AAAI 2025 Papers

Existing studies explore the explainability of Grammatical Error Correction (GEC) in a limited scenario, where they ignore the interaction between corrections and explanations and have not established a corresponding comprehensive benchmark. To bridge the gap, this paper first introduces the task of EXplainable GEC (EXGEC), which focuses on the integral role of correction and explanation tasks. To facilitate the task, we propose EXCGEC, a tailored benchmark for Chinese EXGEC consisting of 8,216 explanation-augmented samples featuring the design of hybrid edit-wise explanations. We then benchmark several series of LLMs in multi-task learning settings, including post-explaining and pre-explaining. To promote the development of the task, we also build a comprehensive evaluation suite by leveraging existing automatic metrics and conducting human evaluation experiments to demonstrate the human consistency of the automatic metrics for free-text explanations. Our experiments reveal the effectiveness of evaluating free-text explanations using traditional metrics like METEOR and ROUGE, and the inferior performance of multi-task models compared to the pipeline solution, indicating its challenges to establish positive effects in learning both tasks.

Weather forecasting is a crucial task for meteorologic research, with direct social and economic impacts. Recently, data-driven weather forecasting models based on deep learning have shown great potential, achieving superior performance compared with traditional numerical weather prediction methods. However, these models often require massive training data and computational resources. In this paper, we propose EWMoE, an effective model for accurate global weather forecasting, which requires significantly less training data and computational resources. Our model incorporates three key components to enhance prediction accuracy: 3D absolute position embedding, a core Mixture-of-Experts (MoE) layer, and two specific loss functions. We conduct our evaluation on the ERA5 dataset using only two years of training data. Extensive experiments demonstrate that EWMoE outperforms current models such as FourCastNet and ClimaX at all forecast time, achieving competitive performance compared with the state-of-the-art models Pangu-Weather and GraphCast in evaluation metrics such as Anomaly Correlation Coefficient (ACC) and Root Mean Square Error (RMSE). Additionally, ablation studies indicate that applying the MoE architecture to weather forecasting offers significant advantages in improving accuracy and resource efficiency.

Conversational Recommender Systems (CRS) aim to provide tailored recommendation responses via a chat interface, including both the user's preferred item and its accompanying explanation. However, due to its generative nature, CRS are prone to responding with factually incorrect explanations (i.e., hallucinations). To solve this problem, we propose incorporating a passage retrieval module into CRS with the objective of enhancing the factuality and informativeness of system responses. Specifically, we outline essential directions for employing a passage retrieval module in CRS to address the following critical issues: (1) the risk of passage retrieval not aligning with the user preference; (2) the absence of supervision for training a passage retrieval module. As a solution, we introduce ESPRESSO, a novel passage retrieval approach for CRS, to effectively tackle the above issues with two core ideas: adaptive item selection and relevance-based groupwise learning. Our extensive experiments show that ESPRESSO effectively resolves issues, achieving up to 36% higher Hit@3 accuracy than the best of 8 competing methods. Additionally, we verify that leveraging passages retrieved by ESPRESSO significantly improves the response quality of CRS.

Event-based semantic segmentation (ESS) has attracted researchers' attention recently, as event cameras can solve problems such as under/over-exposure or motion blur that are difficult for RGB cameras to handle. However, event data are noisy and sparse, resulting in difficulties for the model to locate and extract reliable cues from their sparse representations, especially when performing pixel-level tasks. In this paper, we propose a novel framework ESEG to alleviate the dilemma. Given that event signals relate closely to moving edges, instead of proposing complex structures to expect them to recognize those reliable edge regions behind event signals on their own, we introduce the explicit edge-semantic supervision as a reference to let the ESS model globally optimize semantics, considering the high confidence of event data in edge regions. In addition, we propose a fusion module named Density-Aware Dynamic-Window Cross Attention Fusion (D\textsuperscript{2}CAF), in which the density perception, cross-attention, and dynamic window masking mechanisms are jointly imposed to optimize edge-dense feature fusion, leveraging the characteristics of event cameras. Experimental results on DSEC and DDD17 datasets demonstrate the efficacy of the ESEG framework and its core designs.

