--Audio-Visual Question Answering (A VQA) is a challenging multimodal reasoning task requiring intelligent systems to answer natural language queries based on paired audio-video inputs accurately. However, existing A VQA approaches often suffer from overfitting to dataset biases, leading to poor robustness. T o address these challenges, we first introduce a novel dataset, FortisA VQA, constructed in two stages: (1) rephrasing questions in the test split of the public MUSIC-A VQA dataset and (2) introducing distribution shifts across questions. The first stage expands the test space with greater diversity, while the second enables a refined robustness evaluation across rare, frequent, and overall question distributions. Second, we introduce a robust Multimodal Audio-Visual Epistemic Network (MA VEN) that leverages a multifaceted cycle collaborative debiasing strategy to mitigate bias learning. Experimental results demonstrate that our architecture achieves state-of-the-art performance on FortisA VQA, with a notable improvement of 7.81%. Additionally, our evaluation reveals the limited robustness of existing multimodal QA methods. We also verify the plug-and-play capability of our strategy by integrating it with various baseline models across both datasets. UMANS possess the extraordinary capacity to seam-lessly integrate auditory and visual cues, effectively establishing a cohesive relationship between visual and auditory stimuli [1-3]. Jie Ma, Pinghui Wang, Jing Tao and Zhou Su are with the Ministry of Education of Key Laboratory for Intelligent Networks and Network Security, School of Cyber Science and Engineering, Xi'an Jiaotong University, Xi'an, Shaanxi 710049, China. Zhitao Gao and Jun Liu are with the Shannxi Provincial Key Laboratory of Big Data Knowledge Engineering, School of Computer Science and Technology, Xi'an Jiaotong University, Xi'an, Shaanxi 710049, China. Qi Chai is with the Information Hub, Hong Kong University of Science and Technology (Guangzhou), Guangzhou, Guangdong, 510000, China. The question in current A VQA datasets is generated by a limited set of predefined templates, which may not be in line with the real-world scenario. Our findings indicate that existing methods such as STG [6] are not robust, which may be attributed to excessive bias learning, such as memorizing statistical regularities between critical question words and answers. It requires the system to learn high-order interaction representations of the concepts encompassed with audio, video, and language modalities. As is known to us [8-10], the high-level reasoning ability of the system mainly relies on large-scale data that does not contain harmful biases or statistical regularities. However, completely avoiding the negative bias in datasets seems challenging [11] due to the inherent skewness in real-world data distributions.

Retrieval-Augmented Generation (RAG) has gained significant attention in recent years for its potential to enhance natural language understanding and generation by combining large-scale retrieval systems with generative models. RAG leverages external knowledge sources, such as documents, databases, or structured data, to improve model performance and generate more accurate and contextually relevant outputs. This survey aims to provide a comprehensive overview of RAG by examining its fundamental components, including retrieval mechanisms, generation processes, and the integration between the two. We discuss the key characteristics of RAG, such as its ability to augment generative models with dynamic external knowledge, and the challenges associated with aligning retrieved information with generative objectives. We also present a taxonomy that categorizes RAG methods, ranging from basic retrieval-augmented approaches to more advanced models incorporating multi-modal data and reasoning capabilities. Additionally, we review the evaluation benchmarks and datasets commonly used to assess RAG systems, along with a detailed exploration of its applications in fields such as question answering, summarization, and information retrieval. Finally, we highlight emerging research directions and opportunities for improving RAG systems, such as enhanced retrieval efficiency, model interpretability, and domain-specific adaptations. This paper concludes by outlining the prospects for RAG in addressing real-world challenges and its potential to drive further advancements in natural language processing.

