Recent advances in parameter-efficient transfer learning have demonstrated the utility of composing LoRA adapters from libraries of pretrained modules. However, most existing approaches rely on simple retrieval heuristics or uniform averaging, which overlook the latent structure of task relationships in representation space. We propose a new framework for adapter reuse that moves beyond retrieval, formulating adapter composition as a geometry-aware sparse reconstruction problem. Specifically, we represent each task by a latent prototype vector derived from the base model’s encoder and aim to approximate the target task prototype as a sparse linear combination of retrieved reference prototypes, under an L1-regularized optimization objective. The resulting combination weights are then used to blend the corresponding LoRA adapters, yielding a composite adapter tailored to the target task. This formulation not only preserves the local geometric structure of the task representation manifold, but also promotes interpretability and efficient reuse by selecting a minimal set of relevant adapters. We demonstrate the effectiveness of our approach across multiple domains—including medical image segmentation, medical report generation and image synthesis. Our results highlight the benefit of coupling retrieval with latent geometry-aware optimization for improved zero-shot generalization.

Accurate detection of offensive content on social media demands high-quality labeled data; however, such data is often scarce due to the low prevalence of offensive instances and the high cost of manual annotation. To address this low-resource challenge, we propose a self-training framework that leverages abundant unlabeled data through collaborative pseudo-labeling. Starting with a lightweight classifier trained on limited labeled data, our method iteratively assigns pseudo-labels to unlabeled instances with the support of Multi-Agent Vision-Language Models (MA-VLMs). Unlabeled data on which the classifier and MA-VLMs agree are designated as the Agreed-Unknown set, while conflicting samples form the Disagreed-Unknown set. To enhance label reliability, MA-VLMs simulate dual perspectives, moderator and user, capturing both regulatory and subjective viewpoints. The classifier is optimized using a novel Positive-Negative-Unlabeled (PNU) loss, which jointly exploits labeled, Agreed-Unknown, and Disagreed-Unknown data while mitigating pseudo-label noise. Experiments on benchmark datasets demonstrate that our framework substantially outperforms baselines under limited supervision and approaches the performance of large-scale models.

Hateful videos have become prevalent on online platforms, highlighting an urgent need for effective detection. However, existing studies primarily focus on binary classification and fail to provide contextual rationales that reveal the implicit meanings behind these judgments, significantly undermining model explainability. To fill this gap, we aim to achieve explainable hateful video detection, enabling models to provide contextual rationales that integrate relevant evidence and logical reasoning alongside decisions. This approach can comprehensively enhance the understanding of video content and the explainability of the decision-making process. We first introduce two datasets, Ex-HateMM and Ex-ImpliHateVid, for explainable hateful video detection. Each dataset provides fine-grained annotations of multimodal harmful elements, along with contextual rationales. We then propose an Information Augmentation and Reasoning Enhancement (IARE) framework designed for explainable detection. The framework employs an information augmentation phase that leverages the multimodal chain-of-thought to integrate harmful elements, thereby enriching rationale evidence. Additionally, IARE incorporates a reasoning enhancement phase, in which Direct Preference Optimization guides the model toward correct reasoning paths and away from incorrect ones, thereby improving the logical coherence of its justifications. We conduct extensive experiments on the two datasets, comparing multiple baselines with our proposed IARE framework. The results demonstrate that IARE achieves state-of-the-art performance while also generating accurate rationales.