Solving jigsaw puzzles has been extensively studied. While most existing models focus on solving either small-scale puzzles or puzzles with no gap between fragments, solving large-scale puzzles with gaps presents distinctive challenges in both image understanding and combinatorial optimization. To tackle these challenges, we propose a framework of Evolutionary Reinforcement Learning with Multi-head Puzzle Perception (ERL-MPP) to derive a better set of swapping actions for solving the puzzles. Specifically, to tackle the challenges of perceiving the puzzle with gaps, a Multi-head Puzzle Perception Network (MPPN) with a shared encoder is designed, where multiple puzzlet heads comprehensively perceive the local assembly status, and a discriminator head provides a global assessment of the puzzle. To explore the large swapping action space efficiently, an Evolutionary Reinforcement Learning (EvoRL) agent is designed, where an actor recommends a set of suitable swapping actions from a large action space based on the perceived puzzle status, a critic updates the actor using the estimated rewards and the puzzle status, and an evaluator coupled with evolutionary strategies evolves the actions aligning with the historical assembly experience. The proposed ERL-MPP is comprehensively evaluated on the JPLEG-5 dataset with large gaps and the MIT dataset with large-scale puzzles. It significantly outperforms all state-of-the-art models on both datasets.

We propose ERFSL, an efficient reward function searcher using large language models (LLMs) for custom-environment, multi-objective reinforcement learning (RL). ERFSL generates reward components based on explicit user requirements and rectifies them, and iteratively optimizes the weights of these components based on textual context. Applied to an underwater data collection RL task, ERFSL corrects reward codes with only one feedback iteration per requirement, and acquires diverse reward functions within the Pareto set. ERFSL also presents robust capability for deviated weights and small-size LLMs such as GPT-4o mini. The full-text prompts, examples of LLM-generated answers, and source code are available at https://360zmem.github.io/LLMRsearcher/ .

As climate change reshapes global weather patterns, the increasing frequency and intensity of extreme rainfall events have amplified the safety imperatives for autonomous driving systems. During such events, rainfall can escalate from heavy to violent, as defined by the World Meteorological Organization, severely impairing images with diverse and significant degradations. Many existing semantic segmentation models perform well under light to heavy rain, but there is a notable absence of datasets addressing violent rain conditions for these models to validate and learn from. In this paper, we introduce the Extreme RainFall (ERF) dataset for semantic segmentation in both image and video tasks under violent rain conditions. Our dataset comprises 14,757 unlabeled frames and 100 labeled frames, all captured during four different violent rainfall periods. We use our dataset to evaluate the robustness of various methods against violent rainfall, focusing on four approaches: 1) image-based foundation models, 2) image-based domain generalization methods, 3) image-based domain adaptation methods, and 4) video-based methods. The results reveal that none of the existing models tested is capable of withstanding the extreme challenges posed by violent rainfall conditions. By analyzing the results, we offer insights and suggestions for developing more robust models under extreme rainfall events.

Deep reinforcement learning (DRL) has gained significant attention in autonomous systems, yet its black-box nature and lack of explainability hinder user trust in safety-critical domains such as autonomous driving. Existing experience replay approaches enhance sample efficiency but often fail to capture the internal causality of training data, leading to a convoluted training process that is difficult for humans to explain. In this work, we introduce Experience Replay with Causal Inference (ERCI), an explainable approach that integrates time series representation and causal inference to offer human-aligned explanations for DRL. Specifically, ERCI 1) introduces a novel multivariate time series representation to extract explainable Time Series Causal Factors (TSCF) from experimental data and 2) leverages internal causality in TSCFs with causal inference as a crucial standard for experience replay in DRL training. We evaluate ERCI using multiple baseline algorithms across diverse environments. Results show that ERCI provides human-aligned explanations and further improves sample efficiency through enhanced explainability. Notably, ERCI outperforms other state-of-the-art approaches by 15% in average performance, highlighting its effectiveness and generalizability.

