AAAI 2025 论文列表

The development of Large Language Models (LLMs) relies on extensive text corpora, which are often unevenly distributed across languages. This imbalance results in LLMs performing significantly better on high-resource languages like English, German, and French, while their capabilities in low-resource languages remain inadequate. Currently, there is a lack of quantitative methods to evaluate the performance of LLMs in these low-resource languages. To address this gap, we propose the Language Ranker, an intrinsic metric designed to benchmark and rank languages based on LLM performance using internal representations. By comparing the LLM's internal representation of various languages against a baseline derived from English, we can assess the model's multilingual capabilities in a robust and language-agnostic manner. Our analysis reveals that high-resource languages exhibit higher similarity scores with English, demonstrating superior performance, while low-resource languages show lower similarity scores, underscoring the effectiveness of our metric in assessing language-specific capabilities. Besides, the experiments show that there is a strong correlation between the LLM’s performance in different languages and the proportion of those languages in its pre-training corpus. These insights underscore the efficacy of the Language Ranker as a tool for evaluating LLM performance across different languages, particularly those with limited resources.

A new trend in the computer vision community is to capture objects of interest following flexible human command represented by a natural language prompt. However, the progress of using language prompts in driving scenarios is stuck in a bottleneck due to the scarcity of paired prompt-instance data. To address this challenge, we propose the first object-centric language prompt set for driving scenes within 3D, multi-view, and multi-frame space, named NuPrompt. It expands nuScenes dataset by constructing a total of 40,147 language descriptions, each referring to an average of 7.4 object tracklets. Based on the object-text pairs from the new benchmark, we formulate a novel prompt-based driving task, \ie, employing a language prompt to predict the described object trajectory across views and frames. Furthermore, we provide a simple end-to-end baseline model based on Transformer, named PromptTrack. Experiments show that our PromptTrack achieves impressive performance on NuPrompt. We hope this work can provide some new insights for the self-driving community.

The representation learning of time series has a wide range of downstream tasks and applications in many practical scenarios. However, due to the complexity, spatiotemporality, and continuity of sequential stream data, compared with the representation learning of structural data such as images/videos, the time series self-supervised representation learning is even more challenging. Besides, the direct application of existing contrastive learning and masked autoencoder based approaches to time series representation learning encounters inherent theoretical limitations, such as ineffective augmentation and masking strategies. To this end, we propose a Language Pre-training guided Masking Representation Learning (LPMRL) for times series classification. Specifically, we first propose a novel language pre-training guided masking encoder for adaptively sampling semantic spatiotemporal patches via natural language descriptions and improving the discriminability of latent representations. Furthermore, we present the dual-information contrastive learning mechanism to explore both local and global information by meticulously designing high-quality hard negative samples of time series data samples. As a result, we also design various experiments, such as visualization of masking position and distribution and reconstruction error to verify the reasonability of proposed language guided masking technique. Last, we evaluate the performance of proposed representation learning via classification task conducted on 106 time series datasets, which demonstrates the effectiveness of proposed method.

Document summarization has greatly benefited from advances in large language models (LLMs). In real-world situations, summaries often need to be generated from multiple documents with diverse sources and authors, lacking a clear information flow. Naively concatenating these documents and generating a summary can lead to poorly structured narratives and redundancy. Additionally, attributing each part of the generated summary to a specific source is crucial for reliability. In this study, we address multi-document summarization with attribution using our proposed solution ***MiDAS-PRo***, consisting of three stages: (i) Planning the hierarchical organization of source documents, (ii) Reasoning by generating relevant entities/topics, and (iii) Summary Generation. We treat the first two sub-problems as a code completion task for LLMs. By incorporating well-selected in-context learning examples through a graph attention network, LLMs effectively generate plans and reason topics for a document collection. Experiments on summarizing scientific articles from public datasets show that our approach outperforms state-of-the-art baselines in both automated and human evaluations.

Pretrained transformer-based Language Models (LMs) are well-known for their ability to achieve significant improvement on NLP tasks, but their black-box nature, which leads to a lack of interpretability, has been a major concern. My dissertation focuses on developing intrinsically interpretable models when using LMs as encoders while maintaining their superior performance via prototypical networks. I initiated my research by investigating enhancements in performance for interpretable models of sarcasm detection. My proposed approach focuses on capturing sentiment incongruity to enhance accuracy while offering instance-based explanations for the classification decisions. Later, we develop a novel white-box multi-head graph attention-based prototypical framework designed to explain the decisions of text classification models without sacrificing the accuracy of the original black-box LMs. In addition, I am working on extending the attention-based prototypical framework with contrastive learning to redesign an interpretable graph neural network for document classification, aiming to enhance both the interpretability and performance of the model in document classification.

