AAAI 2025 论文列表

Molecular design inherently involves the optimization of multiple conflicting objectives, such as enhancing bio-activity and ensuring synthesizability. Evaluating these objectives often requires resource-intensive computations or physical experiments. Current molecular design methodologies typically approximate the Pareto set using a limited number of molecules. In this paper, we present an innovative approach, called Multi-Objective Molecular Design through Learning Latent Pareto Set (MLPS). MLPS initially utilizes an encoder-decoder model to seamlessly transform the discrete chemical space into a continuous latent space. We then employ local Bayesian optimization models to efficiently search for local optimal solutions (i.e., molecules) within predefined trust regions. Using surrogate objective values derived from these local models, we train a global Pareto set learning model to understand the mapping between direction vectors (called “preferences”) in the objective space and the entire Pareto set in the continuous latent space. Both the global Pareto set learning model and local Bayesian optimization models collaborate to discover high-quality solutions and adapt the trust regions dynamically. Our work is an effective endeavor towards learning the Pareto set for multi-objective molecular design, providing decision-makers with the capability to fine-tune their preferences and thoroughly explore the Pareto set. Experimental results demonstrate that MLPS achieves state-of-the-art performance across various multi-objective scenarios, encompassing diverse objective types and varying numbers of objectives. The effectiveness of MLPS was further validated through real-world challenges in discovering antifungal peptides with low toxicity and high activity.

Heuristics are commonly used to tackle various search and optimization problems. Design heuristics usually require tedious manual crafting with domain knowledge. Recent works have incorporated Large Language Models (LLMs) into automatic heuristic search, leveraging their powerful language and coding capacity. However, existing research focuses on the optimal performance on the target problem as the sole objective, neglecting other criteria such as efficiency and scalability, which are vital in practice. To tackle this challenge, we propose to model the heuristic search as a multi-objective optimization problem and consider introducing additional practical criteria beyond optimal performance. Due to the complexity of the search space, conventional multi-objective optimization methods struggle to effectively handle LLM-based multi-objective heuristic search. We propose the first LLM-based multi-objective heuristic search framework, Multi-objective Evolution of Heuristic (MEoH), which integrates LLMs in a zero-shot manner to generate a non-dominated set of heuristics to meet multiple design criteria. We design a new dominance-dissimilarity mechanism for effective population management and selection, which incorporates both code dissimilarity in the search space and dominance in the objective space. MEoH is demonstrated in two well-known combinatorial optimization problems: the online Bin Packing Problem (BPP) and the Traveling Salesman Problem (TSP). The results indicate that a variety of elite heuristics are automatically generated in a single run, offering more trade-off options than the existing methods. It successfully achieves competitive or superior performance while improving efficiency up to 10 times. Moreover, we also observe that the multi-objective search introduces novel insights into heuristic design and leads to the discovery of diverse heuristics.

User data spread across multiple modalities has popularized multi-modal recommender systems (MMRS). They recommend diverse content such as products, social media posts, TikTok reels, etc., based on a user-item interaction graph. With rising data privacy demands, recent methods propose unlearning private user data from uni-modal recommender systems (RS). However, methods for unlearning item data related to outdated user preferences, revoked licenses, and legally requested removals are still largely unexplored. Previous RS unlearning methods are unsuitable for MMRS due to the incompatibility of their matrix-based representation with the multi-modal user-item interaction graph. Moreover, their data partitioning step degrades performance on each shard due to poor data heterogeneity and requires costly performance aggregation across shards. This paper introduces MMRecUn, the first approach known to us for unlearning in MMRS and unlearning item data. Given a trained RS model, MMRecUn employs a novel Reverse Bayesian Personalized Ranking (BPR) objective to enable the model to forget marked data. The reverse BPR attenuates the impact of user-item interactions within the forget set, while the forward BPR reinforces the significance of user-item interactions within the retain set. Our experiments demonstrate that MMRecUn outperforms baseline methods across various unlearning requests when evaluated on benchmark MMRS datasets. MMRecUn achieves recall performance improvements of up to 49.85% compared to baseline methods and is up to 1.3× faster than the Gold model, which is trained on retain set from scratch. MMRecUn offers significant advantages, including superiority in removing target interactions, preserving retained interactions, and zero overhead costs compared to previous methods.