The integration of Large Language Models (LLMs) with retrieval systems has shown promising potential in retrieving documents (docs) or advertisements (ads) for a given query.Existing LLM-based retrieval methods generate numeric or content-based DocIDs to retrieve docs or ads.However, the one-to-few mapping between numeric IDs and docs, along with the time-consuming content extraction, leads to semantic inefficiency and limits the scalability of existing methods on large-scale corpora.In this paper, we propose the Realtime Ad REtrieval (RARE) framework, which leverages LLM-generated text called Commercial Intentions (CIs) as an intermediate semantic representation to directly retrieve ads for queries in real-time.These CIs are generated by a customized LLM injected with commercial knowledge, enhancing its domain relevance.Each CI corresponds to multiple ads, yielding a lightweight and scalable set of CIs.RARE has been implemented in a real-world online system, handling daily search volumes in billions.The online implementation has yielded significant benefits: a 5.04% increase in consumption, a 6.37% increase in gross merchandise volume (GMV), a 1.28% enhancement in click-through rate (CTR) and a 5.29% increase in shallow conversions.Extensive offline experiments show RARE's superiority over ten competitive baselines in four major categories.

Accurately predicting aortic valve movement under coronary influence is critical for personalized cardiac interventions and virtual surgery planning. Conventional Fluid-Structure Interaction (FSI) models tend to neglect modeling coronary arteries due to their complexities, leading to bias in leaflet motion simulations. To overcome this limitation, we propose a combination of spatiotemporal graph convolution and Transformer based gated (Gated-STGFormer) deep learning framework, which learns to reconstruct coronary-modulated leaflet motion from simulations without explicit coronary arteries. The framework integrates Graph Convolutional Networks (GCNs) for spatial dependency modeling and Transformer for temporal dependency, with both encoding and gating mechanisms to effectively capture spatiotemporal couplings. Quantitative evaluation demonstrated that it reproduces the spatiotemporal movements of leaflets under the coronary arteries with a high degree of fidelity. By addressing anatomical simplifications in conventional simulations, this method provides a physics-informed and computationally efficient surrogate model with strong clinical applicability. Our findings suggest that the Gated-STGFormer effectively incorporates spatiotemporal modal information, and can serve as a module for coronary artery compensation in a preoperative planning system, enabling more realistic and personalized valve biomechanical simulations.

Advertising (ad for short) keyword suggestion is important for sponsored search to improve online advertising and increase search revenue. There are two common challenges in this task. First, the keyword bidding problem: hot ad keywords are very expensive for most of the advertisers because more advertisers are bidding on more popular keywords, while unpopular keywords are difficult to discover. As a result, most ads have few chances to be presented to the users. Second, the inefficient ad impression issue: a large proportion of search queries, which are unpopular yet relevant to many ad keywords, have no ads presented on their search result pages. Existing retrieval-based or matching-based methods either deteriorate the bidding competition or are unable to suggest novel keywords to cover more queries, which leads to inefficient ad impressions.

Summary Multitiered storage systems, which are made up of heterogeneous devices, are widely used in distributed environments to accelerate the I/O performance of upper big data applications. It raises new challenges in efficient data migration through smart caching mechanisms among heterogeneous storage levels, such as MEM‐SSD‐HDD. To optimize the cache policy scheduling mechanism on the distributed tiered storage architecture, we proposed a general framework with five layers, including a tiered storage system layer, a cache migration policy layer, a cache policy adaptive scheduling layer, a data access pattern layer, and a big data application layer. The framework prototype has been designed and implemented on the widely used distributed hybrid storage system named Alluxio. To meet the demands of the big data application layer, on the one hand, we designed a couple of cache eviction policies and promotion policies covering various access patterns on the cache migration policy layer (several proposed eviction policies have been adopted by the Alluxio open‐source community). On the other hand, two adaptive cache policy scheduling algorithms for selecting appropriate policies in various scenarios are designed and implemented on the cache policy adaptive scheduling layer. The scheduling algorithms are designed based on the hit ratio statistics and data access pattern model prediction, respectively. Experimental results show that the proposed cache policies are very effective for various big data applications, such as Spark SQL. The proposed cache policy scheduling algorithms with various eviction policies can improve around 20% hit ratio than that with a single eviction policy.

