Mid Sweden University

Semantic segmentation on resource-constrained hardware remains a key challenge in deep learning, particularly for deployment on edge devices and embedded systems. In this study, we propose FireSegUNet, a lightweight and computationally efficient deep-learning architecture tailored for such environments. The model integrates an optimized inverted bottleneck layer for feature extraction within an encoder–decoder framework, reducing computational complexity by up to 51%. It also improves segmentation accuracy through an efficient squeeze-and-excitation block, while reducing inference time by up to 7.3× and energy consumption by up to 4.6× compared to conventional attention mechanisms. Extensive evaluation on diverse fire segmentation datasets demonstrates that FireSegUNet achieves competitive segmentation accuracy while reducing the number of parameters and storage requirements by up to 81%. Additionally, we provide a detailed analysis of the relationship between model complexity metrics and actual inference time, memory usage, and energy consumption. This comprehensive evaluation confirms that FireSegUNet delivers better performance on edge devices and generalizes well to unseen datasets. These findings position FireSegUNet as a practical solution for efficient image segmentation in resource-constrained environments. Although primarily validated on fire segmentation, the modular design of FireSegUNet makes it adaptable to other computer vision tasks. The source code of FireSegUNet will be publicly available at https://github.com/Realistic3D-MIUN/FireSegUNet.

In today’s complex IT environments, understanding interconnected systems poses significant challenges. The nature of these systems introduces risks, making automatic observability and problem prevention critical. Machine learning and artificial intelligence enable promising solutions for managing this complexity. This paper addresses the challenge of detecting log anomalies within unstructured, continually changing logs that are difficult for traditional methods to parse. We propose a novel approach called LogDEC that utilizes joint clustering and dimensionality reduction together with sequential anomaly detection. Our method employs a sequential auto-encoder to detect anomalies through the reconstruction error of log feature sequences. Anomalies are scored by a threshold on the reconstruction error. LogDEC is trained unsupervised and can effectively identify anomalies, as demonstrated on a dataset collected at the Swedish Social Insurance Agency, as well as with three public log datasets. The model successfully and efficiently detected anomalies arising from manual interference, highlighting its applicability in real-world scenarios. 

Deep learning is widely used in various fields of computervision applications. However, the majority of these state-of-the art deep learning architectures are computationally expensive and hand-engineered, requiring substantial expertise to discover. Recently, neural architecture search has gained significant attention as an automated tool for constructing deep neural networks. Although it has found optimal architecture for various applications, their impact on light field disparity estimation is just to be investigated. This paper introduces the Regressive Differentiable Neural Architecture Search algorithm, which finds the optimal architecture by preserving search and evaluation architecture dimensions and proposes an adaptive dynamic drop strategy based on candidate operation stability for optimization. Furthermore, the parameter-sharing technique facilitates search super-network with rapid convergence to assist in better systematic decisionmaking, enhancing the overall efficiency of the algorithm. The evaluation is conducted on two diverse computer vision tasks  the generalizability of the proposed search strategy. The proposed search strategy discovers an optimal architecture 2.46 times faster, which achieves performance comparable to recent state-of-the-art deep learning architectures. By reducing the time and effort required to find sub-optimal architecture, this study opens up new opportunities for the research community. It could make advanced computer vision more accessible in complex applications, including light field technology. 

Vehicular Ad Hoc Networks (VANETs) are the backbone of intelligent transport and enhanced passenger safety, but they face significant challenges related to authentication, security, and privacy. Existing distributed VANET authentication protocols struggle with issues like privacy preservation during vehicle handovers and inefficiency when handling large volumes of verifications. This paper proposes a novel authentication framework designed to address these limitations. First, we introduce zero-knowledge guarantees for Vehicle-to-Infrastructure (V2I) authentication and improve anonymity and unlinkability in authentication by eliminating explicit vehicle handover, thus enhancing privacy. Second, we propose a batch-verifiable Vehicle-to-Vehicle (V2V) message-sharing method utilizing an elliptic curve digital signatures scheme (ECDSA*). Unlike others, we provide a complete computational and efficiency analysis of batch verification in the presence of faulty signatures. A formal security analysis and proven security in the Scyther security verification tool provide the security guarantees of our proposed scheme. A thorough efficiency analysis shows that our scheme can perform at least 5-times more V2I authentication and can batch verify at least 2-times more V2V messages than other related schemes within a time threshold of 300ms.

Differentiable Architecture Search (DARTS) has shown promising results in automating the design of deep learning models. However, its search process is computationally expensive because it evaluates all candidate operations simultaneously, often leading to an over-parameterized and inefficient search network. To reduce the computational cost, DARTS employs a smaller search network than the final evaluation network, which introduces an architecture optimization gap that limits real-world performance. To overcome this limitation, we introduce CR-DARTS, a multi-stage search framework designed to bridge the architecture optimization gap through an adaptive channel redistribution strategy. CR-DARTS reduces the computational complexity of the search network by compressing the shared input features among candidate operations and restoring the network dimensions via channel-wise feature concatenation. In addition, it progressively eliminates underperforming operations and redistributes the number of channels for more relevant feature extraction, thereby narrowing the gap between the search and evaluation networks. We validated CR-DARTS on two diverse computer vision tasks to assess its generalizability. Experimental results show that the proposed search framework reduces the memory requirement of the DARTS algorithm by up to 4.3×, while addressing the architecture optimization gap. Moreover, in the evaluation phase, the discovered architecture achieves up to 25.3% reductions in computational complexity and 50.6% faster inference time compared to state-of-the-art methods, while maintaining comparable accuracy. It also produces a competitive fire segmentation network that outperforms the state-of-the-art methods while maintaining similar computational efficiency. These results demonstrate that CR-DARTS is a practical solution for neural architecture search. Source code will be made publicly available at https://github.com/Realistic3D-MIUN/CR-DARTS.