Prompt tuning is a promising method to fine-tune a pre-trained language model without retraining its large-scale parameters. Instead, it attaches a soft prompt to the input text, whereby downstream tasks can be well adapted by merely learning the embeddings of prompt tokens. Nevertheless, existing methods still suffer from two challenges: (i) they are hard to balance accuracy and efficiency. A longer (shorter) soft prompt generally leads to a better (worse) accuracy but at the cost of more (less) training time. (ii) The performance may not be consistent when adapting to different downstream tasks. We attribute it to the same embedding space but responsible for different requirements of downstream tasks. To address these issues, we propose an Efficient Prompt Tuning method (EPT) by multi-space projection and prompt fusion. Specifically, it decomposes a given soft prompt into a shorter prompt and two low-rank matrices, significantly reducing the training time. Accuracy is also enhanced by leveraging low-rank matrices and the short prompt as additional knowledge sources to enrich the semantics of the original short prompt. In addition, we project the soft prompt into multiple subspaces to improve the performance consistency, and then adaptively learn the combination weights of different spaces through a gating network. Experiments on 13 natural language processing downstream tasks show that our method significantly and consistently outperforms 11 comparison methods with the relative percentage of improvements up to 12.9%, and training time decreased by 14%.

Due to the remarkable reasoning ability, Large language models (LLMs) have demonstrated impressive performance in knowledge graph question answering (KGQA) tasks, which find answers to natural language questions over knowledge graphs (KGs). To alleviate the hallucinations and lack of knowledge issues of LLMs, existing methods often retrieve the question-related information from KGs to enrich the input context. However, most methods focus on retrieving the relevant information while ignoring the importance of different types of knowledge in reasoning, which degrades their performance. To this end, this paper reformulates the KGQA problem as a graphical model and proposes a three-stage framework named the Evidence Path Enhanced Reasoning Model (EPERM) for KGQA. In the first stage, EPERM uses the fine-tuned LLM to retrieve a subgraph related to the question from the original knowledge graph. In the second stage, EPERM filters out the evidence paths that faithfully support the reasoning of the questions, and score their importance in reasoning. Finally, EPERM uses the weighted evidence paths to reason the final answer. Since considering the importance of different structural information in KGs for reasoning, EPERM can improve the reasoning ability of LLMs in KGQA tasks. Extensive experiments on benchmark datasets demonstrate that EPERM achieves superior performances in KGQA tasks.

Open-vocabulary panoptic segmentation aims to segment and classify everything in diverse scenes across an unbounded vocabulary. Existing methods typically employ two-stage or single-stage framework. The two-stage framework involves cropping the image multiple times using masks generated by a mask generator, followed by feature extraction, while the single-stage framework relies on a heavyweight mask decoder to make up for the lack of spatial position information through self-attention and cross-attention in multiple stacked Transformer blocks. Both methods incur substantial computational overhead, thereby hindering the efficiency of model inference. To fill the gap in efficiency, we propose EOV-Seg, a novel single-stage, shared, efficient, and spatialaware framework designed for open-vocabulary panoptic segmentation. Specifically, EOV-Seg innovates in two aspects. First, a Vocabulary-Aware Selection (VAS) module is proposed to improve the semantic comprehension of visual aggregated features and alleviate the feature interaction burden on the mask decoder. Second, we introduce a Two-way Dynamic Embedding Experts (TDEE), which efficiently utilizes the spatial awareness capabilities of ViT-based CLIP backbone. To the best of our knowledge, EOV-Seg is the first open-vocabulary panoptic segmentation framework towards efficiency, which runs faster and achieves competitive performance compared with state-of-the-art methods. Specifically, with COCO training only, EOV-Seg achieves 24.5 PQ, 32.1 mIoU, and 11.6 FPS on the ADE20K dataset and the inference time of EOV-Seg is 4-19 times faster than state-of-the-art methods. Especially, equipped with ResNet50 backbone, EOV-Seg runs 23.8 FPS with only 71M parameters on a single RTX 3090 GPU.

