landscape for radio astronomy, we report results from a survey conducted via semi-structured interviews with 14 Australian scientists and providers with experience of commercial HPC in astronomy and similar data intensive fields. We supplement these data with learnings from two earlier studies in which we investigated the application of commercial HPC to radio astronomy data processing, using cases with very different data and processing considerations. We use the established qualitative research approach of thematic analysis to extract key messages from our interviews. We find that commercial HPC can provide major advantages in accessibility and availability, and may contribute to increasing researchers’ career productivity. Significant barriers exist, however, including the need for access to increased expertise in systems programming and parallelisation, and a need for recognition in research funding. We comment on potential solutions to these issues.

An optical beam propagating through the atmosphere is modified by spatially-dependent variations in refractive index (turbulence) within the atmosphere. This turbulence impacts the beam propagation in two ways. Firstly, there is spatially dependent irradiance variations (scintillation), with random peaks and nulls distributed across the propagating beam profile. Secondly, the beam can wander away from the optical axis between the source and receiver for extended periods of time. This paper presents extensions to the general split-step angular-spectrum model to allow extended propagation distances. This model is then applied to explore options to mitigate the impact of the atmosphere in the Earth-to-LEO physical channel.

A Bayesian approach for the inverse design of acoustic coatings for broadband sound absorption is presented. The acoustic coating comprises a soft viscoelastic material submerged in water and attached to a steel backing plate. The coating is embedded with layers of resonant inclusions corresponding to cavities and hard scatterers yielding monopole and dipole sound scattering, respectively. The acoustic performance of the coating is primarily governed by strong wave scattering at frequencies around the resonance frequency of the inclusions, and enhanced wave scattering due to resonance coupling between adjacent inclusions. Designing an acoustic coating for a targeted performance presents a challenge due to the competing and often conflicting influence of several coating parameters corresponding to the material properties of the coating, the geometric and material properties of the inclusions, and the number of inclusion layers. This challenge is addressed using a dual-level Bayesian inference approach for targeted performance based on sound absorption by the coating. Parameter estimation is employed to determine the geometric dimensions of the inclusions within each layer, while model selection identifies the number of inclusion layers and material of the inclusions within each layer. The dual-level Bayesian approach is validated against coating designs of increasing complexity. By exploiting the resonances associated with different designs, a physics informed design of a coating with high broadband sound absorption is proposed. • A dual-level Bayesian approach to design high performance resonant coatings is employed. • Coating sound absorption arises from strong wave scattering at inclusion resonances. • Multiple layers and inclusion material are inferred using Bayesian model selection. • Inclusion geometric values are determined using Bayesian parameter estimation. • High broadband absorption is achieved by exploiting coating inclusion resonances.

Abstract As cyber threats become increasingly automated and sophisticated, novel solutions must be introduced to improve defense of enterprise networks. Deep reinforcement learning (DRL) has demonstrated potential in mitigating these advanced threats. Single DRL agents have proven utility toward execution of autonomous cyber defense. Despite the success of employing single DRL agents, this approach presents significant limitations, especially regarding scalability within large enterprise networks. An attractive alternative to the single‐agent approach is the use of multi‐agent reinforcement learning (MARL). However, developing MARL agents is costly with few options for examining MARL cyber defense techniques against adversarial agents. This paper presents a MARL network security environment, the fourth iteration of the cyber autonomy gym for experimentation (CAGE) challenges. This challenge was specifically designed to test the efficacy of MARL algorithms in an enterprise network. Our work aims to evaluate the potential of MARL as a robust and scalable solution for autonomous network defense.