香港科技大学

Instruction tuning of Large Vision-language Models (LVLMs) has revolutionized the development of versatile models with zero-shot generalization across a wide range of downstream vision-language tasks. However, the diversity of different training tasks from various sources and formats would lead to inevitable task conflicts, where different tasks conflict for the same set of model parameters, resulting in sub-optimal instruction-following abilities. To address that, we propose the Mixture of Cluster-conditional LoRA Experts (MoCLE), a novel Mixture of Experts (MoE) architecture designed to activate task-customized model parameters based on instruction clusters. A separate universal expert is further incorporated to improve generalization abilities of MoCLE for novel instructions. Extensive experiments on InstructBLIP and LLaVA demonstrate the effectiveness of MoCLE.

Controllable generative models for images and videos have seen significant success, yet 3D scene generation, especially in unbounded scenarios like autonomous driving, remains underdeveloped. Existing methods lack flexible controllability and often rely on dense view data collection in controlled environments, limiting their generalizability across common datasets (e.g., nuScenes). In this paper, we introduce MagicDrive3D, a novel framework for controllable 3D street scene generation that combines video-based view synthesis with 3D representation (3DGS) generation. It supports multi-condition control, including road maps, 3D objects, and text descriptions. Unlike previous approaches that require 3D representation before training, MagicDrive3D first trains a multi-view video generation model to synthesize diverse street views. This method utilizes routinely collected autonomous driving data, reducing data acquisition challenges and enriching 3D scene generation. In the 3DGS generation step, we introduce Fault-Tolerant Gaussian Splatting to address minor errors and use monocular depth for better initialization, alongside appearance modeling to manage exposure discrepancies across viewpoints. Experiments show that MagicDrive3D generates diverse, high-quality 3D driving scenes, supports any-view rendering, and enhances downstream tasks like BEV segmentation, demonstrating its potential for autonomous driving simulation and beyond.

Online retailing has seen steady growth over the last decade. According to the Digital Commerce (formerly Internet Retailer) analysis of the US Commerce Department's year-end retail data, online sales constituted 16% of all retail sales in 2019, and is forecast to reach higher levels in the next years due to the impact of COVID-19. For an online retailer, one of the most important decisions is the products' display positioning as it plays a crucial role in shaping customers' shopping behavior. Empirical evidence abounds. Baye et al. [2] find that a consumer's likelihood of purchasing from a firm is strongly related to the order in which the firm is listed on a webpage by a search engine. In the online advertising industry, it has been widely observed that ads placed higher on a webpage attract more clicks from consumers [1]. Given the importance of product ranking positions, the key question for online retailers is how to rank the products to maximize the revenue. The question cannot be answered definitively, unless we can characterize and quantify how exactly customers react to products ranked in different positions. There are a number of reasons to explain the so-called position bias. The first reason is the limited attention of consumers. Eyetracking experiments show that the users are less likely to examine results near the bottom of the list. Besides limited attention, a customer seems to be more likely to buy a product ranked at the top, even though there is another similar product below inside her attention span. What explains this phenomenon at the individual level? Craswell et al. [3] provide a second explanation to the position bias using experiments, which is related to the satisficing behavior of customers. In particular, the customer views product sequentially and directly proceeds to purchasing a product once the utility of the product exceeds an acceptable threshold. The remaining products in the attention span are thus never viewed. Thus, positioning a brand or product at a top position on a list can improve both consumer attention and consumer selection of the brand.

Atmospheric deposition of particulate organic nitrogen (ONp) is a significant process in the global nitrogen cycle and may be pivotally important for N-limited ecosystems. However, past models largely overlooked the spatial and chemical inhomogeneity of atmospheric ONp and were thus deficient in assessing global ONp impacts. We constructed a comprehensive global model of atmospheric gaseous and particulate organic nitrogen (ON), including the latest knowledge on emissions and secondary formations. Using this model, we simulated global atmospheric ONp abundances consistent with observations. Our estimated global atmospheric ON deposition was 26 Tg N yr-1, predominantly in the form of ONp (23 Tg N yr-1) and mostly from wildfires (37%), oceans (22%) and aqueous productions (17%). Globally, ONp contributed as much as 40% to 80% of the total N deposition downwind of biomass-burning regions. Atmospheric ONp deposition thus constituted the dominant external N supply to the N-limited boreal forests, tundras and the Arctic Ocean, and its importance may be amplified in a future warming climate.

Trust relationships play a crucial role in various domains, such as social spam detection, retweet behavior analytics, and recommendation systems. Trust is often implicit and difficult to observe directly in the real world, as it is driven by people's underlying intentions and motivations. Therefore, when evaluating trust, it is critical to analyze not only user behavior data but also the intentions behind these behaviors that lead to trust. Existing trust evaluation methods often neglect the underlying reasons behind connections, such as shared hobbies or belonging to the same community. Therefore, these methods cannot differentiate the genuine intentions that lead to trust, resulting in an inaccurate evaluation of hidden trust relationships. To address this issue, we propose a novel Intent-based model for Trust Evaluation (INTRUST). This model can distinguish the intent behind high-order information in social communities using hypergraphs. Initially, we used hyperedges to represent high-order correlations between user-to-item and user-to-user interactions. Then, we construct K intent prototypes, which serve as foundational elements to build trust. Furthermore, we distinguish K-independent intent subgraphs from these high order correlations. To enhance the generalization and robustness of the model, we employ self-supervised learning and construct contrastive views at the node-level, hyperedge-level, and node hyperedge-level. Extensive experiments on real-world datasets demonstrate that our model outperforms state-of-the-art approaches in terms of trust evaluation accuracy and efficiency.