Huixue Zhou, Hengrui Gu, Zaifu Zhan, Xi Liu, Kaixiong Zhou, Yongkang Xiao, Mingfu Liang, Srinivas Prasad Govindan, Piyush Chawla, Jiyan Yang, Xiangfei Meng, Huayu Li, Buyun Zhang, Liang Luo, Wen-Yen Chen, Yiping Han, Bo Long, Rui Zhang, Tianlong Chen. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Existing progress in object keypoint estimation primarily benefits from the conventional supervised learning paradigm based on numerous data labeled with pre-defined keypoints. However, these well-trained models can hardly detect the undefined new keypoints in test time, which largely hinders their feasibility for diverse downstream tasks. To handle this, various solutions are explored but still suffer from either limited generalizability or transferability. Therefore, in this paper, we explore a novel keypoint learning paradigm in that we only annotate new keypoints in the new data and incrementally train the model, without retaining any old data, called Incremental object Keypoint Learning (IKL). A two-stage learning scheme as a novel baseline tailored to IKL is developed. In the first Knowledge Association stage, given the data labeled with only new keypoints, an auxiliary KA-Net is trained to automatically associate the old keypoints to these new ones based on their spatial and intrinsic anatomical relations. In the second Mutual Promotion stage, based on a keypoint-oriented spatial distillation loss, we jointly leverage the auxiliary KA-Net and the old model for knowledge consolidation to mutually promote the estimation of all old and new keypoints. Owing to the investigation of the correlations between new and old keypoints, our proposed method can not just effectively mitigate the catastrophic forgetting of old keypoints, but may even further improve the estimation of the old ones and achieve a positive transfer beyond anti-forgetting. Such an observation has been solidly verified by extensive experiments on different keypoint datasets, where our method exhibits superiority in alleviating the forgetting issue and boosting performance while enjoying labeling efficiency even under the low-shot data regime.

Prevailing Multimodal Large Language Models (MLLMs) encode the input image(s) as vision tokens and feed them into the language backbone, similar to how Large Language Models (LLMs) process the text tokens. However, the number of vision tokens increases quadratically as the image resolutions, leading to huge computational costs. In this paper, we consider improving MLLM’s efficiency from two scenarios, (I) Reducing computational cost without degrading the performance. (II) Improving the performance with given budgets. We start with our main finding that the ranking of each vision token sorted by attention scores is similar in each layer except the first layer. Based on it, we assume that the number of essential top vision tokens does not increase along layers. Accordingly, for Scenario I, we propose a greedy search algorithm (G-Search) to find the least number of vision tokens to keep at each layer from the shallow to the deep. Interestingly, G-Search is able to reach the optimal reduction strategy based on our assumption. For Scenario II, based on the reduction strategy from G-Search, we design a parametric sigmoid function (P-Sigmoid) to guide the reduction at each layer of the MLLM, whose parameters are optimized by Bayesian Optimization. Extensive experiments demonstrate that our approach can significantly accelerate those popular MLLMs, e.g. LLaVA, and InternVL2 models, by more than 2⇥ without performance drops. Our approach also far outperforms other token reduction methods when budgets are limited, achieving a better trade-off between efficiency and effectiveness.

Recent advancements in multimodal large language models (MLLMs) have garnered significant attention, offering a promising pathway toward artificial general intelligence (AGI).Among the essential capabilities required for AGI, creativity has emerged as a critical trait for MLLMs, with association serving as its foundation.Association reflects a model's ability to think creatively, making it vital to evaluate and understand.While several frameworks have been proposed to assess associative ability, they often overlook the inherent ambiguity in association tasks, which arises from the divergent nature of associations and undermines the reliability of evaluations.To address this issue, we decompose ambiguity into two types-internal ambiguity and external ambiguity-and introduce AssoCiAm, a benchmark designed to evaluate associative ability while circumventing the ambiguity through a hybrid computational method.We then conduct extensive experiments on MLLMs, revealing a strong positive correlation between cognition and association.Additionally, we observe that the presence of ambiguity in the evaluation process causes MLLMs' behavior to become more random-like.Finally, we validate the effectiveness of our method in ensuring more accurate and reliable evaluations.See Project Page for the data and codes.

This paper proposes AutoScape, a long-horizon driving scene generation framework. At its core is a novel RGB-D diffusion model that iteratively generates sparse, geometrically consistent keyframes, serving as reliable anchors for the scene's appearance and geometry. To maintain long-range geometric consistency, the model 1) jointly handles image and depth in a shared latent space, 2) explicitly conditions on the existing scene geometry (i.e., rendered point clouds) from previously generated keyframes, and 3) steers the sampling process with a warp-consistent guidance. Given high-quality RGB-D keyframes, a video diffusion model then interpolates between them to produce dense and coherent video frames. AutoScape generates realistic and geometrically consistent driving videos of over 20 seconds, improving the long-horizon FID and FVD scores over the prior state-of-the-art by 48.6\% and 43.0\%, respectively.