Elucidating the functional mechanisms of the primary visual cortex (V1) remains a fundamental challenge in systems neuroscience. Current computational models face two critical limitations, namely the challenge of cross-modal integration between partial neural recordings and complex visual stimuli, and the inherent variability in neural characteristics across individuals, including differences in neuron populations and firing patterns. To address these challenges, we present a multi-modal identifiable variational autoencoder (miVAE) that employs a two-level disentanglement strategy to map neural activity and visual stimuli into a unified latent space. This framework enables robust identification of cross-modal correlations through refined latent space modeling. We complement this with a novel score-based attribution analysis that traces latent variables back to their origins in the source data space. Evaluation on a large-scale mouse V1 dataset demonstrates that our method achieves state-of-the-art performance in cross-individual latent representation and alignment, without requiring subject-specific fine-tuning, and exhibits improved performance with increasing data size. Significantly, our attribution algorithm successfully identifies distinct neuronal subpopulations characterized by unique temporal patterns and stimulus discrimination properties, while simultaneously revealing stimulus regions that show specific sensitivity to edge features and luminance variations. This scalable framework offers promising applications not only for advancing V1 research but also for broader investigations in neuroscience.

Health-risk behaviors such as overeating and smoking have a profound impact on public health, making their monitoring and mitigation critical. Wearable RGB-Thermal cameras are being employed to monitor these behaviors by capturing hand-to-mouth (HTM) gestures, which are central to them. However, detection models relying on single modalities—either RGB or thermal—often struggle to accurately distinguish these confounding gestures due to inherent sensor limitations, such as sensitivity to lighting conditions or thermal occlusions. We present a family of fusion models that integrate RGB and thermal video data using early-, decision- , and a novel mid-fusion architecture, RGB-Thermal Fusion Video Network (RTFVNet), designed to enhance the recognition of HTM gestures associated with eating and smoking. Our evaluation shows that while decision fusion achieves the highest F1-score of 88% (0.44 TFLOPs), RTFVNet offers an optimal balance between performance (85%) and complexity (0.37 TFLOPs) for gesture classification of eating, smoking, and non-gesture activities.

Learning a perception and reasoning module for robotic assistants to plan steps to perform complex tasks based on natural language instructions often requires large free-form language annotations, especially for short high-level instructions. To reduce the cost of annotation, large language models (LLMs) are used as a planner with few data. However, when elaborating the steps, even the state-of-the-art planner that uses LLMs mostly relies on linguistic common sense, often neglecting the status of the environment at command reception, resulting in inappropriate plans. To generate plans grounded in the environment, we propose FLARE (Few-shot Language with environmental Adaptive Replanning Embodied agent), which improves task planning using both language command and environmental perception. As language instructions often contain ambiguities or incorrect expressions, we additionally propose to correct the mistakes using visual cues from the agent. The proposed scheme allows us to use a few language pairs thanks to the visual cues and outperforms state-of-the-art approaches. Our code and the dataset are publicly available to facilitate further research.

Knowledge distillation (KD) has become a prevalent technique for compressing large language models (LLMs). Existing KD methods are constrained by the need for identical tokenizers (i.e., vocabularies) between teacher and student models, limiting their versatility in handling LLMs of different architecture families. In this paper, we introduce the Multi-Level Optimal Transport (MultiLevelOT), a novel approach that advances the optimal transport for universal cross-tokenizer knowledge distillation. Our method aligns the logit distributions of the teacher and the student at both token and sequence levels using diverse cost matrices, eliminating the need for dimensional or token-by-token correspondence. At the token level, MultiLevelOT integrates both global and local information by jointly optimizing all tokens within a sequence to enhance robustness. At the sequence level, we efficiently capture complex distribution structures of logits via the Sinkhorn distance, which approximates the Wasserstein distance for divergence measures. Extensive experiments on tasks such as extractive QA, generative QA, and summarization demonstrate that the MultiLevelOT outperforms state-of-the-art cross-tokenizer KD methods under various settings. Our approach is robust to different student and teacher models across model families, architectures, and parameter sizes.

