Current multimodal template protection methods typically require encryption or transformation of the original biometric features. However, these operations carry certain risks, as attackers may reverse-engineer or decrypt the protected multimodal templates to retrieve partial or complete information about the original templates, leading to the leakage of the original biometric features. To address this issue, we propose a cancelable multimodal template protection method based on random indexing. First, hash functions are used to generate integer sequences as index values, which are then employed to create single-modal cancelable templates using random binary vectors. Second, the single-modal cancelable templates are used as indices for random binary sequences, which locate the corresponding template information and are filled into the fusion cancelable template at the respective positions, achieving template fusion. The resulting template is unrelated to the original biometric features. Finally, without directly storing the binary factor sequences, an XOR operation is performed on the extended biometric feature vectors and random binary sequences to generate the encoded key. Experimental results demonstrate that the proposed method significantly enhances performance on the FVC2002DB1 fingerprint, MMCBNU_6000 finger-vein, and NUPT_FPV databases, while also satisfying the standards for cancelable biometric feature design. We also analyze four privacy and security attacks against this scheme.

Fake identity in social networks is a phenomenon that is strongly increasing, and it is used for discovering personal information, identity theft, influencing people, spreading fake news, fraud, and so on. In this paper, we face this problem by introducing the concept of certified social profiles and by propagating this property through a collaborative approach that exploits keystroke-dynamic-recognition techniques to identify illegal access to certified profiles. We propose a decentralized approach to compute the trust level of a social profile, and we show the robustness of the proposal by analyzing the security of the trust mechanism through experimental validation.

Recently, the number of publicized attacks on IoT devices has noticeably grown. This is in part due to the increasing deployment of embedded systems into various domains, including critical infrastructure, which makes them a valuable asset and a compromise can cause significant damages. In this case, it is often required to send an engineer to manually recover the devices, as the attack leaves them out of reach of standard remote management solutions. To avoid this costly process, the concept of cyber resilience has gained traction in recent years in both academia and industry. Its core idea is to enable compromised devices to recover themselves to a trusted state without human intervention. Initial guidelines and architectures to realize cyber resilience have been published by standardization entities like NIST and TCG, and in multiple academic article. While the initial works focused on guaranteed recovery, recent proposals included attack detection to speed up the recovery process. In this work, we build on top of these ideas and present an extended resilience architecture. We present new implementations of resilience engines with a focus on secure and reliable data acquisition for attack detection and classification. Our attack classification engine enables tailored, more efficient recovery responses.

The class of Trustworthy Autonomous Systems (TAS) includes cyber-physical systems leveraging on self-x technologies that make them capable to learn, adapt to changes, and reason under uncertainties in possibly critical applications and evolving environments. In the last decade, there has been a growing interest in enabling artificial intelligence technologies, such as advanced machine learning, new threats, such as adversarial attacks, and certification challenges, due to the lack of sufficient explainability. However, in order to be trustworthy, those systems also need to be dependable, secure, and resilient according to well-established taxonomies, methodologies, and tools. Therefore, several aspects need to be addressed for TAS, ranging from proper taxonomic classification to the identification of research opportunities and challenges. Given such a context, in this paper address relevant taxonomies and research perspectives in the field of TAS. We start from basic definitions and move towards future perspectives, regulations, and emerging technologies supporting development and operation of TAS.

Today, the deep learning-based side-channel analysis represents a widely researched topic, with numerous results indicating the advantages of such an approach. Indeed, breaking protected implementations while not requiring complex feature selection made deep learning a preferred option for profiling side-channel analysis. Still, this does not mean it is trivial to mount a successful deep learning-based side-channel analysis. One of the biggest challenges is to find optimal hyperparameters for neural networks resulting in powerful side-channel attacks. This work proposes an automated way for deep learning hyperparameter tuning based on Bayesian optimization. We build a custom framework denoted AutoSCA supporting machine learning and side-channel metrics. Our experimental analysis shows that our framework performs well regardless of the dataset, leakage model, or neural network type. We find several neural network architectures outperforming state-of-the-art attacks. Finally, while not considered a powerful option, we observe that neural networks obtained via random search can perform well, indicating that the publicly available datasets are relatively easy to break.