In multimodal recommender systems (MRS), integrating various modalities helps to model user preferences and item characteristics more accurately, thereby assisting users in discovering items that match their interests. Although the introduction of multimodal information offers opportunities for performance improvement, it will increase the risks of inherent noise and information redundancy, posing challenges to the robustness of MRS. Many existing methods typically address these two issues separately either by introducing perturbations at the model input for robust training to handle noise or by designing complex network structures to filter out redundant information. In contrast, we propose the DVIB framework to simultaneously address both issues in a simple manner. We found that moving the perturbations from the input layer to the hidden layer, combined with feature self-distillation, can mitigate noise and handle information redundancy without altering the original network architecture. Additionally, we also provide theoretical evidence for the effectiveness of DVIB, demonstrating that the framework not only explicitly enhances the robustness of model training but also implicitly exhibits an information bottleneck effect, which effectively reduces redundant information during multimodal fusion and improves feature extraction quality. Extensive experiments show that DVIB consistently improves the performance of MRS across different datasets and model settings, and it can complement existing robust training methods, representing a promising new paradigm in MRS. See code at https://github.com/MarshmallowLight/DVIB.git.

Active recognition enables robots to intelligently explore novel observations, thereby acquiring more information while circumventing undesired viewing conditions. Recent approaches favor learning policies from simulated or collected data, wherein appropriate actions are more frequently selected when the recognition is accurate. However, most recognition modules are developed under the closed-world assumption, which makes them ill-equipped to handle unexpected inputs, such as the absence of the target object in the current observation. To address this issue, we propose treating active recognition as a sequential evidence-gathering process, providing by-step uncertainty quantification and reliable prediction under the evidence combination theory. Additionally, the reward function developed in this paper effectively characterizes the merit of actions when operating in open-world environments. To evaluate the performance, we collect a dataset from an indoor simulator, encompassing various recognition challenges such as distance, occlusion levels, and visibility. Through a series of experiments on recognition and robustness analysis, we demonstrate the necessity of introducing uncertainties to active recognition and the superior performance of the proposed method.

Autonomous vehicle (AV) systems rely on robust perception models as a cornerstone of safety assurance. However, objects encountered on the road exhibit a long-tailed distri-bution, with rare or unseen categories posing challenges to a deployed perception model. This necessitates an expen-sive process of continuously curating and annotating data with significant human effort. We propose to leverage recent advances in vision-language and large language models to design an Automatic Data Engine (AIDE) that automati-cally identifies issues, efficiently curates data, improves the model through auto-labeling, and verifies the model through generation of diverse scenarios. This process operates it-eratively, allowing for continuous self-improvement of the model. We further establish a benchmark for open-world detection on AV datasets to comprehensively evaluate vari-ous learning paradigms, demonstrating our method's supe-rior performance at a reduced cost.

Recently many plug-and-play self-attention modules (SAMs) are proposed to enhance the model performance by exploiting the internal information of deep convolutional neural networks. However, most SAMs connect individually with each block of the backbone for granted, leading to incremental computational cost and the number of parameters with the growth of network depth. To address this issue, we first empirically and theoretically explore the Lottery Ticket Hypothesis for Self-attention Networks (LTH4SA): a full self-attention network contains a subnetwork with sparse self-attention connections that can (1) accelerate inference, (2) reduce extra parameter increment, and (3) maintain accuracy. Furthermore, we propose a simple yet effective policy-gradient-based baseline method to search the ticket, i.e., the connection scheme that satisfies the three above-mentioned conditions. Extensive experiments on widely-used benchmarks and popular self-attention networks show the effectiveness of our method. Besides, our experiments illustrate that our searched ticket has the capacity of transferring to other vision tasks, e.g., crowd counting and segmentation. https://github.com/gbup-group/EAN-efficient-attention-network.

We propose the attention-inspired numerical solver (AttNS), a concise method that helps the generalization and robustness issues faced by the AI-Hybrid numerical solver in solving differential equations due to limited data. AttNS is inspired by the effectiveness of attention modules in Residual Neural Networks (ResNet) in enhancing model generalization and robustness for conventional deep learning tasks. Drawing from the dynamical system perspective of ResNet, We seamlessly incorporate attention mechanisms into the design of numerical methods tailored for the characteristics of solving differential equations. Our results on benchmarks, ranging from high-dimensional problems to chaotic systems, showcase AttNS consistently enhancing various numerical solvers without any intricate model crafting. Finally, we analyze AttNS experimentally and theoretically, demonstrating its ability to achieve strong generalization and robustness while ensuring the convergence of the solver. This includes requiring less data compared to other advanced methods to achieve comparable generalization errors and better prevention of numerical explosion issues when solving differential equations.