The common matrix completion methods minimize the rank of the matrix to be completed in addition to the Hamming loss between the incomplete and completed matrices. The rank of matrix measures the linear relation among the vectors of matrix, which may introduce ambiguity for data recovery. To cope with this issue, we extend multi-label ranking loss into matrix completion, and employ multi-label ranking loss minimization (MLRM) in this paper to exploit the relative correlation among matrix vectors. In MLRM, the original incomplete matrix is converted into a pairwise ranking matrix, and the approximation on this newly generated matrix can be viewed as a surrogate of multi-label ranking loss to replace the Hamming loss pattern in the existing methods. Extensive experiments demonstrate that MLRM outperforms the state-of-the-art matrix completion methods in varies of applications, including movie recommendation, drug-target interaction prediction and multi-label learning.

Multi-label few-shot image classification is a crucial and challenging task due to limited annotated data and elusive category specificity. However, research on this topic is still in the rudimentary stage and few methods are available. Existing methods either leverage data augmentation to alleviate data scarcity or utilize label features as auxiliary knowledge to eliminate the negative effect caused by irrelevant categories, but they ignore the utilization of image region features for data augmentation, and overlook to learn appropriate text feature to better match the image features of specific categories. Moreover, these methods only focus on one side and do not effectively tackle the above two issues simultaneously. In this paper, we introduce a novel prototype-based multi-label few-shot learning framework that seamlessly integrates pairwise feature augmentation and flexible prompt learning. Specifically, by pairwise feature augmentation, we leverage the region features of images in the support set to generate more image features and construct image prototypes, thus alleviating the issue of data scarcity. By flexible prompt learning, we adaptively acquire class-specific prompts to build text prototypes that highly match the image features of specific classes, thereby mitigating the impact of irrelevant classes. Finally, with adaptive learnable parameters, we merge image and text prototypes to obtain the final prototypes, achieving a more powerful classifier for multi-label few-shot image classification. Extensive experimental results demonstrate that our proposed method can push the performance to a higher level.

Multi-instance multi-label classification (MIML) is a fundamental task in machine learning, where each data sample comprises a bag containing several instances and multiple binary labels. Despite its wide applications, the data collection process involves matching multiple instances and labels, typically resulting in high annotation costs. In this paper, we study a novel yet practical crowdsourced multi-instance multi-label classification (CMIML) setup, where labels are collected from multiple crowd sources. To address this problem, we first propose a novel data generation process for CMIML, i.e., cross-label transition, where cross-label annotation error is more likely to appear rather than previous single-label transition assumption, due to the inherent similarity of localized instances from different classes. Then, we formally define the cross-label transition by cross-label transition matrices which are dependent across classes. Subsequently, we establish the first unbiased risk estimator for CMIML and further improve it through aggregation techniques, along with a rigorous generalization error bound. We also provide a practical implementation of cross-label transition matrix estimation. Comprehensive experiments on six benchmark datasets under various scenarios demonstrate that our algorithm outperforms the baselines by a large margin, validating its effectiveness in handling the CMIML problem.

Current benchmarks for video segmentation are limited to annotating only salient objects (i.e., foreground instances). Despite their impressive architectural designs, previous works trained on these benchmarks have struggled to adapt to realworld scenarios. Thus, developing a new video segmentation dataset aimed at tracking multi-granularity segmentation target in the video scene is necessary. In this work, we aim to generate multi-granularity video segmentation dataset that is annotated for both salient and non-salient masks. To achieve this, we propose a large-scale, densely annotated multi-granularity video object segmentation (MUG-VOS) dataset that includes various types and granularities of mask annotations. We automatically collected a training set that assists in tracking both salient and non-salient objects, and we also curated a human-annotated test set for reliable evaluation. In addition, we present memory-based mask propagation model (MMPM), trained and evaluated on MUG-VOS dataset, which leads to the best performance among the existing video object segmentation methods and Segment SAM-based video segmentation methods.