DNN-based watermarking methods have rapidly advanced, with the ``Encoder-Noise Layer-Decoder'' (END) framework being the most widely used. To ensure end-to-end training, the noise layer in the framework must be differentiable. However, real-world distortions are often non-differentiable, leading to challenges in end-to-end training. Existing solutions only treat the distortion perturbation as additive noise, which does not fully integrate the effect of distortion in training. To better incorporate non-differentiable distortions into training, we propose a novel dual-decoder architecture (END^2). Unlike conventional END architecture, our method employs two structurally identical decoders: the Teacher Decoder, processing pure watermarked images, and the Student Decoder, handling distortion-perturbed images. The gradient is backpropagated only through the Teacher Decoder branch to optimize the encoder thus bypassing the problem of non-differentiability. To ensure resistance to arbitrary distortions, we enforce alignment of the two decoders' feature representations by maximizing the cosine similarity between their intermediate vectors on a hypersphere. Extensive experiments demonstrate that our scheme outperforms state-of-the-art algorithms under various non-differentiable distortions. Moreover, even without the differentiability constraint, our method surpasses baselines with a differentiable noise layer. Our approach is effective and easily implementable across all END architectures, enhancing practicality and generalizability.

The objective of Composed Image Retrieval (CIR) is to identify a target image that meets the requirement based on a multimodal query (including the reference image and the modification text) provided by the user. Despite the notable success of existing approaches, they fail to adequately address the modification relation between visual entities and modification actions. This limitation is non-trivial due to three challenges: 1) irrelevant factor perturbation, 2) vague semantic boundaries, and 3) implicit modification relations. To address the above challenges, we propose an Entity miNing and modifiCation relatiOn binDing nEtwoRk (ENCODER), which has been designed to mine visual entities and modification actions, and then bind modification relations. Among the various components of the proposed ENCODER, we have initially designed the Latent Factor Filter (LFF) module to filter visual and textual latent factors related to modification semantics based on a threshold gating mechanism. Secondly, we propose Entity-Action Binding (EAB), which comprises modality-shared Learnable Relation Queries (LRQ) that are capable of mining visual entities and modification actions, as well as learning implicit modification relations for entity-action binding. Finally, the Multi-scale Composition module is introduced to achieve multi-scale feature composition, with guidance provided by entity-action binding. Extensive experiments on four benchmark datasets demonstrate the superiority of our proposed method.

Urban change is a constant process that influences the perception of neighbourhoods and the lives of the people within them. The field of Urban Scene Change Detection (USCD) aims to capture changes in street scenes using computer vision and can help raise awareness of changes that make it possible to better understand the city and its residents. Traditionally, the field of USCD has used supervised methods with small scale datasets. This constrains methods when applied to new cities, as it requires labour-intensive labeling processes and forces a priori definitions of relevant change. In this paper we introduce AC-1M the largest USCD dataset by far of over 1.1M images, together with EMPLACE, a self-supervising method to train a Vision Transformer using our adaptive triplet loss. We show EMPLACE outperforms SOTA methods both as a pre-training method for linear fine-tuning as well as a zero-shot setting. Lastly, in a case study of Amsterdam, we show that we are able to detect both small and large changes throughout the city and that changes uncovered by EMPLACE, depending on size, correlate with housing prices - which in turn is indicative of inequity.

Egocentric human pose estimation (HPE) using wearable sensors is essential for VR/AR applications. Most methods rely solely on either egocentric-view images or sparse Inertial Measurement Unit (IMU) signals, leading to inaccuracies due to self-occlusion in images or the sparseness and drift of inertial sensors. Most importantly, the lack of real-world datasets containing both modalities is a major obstacle to progress in this field. To overcome the barrier, we propose EMHI, a multimodal Egocentric human Motion dataset with Head-Mounted Display (HMD) and body-worn IMUs, with all data collected under the real VR product suite. Specifically, EMHI provides synchronized stereo images from downward-sloping cameras on the headset and IMU data from body-worn sensors, along with pose annotations in SMPL format. This dataset consists of 885 sequences captured by 58 subjects performing 39 actions, totaling about 28.5 hours of recording. We evaluate the annotations by comparing them with optical marker-based SMPL fitting results. To substantiate the reliability of our dataset, we introduce MEPoser, a new baseline method for multimodal egocentric HPE, which employs a multimodal fusion encoder, temporal feature encoder, and MLP-based regression heads. The experiments on EMHI show that MEPoser outperforms existing single-modal methods and demonstrates the value of our dataset in solving the problem of egocentric HPE. We believe the release of EMHI and the method could advance the research of egocentric HPE and expedite the practical implementation of this technology in VR/AR products.