Dialogue serves as the most natural manner of human-computer interaction (HCI). Recent advancements in speech language models (SLM), have significantly enhanced speech-based conversational AI. However, these models are limited to turn-based conversation, lacking the ability to interact with humans in real-time spoken scenarios, for example, being interrupted when the generated content is not satisfactory. To address these limitations, we explore full duplex modeling (FDM) in interactive speech language models (iSLM), focusing on enhancing real-time interaction and, more explicitly, exploring the quintessential ability of interruption. We introduce a novel model design, namely listening-while-speaking language model (LSLM), an end-to-end system equipped with both listening and speaking channels. Our LSLM employs a token-based decoder-only TTS for speech generation and a streaming self-supervised learning (SSL) encoder for real-time audio input. LSLM fuses both channels for autoregressive generation and detects turn-taking in real time. Three fusion strategies—early fusion, middle fusion, and late fusion—are explored, with middle fusion achieving an optimal balance between speech generation and real-time interaction. Two experimental settings, command-based FDM and voice-based FDM, demonstrate LSLM’s robustness to noise and sensitivity to diverse instructions. Our results highlight LSLM’s capability to achieve duplex communication with minimal impact on existing systems. This study aims to advance the development of interactive speech dialogue systems, enhancing their applicability in real-world contexts.

We consider nonconvex optimization problem over simplex, and more generally, a product of simplices. We provide an algorithm, Langevin Multiplicative Weights Update (LMWU) for solving global optimization problems by adding a noise scaling with the non-Euclidean geometry in the simplex. Non-convex optimization has been extensively studied by machine learning community due to its application in various scenarios such as neural network approximation and finding Nash equilibrium. Despite recent progresses on provable guarantee of escaping and avoiding saddle point (convergence to local minima) and global convergence of Langevin gradient based method without constraints, the global optimization with constraints is less studied. We show that LMWU algorithm is provably convergent to interior global minima with a non-asymptotic convergence analysis. We verify the efficiency of the proposed algorithm in real data set from polynomial portfolio management, where optimization of a highly non-linear objective function plays a crucial role.

We present a significant advancement in the field of Langevin Monte Carlo (LMC) methods by introducing the Inexact Proximal Langevin Algorithm (IPLA). This novel algorithm broadens the scope of problems that LMC can effectively address while maintaining controlled computational costs. IPLA extends LMC's applicability to potentials that are convex, strongly convex in the tails, and exhibit polynomial growth, beyond the conventional L-smoothness assumption. Moreover, we extend LMC's applicability to super-quadratic potentials and offer improved convergence rates over existing algorithms. Additionally, we provide bounds on all moments of the Markov chain generated by IPLA, enhancing its analytical robustness.

Vertical Federated Learning (VFL) involves multiple clients collaborating to train a global model, with distributed features of shared samples. While it becomes a critical privacy-preserving learning paradigm, its security can be significantly compromised by backdoor attacks, where a malicious client injects a target backdoor by manipulating local data. Existing attack methods in VFL rely on the assumption that the malicious client can obtain additional knowledge about task labels, which is not applicable in VFL. In this work, we investigate a new backdoor attack paradigm in VFL, Label-Free Backdoor Attacks (LFBA), which does not require any additional task label information and is feasible in VFL settings. Specifically, while existing methods assume access to task labels or target-class samples, we demonstrate that the gradients of local embeddings reflect the semantic information of labels. It can be utilized to construct the target poison sample set. Besides, we uncover that backdoor triggers tend to be ignored and under-fitted due to the learning of original features, which hinders backdoor task optimization. To address this, we propose selectively switching poison samples to disrupt feature learning, promoting backdoor task learning while maintaining accuracy on clean data. Extensive experiments demonstrate the effectiveness of our method in various settings.

The accurate segmentation of guidewires in interventional cardiac fluoroscopy videos is crucial for computer-aided navigation tasks. Although deep learning methods have demonstrated high accuracy and robustness in wire segmentation, they require substantial annotated datasets for generalizability, underscoring the need for extensive labeled data to enhance model performance. To address this challenge, we propose the Segmentation-guided Frame-consistency Video Diffusion Model (SF-VD) to generate large collections of labeled fluoroscopy videos, augmenting the training data for wire segmentation networks. SF-VD leverages videos with limited annotations by independently modeling scene distribution and motion distribution. It first samples the scene distribution by generating 2D fluoroscopy images with wires positioned according to a specified input mask, and then samples the motion distribution by progressively generating subsequent frames, ensuring frame-to-frame coherence through a frame-consistency strategy. A segmentation-guided mechanism further refines the process by adjusting wire contrast, ensuring a diverse range of visibility in the synthesized image. Evaluation on a fluoroscopy dataset confirms the superior quality of the generated videos and shows significant improvements in guidewire segmentation.

