Continual Visual Instruction Tuning (CVIT) enables Multimodal Large Language Models (MLLMs) to incrementally learn new tasks over time. However, this process is challenged by catastrophic forgetting, where performance on previously learned tasks deteriorates as the model adapts to new ones. A common approach to mitigate forgetting is architecture expansion, which introduces task-specific modules to prevent interference. Yet, existing methods often expand entire layers for each task, leading to significant parameter overhead and poor scalability. To overcome these issues, we introduce LoRA in LoRA (LiLoRA), a highly efficient architecture expansion method tailored for CVIT in MLLMs. LiLoRA shares the LoRA matrix A across tasks to reduce redundancy, applies an additional low-rank decomposition to matrix B to minimize task-specific parameters, and incorporates a cosine-regularized stability loss to preserve consistency in shared representations over time. Extensive experiments on a diverse CVIT benchmark show that LiLoRA consistently achieves superior performance in sequential task learning while significantly improving parameter efficiency compared to existing approaches.

Modern network systems have grown considerably in scale and complexity. Deep learning technology has thus become indispensable for unlocking the full potential. Therefore, the demand for high-precision fine-grained data has grown significantly. In this paper, we introduce the critical yet unexplored problem of super-resolution for network traffic data, aiming to reconstruct fine-grained data (i.e., data sampled at high frequencies) from coarse-grained data (i.e., data sampled at low frequencies). Inspired by image super-resolution techniques, we first transform network traffic data into images to expose their inherent periodic patterns. Moreover, we successfully migrate vision backbone network to the temporal super-resolution task. Based on this foundational network, we design a novel cross-attention super-resolution generative adversarial network (SRGAN) model that integrates our cross-attention mechanism to jointly capture local-global temporal correlations. Experimental results on real-world network traffic datasets demonstrate that our model effectively performs super-resolution on traffic network data, outperforming several state-of-the-art models.

In the mobile network, digital twin (DT)-aided artificial intelligence (AI)-empowered link control is vital to enhance the performance of wireless communication. This paper proposes a deep reinforcement learning (DRL)-convex optimization enhanced time-frequency domain power allocation scheme to reduce the long-term average bit error rate (BER) in multi-user orthogonal frequency division multiplexing (OFDM) systems. To alleviate performance loss caused by trial-and-error during the training period of DRL algorithms, we design a novel practical DT-aided “prediction-then-decision” autonomous wireless link control framework considering the periodic interaction mechanism between the DT and its physical counterpart. A Transformer-based channel generator Mucomformer is implemented in the DT layer to generate large amounts of multi-user virtual channel state information (CSI) in future transmission frames. In addition, the DRL agent is trained over the DT channel in advance and executed in the real-world OFDM system to generate the optimal transmission strategy by considering the interaction mechanism between the DT and the physical counterpart. The simulation results demonstrate that the proposed Mucomformer has lower average prediction error of 2.51 dB compared to the Transformer baseline. The DRL and convex-based power allocation scheme further outperforms the classic strategy. Moreover, the practical DT-aided autonomous link control framework effectively mitigates the performance impairment, achieves an average BER performance gain 45.65% higher than that without DT and achieves faster convergence during the whole training period.