Recent advances in diffusion models focus on efficiently handling conditional generative tasks without extra training. The process involves decomposing the result into two components: 1. unconditional sample, generated in the absence of conditions; 2. condition correction, adjusting unconditional sample to include the guidance image. This adjustment is quantified by the pixel-level measure, where the latent is decoded back into a pixel image, and the forward operator translates the noisy image into the guidance domain for comparison with the guidance image. To enhance the fidelity of condition correction, we propose a learnable latent forward operator, focusing on latent-space consistency with the expectation that this latent-space consistency approximates the pixel-level fidelity measure. The encoder translates the guidance image into the latent space, and a correctional operator is proposed to rectify model mismatching in the latent guidance model. The determination of the condition term and the correction estimation is akin to solving a blind inverse problem. Our EMControl employs the Expectation-Maximization (EM) algorithm to solve the blind inverse problem during the reverse sampling process. This technique ensures that samples, once consistent with the guidance, are accurately mapped back onto the noisy data manifold, adhering to the data's inherent distribution. The EMControl has proven its effectiveness by delivering superior performance in conditional diffusion generation tasks compared to previous approaches. Moreover, its application to multiple-condition scenarios underscores its versatility and robustness across a range of generative tasks.

The language model (LM) approach based on acoustic and linguistic prompts, such as VALL-E, has achieved remarkable progress in the field of zero-shot audio generation. However, existing methods still have some limitations: 1) repetitions, transpositions, and omissions in the output synthesized speech due to limited alignment constraints between audio and phoneme tokens; 2) challenges of fine-grained control over the synthesized speech with autoregressive (AR) language model; 3) infinite silence generation due to the nature of AR-based decoding, especially under the greedy strategy. To alleviate these issues, we propose ELLA-V, a simple but efficient LM-based zero-shot text-to-speech (TTS) framework, which enables fine-grained control over synthesized audio at the phoneme level. The key to ELLA-V is interleaving sequences of acoustic and phoneme tokens, where phoneme tokens appear ahead of the corresponding acoustic tokens. The experimental findings reveal that our model outperforms baselines in terms of accuracy and delivers more stable results using both greedy and sampling-based decoding strategies.

Large language models (LLMs) require model editing to efficiently update specific knowledge within them and avoid factual errors. Most model editing methods are solely designed for single-time use and result in a significant forgetting effect in lifelong editing scenarios, where sequential edits are conducted over time. Previous approaches manage sequential edits by freezing original parameters and discretely allocating new parameters for each knowledge update. However, these methods lack robustness to minor input variations due to the discrete mapping between data and parameters. To overcome this challenge, we propose ELDER, a novel approach to create a continuous association between data and adapters. ELDER integrates multiple LoRAs through a router network and is trained to establish a smooth data-adapter association, thereby enhancing the edit robustness and generalization of semantically equivalent inputs. To ensure inputs containing the same knowledge will be processed by the same LoRAs, we design a novel loss to guide the model link LoRA allocations with edit knowledge. Furthermore, we propose a deferral mechanism to retain the original LLM capabilities post-edit. Extensive experiments on GPT-2 XL and LLaMA2-7B demonstrate that ELDER effectively edits models in the lifelong setting, outperforming eight baselines while exhibiting strong scalability and preserving LLMs' general abilities on downstream tasks.

3D Gaussian Splatting (3D GS) has gained popularity due to its faster rendering speed and high-quality novel view synthesis. Some researchers have explored using 3D GS for reconstructing driving scenes. However, these methods often rely on various types of data, such as depth maps, 3D bounding boxes, and trajectories of moving objects. Additionally, the lack of annotations for synthesized images limits their direct application in downstream tasks. To address these issues, we propose EGSRAL, a 3D GS-based method that relies solely on training images without extra annotations. EGSRAL enhances 3D GS's capability to model both dynamic objects and static backgrounds and introduces a novel adaptor for auto labeling, generating corresponding annotations based on existing annotations. We also propose a grouping strategy for vanilla 3D GS to address perspective issues in rendering large-scale, complex scenes. Our method achieves state-of-the-art performance on multiple datasets without any extra annotation. For example, the PSNR metric reaches 29.04 on the nuScenes dataset. Moreover, our automated labeling can significantly improve the performance of 2D/3D detection tasks.