The ubiquitous and unavoidable label noise brings great challenges to the generalization performance of learning methods.Label noise correction aims to detect and correct label noise in the data, which is one of the most potential methods to address this challenge.Current methods for label noise filtering that utilize primitive features primarily concentrate on identifying noise, which often limits their capacity to adaptively learn features crucial for specific tasks, thereby resulting in a higher rate of noise identification within the noise recognition process. On the other hand, deep neural networks, endowed with robust feature extraction capabilities, typically exhibit lower noise identification, as they are prone to fitting noise patterns during the recognition process, potentially undermining their overall efficacy. Moreover, Fuzzy Learning Machine (FLM) excels not only in feature extraction but also in noise tolerance, adeptly navigating data uncertainties. FLM enhances the accuracy of the labels by calculating the membership degrees of samples across categories and determining their fuzzy memberships. The introduction of a two-stage FLM-based framework, which employs a secondary learning mechanism for precise noise filtering and correction, has shown substantial improvements in noise correction across various large-scale noisy datasets, thereby significantly enhancing samples' quality and boosting the generalization capabilities of classifiers.

Quality control is a crucial issue of label data collection by crowdsourcing. Typically, aggregation methods to redundant crowd labels are proposed for estimating high-quality labels from noisy crowd labels. Most of the existing works concentrate on the label aggregation for Single Crowd Tasks (SCTs) which have a single object set with homogeneous question types. However, it is useful for a requester to combine multiple relevant but different crowd tasks into a Composite Crowd Task (CCT) which have heterogeneous question types and (or) multiple object sets for diverse purposes. Instead of the label aggregation on each crowd task respectively, label aggregation methods by bridging multiple SCTs in CCTs can potentially improve the label quality of all tasks. In this paper, we propose a general label aggregation approach for such CCTs by worker ability constraint satisfaction and relaxed optimization. We collected real crowd datasets of CCTs with diverse task settings based on heterogeneous question types, including categorization, pairwise preference comparisons, and pairwise similarity comparisons. The results demonstrate that our approach can effectively bridge the worker information of CCTs to improve the quality of aggregated labels and outperforms the baselines proposed for SCTs.

Visual object tracking is essentially crucial for unmanned aerial vehicles (UAVs). Despite the substantial progress, most of the existing UAV trackers are designed for well-conditioned daytime data, while for the scenarios in challenging weather condition, e.g. foggy or nighttime environment, the tremendous domain gap leads to significant performance degradation. To address this issue, in this paper, we propose a novel robust UAV tracker termed LVPTrack, which conducts high quality label-aligned visual prompt tuning to adapt to various challenging weather conditions. Specifically, we first synthesize the sequential foggy and nighttime video frames to assist the model training. A domain adaptive teacher-student network is utilized to distill the hierarchical visual semantic of the target objects in cross-domain scenarios. Then we propose a target-aware pseudo-label voting (PLV) strategy to alleviate the target-level misalignment in the dual domains. Furthermore, we propose a dynamic aggregated prompt (DAP) module to facilitate the appearance variation adaptation of the target object in challenging scenarios. Extensive experiments demonstrate that our tracker achieves superior performance over existing state-of-the-art UAV trackers.

We investigate the synthesis of policies for high-level agent programs expressed in Golog, a language based on situation calculus that incorporates nondeterministic programming constructs. Unlike traditional approaches for program realization that assume full agent control or rely on incremental search, we address scenarios where environmental nondeterminism significantly influences program outcomes. Our synthesis problem involves deriving a policy that successfully realizes a given Golog program while ensuring the satisfaction of a temporal specification, expressed in Linear Temporal Logic on finite traces (LTLf), across all possible environmental behaviors. By leveraging an expressive class of first-order action theories, we construct a finite game arena that encapsulates program executions and tracks the satisfaction of the temporal goal. A game-theoretic approach is employed to derive such a policy. Experimental results demonstrate this approach's feasibility in domains with unbounded objects and non-local effects. This work bridges agent programming and temporal logic synthesis, providing a framework for robust agent behavior in nondeterministic environments.

We study the problem of realizing strategies for an LTLf goal specification while ensuring that at least an LTLf backup specification is satisfied in case of unreliability of certain input variables. We formally define the problem and characterize its worst-case complexity as 2EXPTIME-complete, like standard LTLf synthesis. Then we devise three different solution techniques: one based on direct automata manipulation, which is 2EXPTIME, one disregarding unreliable input variables by adopting a belief construction, which is 3EXPTIME, and one leveraging second-order quantified LTLf (QLTLf), which is 2EXPTIME and allows for a direct encoding into monadic second-order logic, which in turn is worst-case nonelementary. We prove their correctness and evaluate them against each other empirically. Interestingly, theoretical worst-case bounds do not translate into observed performance; the MSO technique performs best, followed by belief construction and direct automata manipulation. As a byproduct of our study, we provide a general synthesis procedure for arbitrary QLTLf specifications.