Convolutional neural networks are widely used forlight field disparity estimation. However, many state-of-the-artdeep learning models are computationally expensive due to their reliance on standard convolutions with varying kernel sizes. In this paper, we analyze the effect of various advanced convolution operations with different kernel sizes for feature extraction in a state-of-the-art light field disparity estimation network. Based on this investigation, we propose an optimized mixed convolution layer to extract relevant features using multiple kernel sizes in parallel, while maintaining significantly lower computational cost. Experimental results demonstrate that our approach reduces model complexity by up to 4.2× while also improving disparity estimation accuracy. These findings make the proposed convolutional operation more practical for lightfield applications, where efficient spatial and angular feature extraction is essential for improved model performance.

Vehicular Ad-Hoc Networks (VANETs) enable vehicles to share critical data for safety and traffic management. To improve efficiency, batch verification is used to authenticate multiple vehicle-to-vehicle (V2V) messages at once. However, proposed solutions avoid error-prone environments with faulty signatures because of the higher analytical complexity, thus considering only error-free scenarios. This paper considers the errorprone scenario and proposes a novel strategy that allows optimal aggregation computations and reuse of such pre-computations to minimize the computational cost of identifying the faulty signature in a batch. Our analysis shows that batch verification outperforms standard methods when the error rate is below 40 %, with advantages up to 63 % in typical scenarios. We provide guidelines for when batch verification is more efficient and suggest improvements to further optimize its performance in VANETs, offering a practical solution for real-world applications.

Log anomaly detection is a critical task for ensuring the reliability and security of modern software systems. In this work, we investigate lightweight, optimization-based approaches for log anomaly detection, focusing on an implementation based on Robust Principal Component Analysis (Robust PCA). In practice, guaranteeing that the training data for log anomaly detection models is entirely free of anomalies is often unrealistic. Therefore, robustness to contaminated training data is essential. We compare our Robust PCA approach with a standard Principal Component Analysis (PCA) baseline and show that Robust PCA more effectively handles minor abnormalities within the training data. Furthermore, we compare our approach trained on data without anomalies with state-of-the-art deep learning methods and achieve comparable results on a simple dataset. Our findings indicate that the proposed optimization-based model is simple to implement and deploy. By design, it relies on easily parsed log templates and disregards event order, enabling high efficiency while trading of the ability to capture sequential or context-dependent patterns present in more complex log data.

The advancement of Internet of Vehicles (IoV) technologies has significantly enhanced road safety and transportation efficiency through smart traffic management and precise control systems. With the advent of 5G and beyond, vehicles within the IoV ecosystem can seamlessly communicate with various smart entities (X) using V2X (Vehicle-to-Entity) communications. However, the openness of IoV networks and the exponential growth of V2X links have expanded potential attack surfaces, increasing the risk of security and privacy breaches. In response to these challenges, this article proposes a privacy-preserving and secure communication framework for IoV networks, addressing critical security challenges in V2X communication. By leveraging lightweight cryptographic mechanisms such as hash functions, quadratic residuosity, and Legendre symbols, the proposed scheme ensures secure authentication, group key sharing, and pseudonym management within IoV networks. The proposed scheme's security and privacy features, along with its correctness, have been rigorously validated against various security threats where other state-of-the-art schemes fail. Comprehensive performance analysis demonstrates that our scheme completes authentication in a fraction of a millisecond, significantly outperforming existing approaches. The design simplicity and efficiency of the proposed authentication structure make it highly suitable for real-world IoV applications. 

Vehicular Ad-Hoc Networks (VANETs) offer enhanced road safety, efficient traffic management, and improved vehicle connectivity while dealing with privacy and security challenges in public communication. In these networks, authentication mechanisms are mandatory to establish trust among communicating entities, such as vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I), without losing identity and location-based privacy. The prevailing conventional authentication mechanisms frequently depend on a centralized trust authority (CA) to ensure the mutual verifiability of transmitted messages. Nevertheless, in scenarios where the density of vehicles within the network is notably high, an overwhelming influx of authentication requests may result in a communication bottleneck at the CA, leading to a single point of failure. This paper proposes a novel distributed authentication scheme in a decentralized VANET with multiple independent CAs connected to multiple local inspectors to eliminate a single point of failure. Furthermore, prior solutions lack the capability to immediately revoke a disputed vehicle that is transmitting malicious messages in the network. In this regard, the proposed scheme also facilitates an immediate revocation of a disputed sender to prevent other vehicles from further receiving malicious messages. As vehicles share time-sensitive data for driving assistance, our scheme minimizes the computation and communication costs for V2I key sharing and direct V2V authenticated message sharing significantly compared to previously proposed schemes. Using comparatively lightweight elliptic curve cryptography and eliminating the direct involvement of CAs in the authentication process, we have reduced the overall delays and achieved a maximum of ≈ 3.9 times faster V2I authenticated key sharing, and a maximum of ≈ 7.5 times faster V2V message sharing compared to state-of-the-art bilinear pairing-based protocols. A comprehensive efficiency analysis validates our scheme's ability to outperform time-sensitive responses, such as sending and receiving an alert within nearly 4 milliseconds.