Little research explores the correlation between the expressive ability and generalization ability of the low-rank adaptation (LoRA). Sharpness-Aware Minimization (SAM) improves model generalization for both Convolutional Neural Networks (CNNs) and Transformers by encouraging convergence to locally flat minima. However, the connection between sharpness and generalization has not been fully explored for LoRA due to the lack of tools to either empirically seek flat minima or develop theoretical methods. In this work, we propose Flat Minima LoRA (FMLoRA) and its efficient version i.e., EFMLoRA, to seek flat minima for LoRA. Concretely, we theoretically demonstrate that perturbations in the full parameter space can be transferred to the low-rank subspace. This approach eliminates the potential interference introduced by perturbations across multiple matrices in the low-rank subspace. Our extensive experiments on large language models and vision-language models demonstrate that EFMLoRA achieves optimization efficiency comparable to that of LoRA while simultaneously attaining comparable or even better performance. For example, on the GLUE dataset with RoBERTa-large, EFMLoRA outperforms LoRA and full fine-tuning by 1.0% and 0.5% on average, respectively. On vision-language models e.g., Qwen-VL-Chat, there are performance improvements of 1.5% and 1.0% on the SQA and VizWiz datasets, respectively. These empirical results also verify that the generalization of LoRA is closely related to sharpness, which is omitted by previous methods.

Leveraging the complementary characteristics of visible (RGB) and infrared (IR) imagery offers significant potential for improving object detection. In this paper, we propose WaveMamba, a cross-modality fusion method that efficiently integrates the unique and complementary frequency features of RGB and IR decomposed by Discrete Wavelet Transform (DWT). An improved detection head incorporating the Inverse Discrete Wavelet Transform (IDWT) is also proposed to reduce information loss and produce the final detection results. The core of our approach is the introduction of WaveMamba Fusion Block (WMFB), which facilitates comprehensive fusion across low-/high-frequency sub-bands. Within WMFB, the Low-frequency Mamba Fusion Block (LMFB), built upon the Mamba framework, first performs initial low-frequency feature fusion with channel swapping, followed by deep fusion with an advanced gated attention mechanism for enhanced integration. High-frequency features are enhanced using a strategy that applies an ``absolute maximum" fusion approach. These advancements lead to significant performance gains, with our method surpassing state-of-the-art approaches and achieving average mAP improvements of 4.5% on four benchmarks.

With the continual growth in user communication needs, classical optical communications are facing developmental bottlenecks. On one hand, the communication capacity of available optical fiber resources is approaching its upper limit. On the other hand, emerging quantum computing technology poses security threats. Quantum key distribution (QKD), as a representative quantum cryptography technology, is being introduced into existing optical infrastructure to mitigate security threats. It also has pioneering application value for next-generation quantum information networks. However, limited optical fiber resources struggle to support the introduction of quantum communication over classical optical networks. There are also incompatible noise factors between the two communication paradigms. This paper proposes transitioning from unilateral adaptive routing and wavelength assignment (ARW A) to bilateral synergistic RW A (SRW A) to facilitate the coexistence of two heterogeneous communication paradigms in optical networks. Simulations have proven SRW A can increase end-to-end key supply rates from bit/s to kbit/s levels. It has an enabling effect on QKD over classical optical networks.

Datacenter networks (DCNs) with cloud–edge collaboration are emerging to satisfy the communication, computation, and caching (3C) requirements of future services such as cloud-based IoT services. However, the enroute data over DCNs with cloud–edge collaboration is likely to suffer from cyberattacks such as eavesdropping. A large number of services require not only 3C resources, but also cryptographic resources for encryption to ensure high security. Quantum key distribution (QKD) is a practical approach to provide secret keys for remote users with information-theoretic security against attacks from quantum computing. A QKD-secured DCN (QKD-DCN) with cloud–edge collaboration can be deployed to satisfy the communication, computation, caching, and cryptographic (4C) requirements of services. This article innovatively solves the new 4C resource-allocation (4CRA) problem in the network to minimize the cryptographic resource consumption. It formulates an integer linear programming (ILP) model and proposes a heuristic cryptographic-dependent 4CRA algorithm to find optimal solutions. The proposed algorithm is compared with two baseline 4CRA algorithms which, respectively, consider the minimized service delivery latency and the first-fit resource availability. Analytical simulations show that the proposed algorithm minimizes the key-resource-consumption ratio and the average key-resource consumption under static and dynamic traffic scenarios in different network topologies.