Session-Based Recommendation (SBR) based on Graph Neural Networks (GNN) has become a new paradigm for recommender systems, and plays a fundamental role in e-commerce and other relevant domains. Existing graph aggregation methods primarily form node representations by capturing basic relationships between neighboring and central nodes. Despite their encouraging results, the global relationships of items and user intentions within sessions typically change over time, which degrades the effectiveness of existing embedding schemes. To resolve this challenge, we propose a Long and Short-Term Temporal Graph Neural Network (LS-TGNN) for SBR. LS-TGNN employs a novel temporal session graph to aggregate neighborhood information, and models user interests from both long and short-term perspectives. Specifically, we design long-term and short-term encoders to model the long and short-term interests of users, respectively. In order to better model the interests of users in different time dimensions, we introduce an item-granularity method that distinguishes between long and short-term interests. Extensive experiments on three widely used datasets demonstrate that LS-TGNN outperforms existing methods with a large margin.

Multilingual large language-vision models (LVLMs), which understand and generate both text and images across multiple languages, have achieved remarkable performance on English-centric multimodal generation tasks. However, their performance on non-English tasks has been underwhelming. One major challenge with multilingual LVLMs is the modality gap between visual inputs and multilingual textual inputs/outputs due to the lack of high-quality multilingual training data. In this paper, we propose LRM-LLaVA, a multilingual large language-vision model designed for low-resource languages to overcome the modality gap. It is composed of four components: a visual encoder, a multilingual large language model, a vision-text representation projector, and a cross-modal regularizer. Both the projector and regularizer aim at reducing the modality gap and improving multilingual performance. To train LRM-LLaVA, we employ a two-stage training strategy including pre-training and instruction fine-tuning. Meanwhile, we construct a multilingual visual question answering dataset based on English open-source datasets and adopt multiple task instructions. To evaluate the performance of LVLMs across various languages, we construct four multilingual benchmarks for 10 languages, based on English open-source benchmarks. Experimental results show that LRM-LLaVA achieves competitive performance compared to other multilingual LVLMs of similar parameters.

Recently, there has been considerable exploration of methods for generating 3D point clouds, which is crucial for numerous 3D vision applications. Though conditional generation methods show promising performance, it depends on the additional paired label. On the other hand, unconditional generation methods usually fail to annotate the generated 3D point cloud. In this paper, we introduce a novel self-conditional architecture that trains on unlabeled data and then generates high-quality labeled 3D point clouds. Specifically, we design a module to extract geometry and view features, and then use a feature fusion module to integrate them as a substitute for label embedding in conditional point cloud generation. Then the point cloud generator is trained using the fused features. LPCG also harnesses CLIP to handle the view features of point clouds for generating label information. Besides, we train two feature diffusion modules to capture the essence of multimodal features and obtain diverse fused features for use as conditions in generating point clouds. Experiments on the ShapeNet dataset demonstrate that LPCG achieves state-of-the-art performance for single class generation. Our experimental results show that the accuracy of our generated label annotations reaches around 97.44% for a two-class generation task.

Vision-based 3D occupancy prediction has become a popular research task due to its versatility and affordability. Nowadays, conventional methods usually project the image-based vision features to 3D space and learn the geometric information through the attention mechanism, enabling the 3D semantic occupancy prediction. However, these works usually face two main challenges: 1) Limited geometric information. Due to the lack of geometric information in the image itself, it is challenging to directly predict 3D space information, especially in large-scale outdoor scenes. 2) Local restricted interaction. Due to the quadratic complexity of the attention mechanism, they often use modified local attention to fuse features, resulting in a restricted fusion. To address these problems, in this paper, we propose a language-assisted 3D semantic occupancy prediction network, named LOMA. In the proposed vision-language framework, we first introduce a VL-aware Scene Generator (VSG) module to generate the 3D language feature of the scene. By leveraging the vision-language model, this module provides implicit geometric knowledge and explicit semantic information from the language. Furthermore, we present a Tri-plane Fusion Mamba (TFM) block to efficiently fuse the 3D language feature and 3D vision feature. The proposed module not only fuses the two features with global modeling but also avoids too much computation costs. Experiments on the SemanticKITTI and SSCBench-KITTI360 datasets show that our algorithm achieves new state-of-the-art performances in both geometric and semantic completion tasks. Our code will be open soon.