The self-attention mechanism (SAM) is widely used in various fields of artificial intelligence and has successfully boosted the performance of different models. However, current explanations of this mechanism are mainly based on intuitions and experiences, while there still lacks direct modeling for how the SAM helps performance. To mitigate this issue, in this paper, based on the dynamical system perspective of the residual neural network, we first show that the intrinsic stiffness phenomenon (SP) in the high-precision solution of ordinary differential equations (ODEs) also widely exists in high-performance neural networks (NN). Thus the ability of NN to measure SP at the feature level is necessary to obtain high performance and is an important factor in the difficulty of training NN. Similar to the adaptive step-size method which is effective in solving stiff ODEs, we show that the SAM is also a stiffness-aware step size adaptor that can enhance the model's representational ability to measure intrinsic SP by refining the estimation of stiffness information and generating adaptive attention values, which provides a new understanding about why and how the SAM can benefit the model performance. This novel perspective can also explain the lottery ticket hypothesis in SAM, design new quantitative metrics of representational ability, and inspire a new theoretic-inspired approach, StepNet. Extensive experiments on several popular benchmarks demonstrate that StepNet can extract fine-grained stiffness information and measure SP accurately, leading to significant improvements in various visual tasks.

Knowledge-based visual question answering (VQA) requires external knowledge to answer the question about an image. Early methods explicitly retrieve knowledge from external knowledge bases, which often introduce noisy information. Recently large language models like GPT-3 have shown encouraging performance as implicit knowledge source and revealed planning abilities. However, current large language models can not effectively understand image inputs, thus it remains an open problem to extract the image information and input to large language models. Prior works have used image captioning and object descriptions to represent the image. However, they may either drop the essential visual information to answer the question correctly or involve irrelevant objects to the task-of-interest. To address this problem, we propose to let large language models make an initial hypothesis according to their knowledge, then actively collect the visual evidence required to verify the hypothesis. In this way, the model can attend to the essential visual information in a task-oriented manner. We leverage several vision modules from the perspectives of spatial attention (i.e., Where to look) and attribute attention (i.e., What to look), which is similar to human cognition. The experiments show that our proposed method outperforms the baselines on open-ended knowledge-based VQA datasets and presents clear reasoning procedure with better interpretability.

Network pruning for deep convolutional neural networks (CNNs) has recently achieved notable research progress on image-level classification. However, most existing pruning methods are not catered to or evaluated on semantic segmentation networks. In this paper, we advocate the importance of contextual information during channel pruning for semantic segmentation networks by presenting a novel Context-aware Pruning framework. Concretely, we formulate the embedded contextual information by leveraging the layer-wise channels interdependency via the Context-aware Guiding Module (CAGM) and introduce the Context-aware Guided Sparsification (CAGS) to adaptively identify the informative channels on the cumbersome model by inducing channel-wise sparsity on the scaling factors in batch normalization (BN) layers. The resulting pruned models require significantly lesser operations for inference while maintaining comparable performance to (at times outperforming) the original models. We evaluated our framework on widely-used benchmarks and showed its effectiveness on both large and lightweight models. On Cityscapes dataset, our framework reduces the number of parameters by 32%, 47%, 54%, and 63%, on PSPNet101, PSPNet50, ICNet, and SegNet, respectively, while preserving the performance.

Attention networks have successfully boosted the performance in various vision problems. Previous works lay emphasis on designing a new attention module and individually plug them into the networks. Our paper proposes a novel-and-simple framework that shares an attention module throughout different network layers to encourage the integration of layer-wise information and this parameter-sharing module is referred to as Dense-and-Implicit-Attention (DIA) unit. Many choices of modules can be used in the DIA unit. Since Long Short Term Memory (LSTM) has a capacity of capturing long-distance dependency, we focus on the case when the DIA unit is the modified LSTM (called DIA-LSTM). Experiments on benchmark datasets show that the DIA-LSTM unit is capable of emphasizing layer-wise feature interrelation and leads to significant improvement of image classification accuracy. We further empirically show that the DIA-LSTM has a strong regularization ability on stabilizing the training of deep networks by the experiments with the removal of skip connections (He et al. 2016a) or Batch Normalization (Ioffe and Szegedy 2015) in the whole residual network.

Batch Normalization (BN) (Ioffe and Szegedy 2015) normalizes the features of an input image via statistics of a batch of images and hence BN will bring the noise to the gradient of training loss. Previous works indicate that the noise is important for the optimization and generalization of deep neural networks, but too much noise will harm the performance of networks. In our paper, we offer a new point of view that the self-attention mechanism can help to regulate the noise by enhancing instance-specific information to obtain a better regularization effect. Therefore, we propose an attention-based BN called Instance Enhancement Batch Normalization (IEBN) that recalibrates the information of each channel by a simple linear transformation. IEBN has a good capacity of regulating the batch noise and stabilizing network training to improve generalization even in the presence of two kinds of noise attacks during training. Finally, IEBN outperforms BN with only a light parameter increment in image classification tasks under different network structures and benchmark datasets.