Open intent classification is critical for the development of dialogue systems, aiming to accurately classify known intents into their corresponding classes while identifying unknown intents. Prior boundary-based methods assumed known intents fit within compact spherical regions, focusing on coarse-grained representation and precise spherical decision boundaries. However, these assumptions are often violated in practical scenarios, making it difficult to distinguish known intent classes from unknowns using a single spherical boundary. To tackle these issues, we propose a Multi-granularity Open intent classification method via adaptive Granular-Ball decision boundary (MOGB). Our MOGB method consists of two modules: representation learning and decision boundary acquiring. To effectively represent the intent distribution, we design a hierarchical representation learning method. This involves iteratively alternating between adaptive granular-ball clustering and nearest sub-centroid classification to capture fine-grained semantic structures within known intent classes. Furthermore, multi-granularity decision boundaries are constructed for open intent classification by employing granular-balls with varying centroids and radii. Extensive experiments conducted on three public datasets demonstrate the effectiveness of our proposed method.

With the continuous emergence of various social media platforms frequently used in daily life, the multimodal meme understanding (MMU) task has been garnering increasing attention. MMU aims to explore and comprehend the meanings of memes from various perspectives by performing tasks such as metaphor recognition, sentiment analysis, intention detection, and offensiveness detection. Despite making progress, limitations persist due to the loss of fine-grained metaphorical visual clue and the neglect of multimodal text-image weak correlation. To overcome these limitations, we propose a multi-granular multimodal clue fusion model (MGMCF) to advance MMU. Firstly, we design an object-level semantic mining module to extract object-level image feature clues, achieving fine-grained feature clue extraction and enhancing the model's ability to capture metaphorical details and semantics. Secondly, we propose a brand-new global-local cross-modal interaction model to address the weak correlation between text and images. This model facilitates effective interaction between global multimodal contextual clues and local unimodal feature clues, strengthening their representations through a bidirectional cross-modal attention mechanism. Finally, we devise a dual-semantic guided training strategy to enhance the model's understanding and alignment of multimodal representations in the semantic space. Experiments conducted on the widely-used MET-MEME bilingual dataset demonstrate significant improvements over state-of-the-art baselines. Specifically, there is an 8.14% increase in precision for offensiveness detection task, and respective accuracy enhancements of 3.53%, 3.89%, and 3.52% for metaphor recognition, sentiment analysis, and intention detection tasks. These results, underpinned by in-depth analyses, underscore the effectiveness and potential of our approach for advancing MMU.

Grounded Multimodal Named Entity Recognition (GMNER) is an emerging information extraction (IE) task, aiming to simultaneously extract entity spans, types, and corresponding visual regions of entities from given sentence-image pairs data. Recent unified methods employing machine reading comprehension or sequence generation-based frameworks show limitations in this difficult task. The former, utilizing human-designed type queries, struggles to differentiate ambiguous entities, such as Jordan (Person) and off-White x Jordan (Shoes). The latter, following the one-by-one decoding order, suffers from exposure bias issues. We maintain that these works misunderstand the relationships of multimodal entities. To tackle these, we propose a novel unified framework named Multi-grained Query-guided Set Prediction Network (MQSPN) to learn appropriate relationships at intra-entity and inter-entity levels. Specifically, MQSPN explicitly aligns textual entities with visual regions by employing a set of learnable queries to strengthen intra-entity connections. Based on distinct intra-entity modeling, MQSPN reformulates GMNER as a set prediction, guiding models to establish appropriate inter-entity relationships from a optimal global matching perspective. Additionally, we incorporate a query-guided Fusion Net (QFNet) as a glue network to boost better alignment of two-level relationships. Extensive experiments demonstrate that our approach achieves state-of-the-art performances in widely used benchmarks.

The rapid progress of multimedia technology has led to an increased focus on enhancing the quality of experience (QoE) for video. Specifically, the demand for low-latency and high-quality decoding has grown significantly. Compressed Video Quality Enhancement (CVQE) methods based on Deep Neural Networks (DNNs) have achieved remarkable success. However, most of the methods suffer from high computational complexity, thereby limiting their practicality in low-latency scenarios. Recently, Look-Up Table (LUT) methods have shown great efficiency, which makes them considerably promising in the field of low-latency CVQE. In this paper, we propose an efficient multi-frame deformable Look-Up Table structure for CVQE. Firstly, we design an efficient CNN to explore the inter-frame correlation and then predict the multi-scale convolution offsets. Secondly, we introduce a temporal feature extraction module and a multi-scale fusion module. We first exploit the predicted offsets to guide sampling for precise temporal alignment and extract multi-frame information. Then, higher quality frames are reconstructed from the fused multi-scale features. During the inference, we convert these two modules into LUTs to achieve a sound trade-off between model performance and computational complexity. Experiments demonstrate that our proposed method dramatically outperforms the state-of-the-art LUT-based methods, and obtains competitive performance compared to CNN-based methods with the capability to run in real-time(30fps) at 1080p resolution.