This paper studies the problem of learning node embeddings (a.k.a. distributed representations) for dynamic networks. The embedding methods allocate each node in network with a d-dimensions vector, which can generalize across various tasks, such as item recommendation, node labeling, and link prediction. In practice, many real-world networks are evolving with nodes/links added or deleted. However, most of the proposed methods are focusing on static networks. Although some previous researches have shown some promising results to handle the dynamic scenario, they just considered the added links and ignored the deleted ones. In this work, we designed a joint learning of added and deleted links model, named RDEM, for dynamic network embedding.

Search task identification aims to understand a user's information needs to improve search quality for applications such as query suggestion, personalized search, and advertisement retrieval. To properly identify the search task within long query sessions, it is important to partition these sessions into segments before further processing. In this paper, we present the first search session segmentation model that uses a long short-term memory (LSTM) network with an attention mechanism. This model considers sequential context knowledge, including temporal information, word and character, and essentially learns which parts of the input query sequence are relevant to the current segmentation task and determines the influence of these parts. This segmentation technique is also combined with an efficient clustering method using a novel query relevance metric for end-to-end search task identification. Using real-world datasets, we demonstrate that our segmentation technique improves task identification accuracy of existing clustering-based algorithms.

Recent works using artificial neural networks based on word distributed representation greatly boost the performance of various natural language learning tasks, especially question answering. Though, they also carry along with some attendant problems, such as corpus selection for embedding learning, dictionary transformation for different learning tasks, etc. In this paper, we propose to straightforwardly model sentences by means of character sequences, and then utilize convolutional neural networks to integrate character embedding learning together with point-wise answer selection training. Compared with deep models pre-trained on word embedding (WE) strategy, our character-sequential representation (CSR) based method shows a much simpler procedure and more stable performance across different benchmarks. Extensive experiments on two benchmark answer selection datasets exhibit the competitive performance compared with the state-of-the-art methods.

Many mobile applications require frequent wireless transmissions between the content provider and mobile devices, consuming much energy in mobile devices. Motivated by the popularity of prefetch-friendly or delay-tolerant apps (e.g., social networking, app updates, cloud storage), we design and implement an application-layer transmission protocol, AppATP, which leverages cloud computing to manage data transmissions for mobile apps, transferring data to and from mobile devices in an energy efficient manner. Measurements show that significantly amount of energy is consumed by mobile devices during poor connectivity. Based on this observation, AppATP adaptively seizes periods of good bandwidth condition to prefetch frequently used data with minimum energy consumption, while deferring delay-tolerant data during poor network connectivity. Using the stochastic control framework, AppATP only relies on the current network information and data queue sizes to make an online decision on transmission scheduling, and performs well under unpredictable wireless network conditions. We implement AppATP on Samsung Note 2 smartphones and Amazon EC2. Results from both trace-driven simulations and extensive real-world experiments show that AppATP can be applied to a variety of application scenarios while achieving 30-50 percent energy savings for mobile devices.

Mobile cloud computing, promising to extend the capabilities of resource-constrained mobile devices, is emerging as a new computing paradigm which has fostered a wide range of exciting applications. In this new paradigm, efficient data transmission between the cloud and mobile devices becomes essential. This, however, is highly unreliable and unpredictable due to several uncontrollable factors, particularly the instability and intermittency of wireless connections, fluctuation of communication bandwidth, and user mobility. Consequently, this puts a heavy burden on the energy consumption of mobile devices. Confirmed by our experiments, significantly more energy is consumed during “bad” connectivity. Inspired by the feasibility to schedule data transmissions for prefetching-friendly or delay-tolerant applications, in this paper, we present eTime, a novel Energy-efficient data Transmission strategy between cloud and Mobile dEvices, based on Lyapunov optimization. It aggressively and adaptively seizes the timing of good connectivity to prefetch frequently used data while deferring delay-tolerant data in bad connectivity. To cope with the randomness and unpredictability of wireless connectivity, eTime only relies on the current status information to make a global energy-delay tradeoff decision. Our evaluations from both trace-driven simulation and realworld implementation show that eTime can be applied to various popular applications while achieving 20%-35% energy saving.