Suggesting complementary clothing items to compose an outfit is a process of emerging interest, yet it involves a fine understanding of fashion trends and visual aesthetics. Previous works have mainly focused on recommendation by scoring visual appeal and representing garments as ordered sequences or as collections of pairwise-compatible items. This limits the full usage of relations among clothes. We attempt to bridge the gap between outfit recommendation and generation by leveraging a graph-based representation of items in a collection. The work carried out in this paper, tries to build a bridge between outfit recommendation and generation, by discovering new appealing outfits starting from a collection of pre-existing ones. We propose a transformer-based architecture, named TGNN, which exploits multi-headed self attention to capture relations between clothing items in a graph as a message passing step in Convolutional Graph Neural Networks. Specifically, starting from a seed, i.e. one or more garments, outfit generation is performed by iteratively choosing the garment that is most compatible with the previously chosen ones. Extensive experimentations are conducted with two different datasets, demonstrating the capability of the model to perform seeded outfit generation as well as obtaining state of the art results on compatibility estimation tasks.

In order to protect the sensitive information in many applications involving neural networks, several privacy-preserving neural networks that operate on encrypted data have been developed. Unfortunately, existing encryption-based privacy-preserving neural networks are mainly built on classical cryptography primitives, which are not secure from the threat of quantum computing. In this paper, we propose the first quantum-resistant solution to protect neural network inferences based on an inner-product functional encryption scheme. The selected state-of-the-art functional encryption scheme based on lattice-based cryptography works with integer-type inputs, which is not directly compatible with neural network computations that operate in the floating point domain. We propose a polynomial-based secure convolution layer to allow a neural network to resolve this problem, along with a technique that reduces memory consumption. The proposed solution, named QuripfeNet, was applied in LeNet-5 and evaluated using the MNIST dataset. In a single-threaded implementation (CPU), QuripfeNet took 107.4 seconds for an inference to classify one image, achieving accuracy of 97.85%, which is very close to the unencrypted version. Additionally, the GPU-optimized QuripfeNet took 25.9 seconds to complete the same task, which is improved by 4.15× compared to the CPU version.

The graph neural network (GNN) has recently become an emerging research topic for processing non-euclidean data structures since the data used in various popular application domains are usually modeled as a graph, such as social networks, recommendation systems, and computer vision. Previous GNN accelerators commonly utilize the hybrid architecture to resolve the issue of “hybrid computing pattern” in GNN training. Nevertheless, the hybrid architecture suffers from poor utilization of hardware resources mainly due to the dynamic workloads between different phases in GNN. To address these issues, existing GNN accelerators adopt a unified structure with numerous processing elements and high bandwidth memory. However, the large amount of data movement between the processor and memory could heavily downgrade the performance of such accelerators in real-world graphs. As a result, the processing-in-memory architecture, such as the ReRAM-based crossbar, becomes a promising solution to reduce the memory overhead of GNN training. In this work, we present the TCAM-GNN, a novel TCAM-based data processing strategy, to enable high-throughput and energy-efficient GNN training over ReRAM-based crossbar architecture. Several hardware co-designed data structures and placement methods are proposed to fully exploit the parallelism in GNN during training. In addition, we propose a dynamic fixed-point formatting approach to resolve the precision issue. An adaptive data reusing policy is also proposed to enhance the data locality of graph features by the bootstrapping batch sampling approach. Overall, TCAM-GNN could enhance computing performance by 4.25× and energy efficiency by 9.11× on average compared to the neural network accelerators.

In the early phases of the COVID-19 pandemic, repurposing of drugs approved for use in other diseases helped counteract the aggressiveness of the virus. Therefore, the availability of effective and flexible methodologies to speed up and prioritize the repurposing process is fundamental to tackle present and future challenges to worldwide health. This work addresses the problem of drug repurposing through the lens of deep learning for graphs, by designing an architecture that exploits both structural and biological information to propose a reduced set of drugs that may be effective against an unknown disease. Our main contribution is a method to repurpose a drug against multiple proteins, rather than the most common single-drug/single-protein setting. The method leverages graph embeddings to encode the relevant proteins' and drugs' information based on gene ontology data and structural similarities. Finally, we publicly release a comprehensive and unified data repository for graph-based analysis to foster further studies on COVID-19 and drug repurposing. We empirically validate the proposed approach in a general drug repurposing setting, showing that it generalizes better than single protein repurposing schemes. We conclude the manuscript with an exemplified application of our method to the COVID-19 use case. All source code is publicly available.