Multi-focus image fusion (MFIF) enhances depth of field in photography by generating an all-in-focus image from multiple images captured at different focal lengths. While deep learning has shown promise in MFIF, most existing methods overlooked the physical properties of defocus blurring in their network design, limiting their interoperability and generalization. This paper introduces a novel framework that integrates explicit defocus blur modelling into the MFIF process, improving both interpretability and performance. Using an atom-based spatially-varying parameterized defocus blurring model, our approach calculates pixel-wise defocus descriptors and initial focused images from multi-focus source images in a scale-recurrent manner to estimate soft decision maps. Fusion is then performed using masks derived from these decision maps, with special treatment for pixels likely defocused in all source images or near boundaries of defocused/focused regions. The model is trained with a fusion loss and a cross-scale defocus estimation loss. Extensive experiments on benchmark datasets demonstrated the effectiveness of our approach.

Edge computing-based video analytics faces data drift issues due to the occurrence of unseen objects or scenes in ever-changing environments. To maintain accuracy, continuous learning (CL) retrains stale models periodically with newly obtained data. However, it leads to unaffordable costs, as we must keep labeling drift data and retraining models. Regarding this concern, we first investigate video patterns across multiple cameras within an area and reveal significant data redundancies. We find that many of the same objects can be captured by multiple edge cameras or appear many times on the same edges. Our quantitative findings suggest that selecting a subset of high-quality data for CL is preferable over using a larger quantity. Yet, existing efforts for data acquisition have only focused on a single static dataset. These methods are not suitable for multi-edge video analytics scenarios, where videos are captured from multiple sources with non-iid data distribution. Hence, we propose a multi-edge collaborative active video acquisition (AVA) framework to collaboratively learn a reinforced video acquisition strategy to identify informative video frames from multiple edge nodes that best enhance model accuracy, avoiding redundancy across edges. Extensive experiments on three video datasets demonstrate that, our method achieves comparable performance to full-set video training while utilizing only 20% of the data in classification tasks. In object detection tasks, our methods can maintain productive accuracy with a reduction of nearly 70% in training video frames.

The autoregressive decoding paradigm endows large language models (LLMs) with superior language generation capabilities; however, its step-by-step decoding process inherently limits decoding speed. To mitigate these constraints, the prevalent “draft and validation” strategy enables parallel validation of candidate drafts, allowing LLMs to decode multiple tokens simultaneously during one model forward propagation. However, existing methodologies for obtaining drafts often incur additional overhead in communication or training process, or statistical biases from the corpus. To this end, we propose an innovative draft generation and maintenance approach that leverages the capabilities of LLM itself. Specifically, we extend the autoregressive decoding paradigm to a multi-branch drafting procedure, which can efficiently generate draft sequences without any additional models or training process, while preserving the quality of the generated content by maintaining LLM parameters. Experiments across various open-source benchmarks show that our method generates 2.0 to 3.2 tokens per forward step and achieves around 2 times improvement of end-to-end throughput compared to the autoregressive decoding strategy.

Current neural networks often employ multi-domain-learning or attribute-injecting mechanisms to incorporate non-independent and identically distributed (non-IID) information for text understanding tasks by capturing individual characteristics and the relationships among samples. However, the extent of the impact of non-IID information and how these methods affect pre-trained language models (PLMs) remains unclear. This study revisits the assumption that non-IID information enhances PLMs to achieve performance improvements from a Bayesian perspective, which unearths and integrates non-IID and IID features. Furthermore, we proposed a multi-attribute multi-grained framework for PLM adaptations (M2A), which combines multi-attribute and multi-grained views to mitigate uncertainty in a lightweight manner. We evaluate M2A through prevalent text-understanding datasets and demonstrate its superior performance, mainly when data are implicitly non-IID, and PLMs scale larger.