Drug combinations are essential in cancer therapy, leveraging synergistic drug-drug interactions (DDI) to enhance efficacy and combat resistance. However, the vast combinatorial space makes experimental screening impractical, and existing computational models struggle to capture the complex, bidirectional nature of DDIs, often relying on independent drug encoding or simplistic fusion strategies. To address this, we propose DPASyn, a novel drug synergy prediction framework featuring a dual-attention mechanism and Precision-Aware Quantization (PAQ). The dual-attention architecture jointly models intra-drug structures and inter-drug interactions via shared projections and cross-drug attention, enabling biologically plausible synergy modeling. Our PAQ strategy dynamically optimizes numerical precision during training based on feature sensitivity—reducing memory usage by 40% and accelerating training threefold without sacrificing accuracy. With LayerNorm-stabilized residual connections for stability, DPASyn outperforms seven state-of-the-art methods on the O'Neil dataset (13,243 combinations) [1] and supports full-batch processing of up to 256 graphs on a single GPU—setting a new standard for efficient and expressive drug synergy prediction. The data and source code are available at https://github.com/Echo-Nie/DPASyn.

Estimating nutrition from food images remains a challenging task, particularly for complex, multi-component dishes. While computer vision methods are effective at recognizing food elements, they typically treat entire meals as monolithic inputs, lacking the ability to decompose visual scenes into individual components. Large language models (LLMs), in contrast, offer strong identification and qualitative reasoning capabilities but struggle with quantitative estimation, especially for assessing volume and mass of individual elements. In this work, we propose FoodAgent, a multi-modal Mixture-of-Experts (MoE) reasoning framework that improves nutrition estimation through a divide-and-conquer strategy. By decomposing dishes into distinct food components, FoodAgent dynamically routes each element to one of three specialized expert modules: (1) monocular volume estimation for nutritionally important and visually clear elements, (2) Retrieval-Augmented Generation (RAG) for important but not clear elements, and (3) direct LLM inference for minor components. This conditional expert selection aligns estimation strategies with the visual and semantic characteristics of each food element, significantly reducing cumulative errors. Experiments show that our element-wise, MoE-driven approach outperforms holistic methods, especially in real-world dietary scenarios involving diverse and complex meals.

Due to the complexity of ocean sensing tasks, buoy detection in traditional ocean observation methods has the disadvantages of high cost and insufficient real-time performance. Ocean mobile crowd sensing technology collects high-resolution data in real-time through shipborne sensors and has significant potential for future development in the ocean Internet of Things. In this article, we consider a cooperative vessel activity that performs sensing tasks, incentivizing multiple vessels through dynamic external task profits. To spontaneously incentivize the team to take part in tasks and maximize the total profit of the vessels, we propose a multiagent task allocation model (MATM-SN) based on deep reinforcement learning (DRL). This model uses a generative adversarial network to predict the trajectory of the vessels, and STGCN is used to extract synergistic features between vessels effectively. DRL training provides an optimal task allocation strategy for vessels. Finally, we compared four baselines and demonstrated that the MATM-SN model outperformed other baselines regarding task completion rate and total efficiency.

The current method of setting up bus lanes is based on historical or current road traffic conditions for control, but these methods lack a mechanism for predicting traffic conditions. This paper constructs an LSTM multi-feature state prediction model to acquire observed states in the environment, utilizes neural networks to predict future traffic conditions, and considers variables affecting actual traffic conditions. Both are used as joint states to train each agent, helping them better understand the environment and take optimal actions. Additionally, to effectively address the differences in characteristics between social vehicles and buses, a multi-type intelligent agent for signal lights and bus lanes is constructed for training. The DPQ (Deep Prediction Q-Network) deep reinforcement learning algorithm is proposed to optimize signal lights and bus lane settings. Furthermore, using mean field theory, the MFPQ (Mean Field Prediction Q-Network) algorithm models the problem as interactions between a single intelligent agent and neighboring intelligent agents’ average effects, reducing the complexity of interaction between agents. Experimental results show that both the DPQ algorithm and the MFPQ algorithm improve evaluation metrics such as travel time. The LSTM multi-feature state prediction model promotes optimal decision-making by predicting future states for intelligent agents and significantly accelerates the convergence speed of the algorithm, enhancing learning performance.

Traditionally, synthetic aperture radar (SAR)-based ship target detection is performed in the image domain, where SAR imaging processing has to be applied first. However, SAR imaging processing is complex and time-consuming, especially in the wide-swath working mode. Actually, for open sea scenes, most echoes are sea surface signals with no ship targets, and there is no need for imaging processing in those areas. Therefore, non-image domain ship target detection is studied in this letter, and a novel Incept-text convolutional neural network (TextCNN) model is proposed for ship target detection in the SAR range-compressed domain (RCD). In the proposed method, the SAR echo data are converted into a 1-D range profile signal first by range compression and mean pooling, and then, the Incept-TextCNN model is proposed and applied, and information about existence of ship targets in relevant range cells will be its output. Finally, the effectiveness and efficiency of the proposed method is testified by simulation and real spaceborne SAR data, and the results demonstrate that the proposed model can filter out the invalid range-compressed data of the sea surface area, which can significantly reduce the amount of data for subsequent SAR imaging and ship classification.

The identification of influential nodes in complex networks is an open issue. To address it, many centrality measures have been proposed, among which the most representative iteration algorithm is the PageRank algorithm. However, it ignores the correlation between nodes and assumes that the jumping probability from a node to its adjacent nodes is the same. To make up it, we proposed a method to improve the PageRank based on the structural similarity of nodes calculated by Kullback–Leibler divergence. The Susceptible-infected (SI) model was used in six real networks, and the results of comparison experiments demonstrate the effectiveness of the proposed method.

The likelihood function is widely used in data processing but the classical likelihood function is too strict to deal with data with extreme values in real applications. To address this issue, a soft likelihood function based on power ordered weighted average (POWA) operator for probability distribution is proposed. Compared with the existing soft likelihood functions, one of the main features of the proposed likelihood function is that the weighted mean is displayed by the geometric mean of the weighted probability products. As a result, the effect on final result caused by extreme data can be efficiently decreased. Numerical examples are used to illustrate the efficiency of the proposed soft likelihood function.