Recent top-performing temporal 3D detectors based on Lidars have increasingly adopted region-based paradigms. They first generate coarse proposals, followed by encoding and fusing regional features. However, indiscriminate sampling and fusion often overlook the varying contributions of individual points and lead to exponentially increased complexity as the number of input frames grows. Moreover, arbitrary result-level concatenation limits the global information extraction. In this paper, we propose a Focal Token Acquring-and-Scaling Transformer (FASTer), which dynamically selects focal tokens and condenses token sequences in an adaptive and lightweight manner. Emphasizing the contribution of individual tokens, we propose a simple but effective Adaptive Scaling mechanism to capture geometric contexts while sifting out focal points. Adaptively storing and processing only focal points in historical frames dramatically reduces the overall complexity. Furthermore, a novel Grouped Hierarchical Fusion strategy is proposed, progressively performing sequence scaling and Intra-Group Fusion operations to facilitate the exchange of global spatial and temporal information. Experiments on the Waymo Open Dataset demonstrate that our FASTer significantly outperforms other state-of-the-art detectors in both performance and efficiency while also exhibiting improved flexibility and robustness. The code is available at https://github.com/MSunDYY/FASTer.git.

The safety alignment ability of Vision-Language Models (VLMs) is prone to be degraded by the integration of the vision module compared to its LLM backbone. We investigate this phenomenon, dubbed as ''safety alignment degradation'' in this paper, and show that the challenge arises from the representation gap that emerges when introducing vision modality to VLMs. In particular, we show that the representations of multi-modal inputs shift away from that of text-only inputs which represent the distribution that the LLM backbone is optimized for. At the same time, the safety alignment capabilities, initially developed within the textual embedding space, do not successfully transfer to this new multi-modal representation space. To reduce safety alignment degradation, we introduce Cross-Modality Representation Manipulation (CMRM), an inference time representation intervention method for recovering the safety alignment ability that is inherent in the LLM backbone of VLMs, while simultaneously preserving the functional capabilities of VLMs. The empirical results show that our framework significantly recovers the alignment ability that is inherited from the LLM backbone with minimal impact on the fluency and linguistic capabilities of pre-trained VLMs even without additional training. Specifically, the unsafe rate of LLaVA-7B on multi-modal input can be reduced from 61.53% to as low as 3.15% with only inference-time intervention. WARNING: This paper contains examples of toxic or harmful language.

The widespread deployment of large language models (LLMs) has led to impressive advancements, yet information about their training data, a critical factor in their performance, remains undisclosed. Membership inference attacks (MIAs) aim to determine whether a specific instance was part of a target model's training data. However, applying MIAs to LLMs presents unique challenges due to the massive scale of pre-training data and the ambiguous nature of membership. Additionally, creating appropriate benchmarks to evaluate MIA methods is not straightforward, as training and test data distributions are often unknown. In this paper, we introduce EM-MIA, a novel MIA method for LLMs that iteratively refines membership scores and prefix scores via an expectation-maximization algorithm, leveraging the duality that the estimates of these scores can be improved by each other. Membership scores and prefix scores assess how each instance is likely to be a member and discriminative as a prefix, respectively. Our method achieves state-of-the-art results on the WikiMIA dataset. To further evaluate EM-MIA, we present OLMoMIA, a benchmark built from OLMo resources, which allows us to control the difficulty of MIA tasks with varying degrees of overlap between training and test data distributions. We believe that EM-MIA serves as a robust MIA method for LLMs and that OLMoMIA provides a valuable resource for comprehensively evaluating MIA approaches, thereby driving future research in this critical area. Large language models (LLMs) (Brown et al., 2020; Touvron et al., 2023b) have recently emerged as a groundbreaking development and have had a transformative impact in many fields.

As large language models (LLMs) become increasingly versatile, numerous large scale benchmarks have been developed to thoroughly assess their capabilities. These benchmarks typically consist of diverse datasets and prompts to evaluate different aspects of LLM performance. However, comprehensive evaluations on hundreds or thousands of prompts incur tremendous costs in terms of computation, money, and time. In this work, we investigate strategies to improve evaluation efficiency by selecting a subset of examples from each benchmark using a learned policy. Our approach models the dependencies across test examples, allowing accurate prediction of the evaluation outcomes for the remaining examples based on the outcomes of the selected ones. Consequently, we only need to acquire the actual evaluation outcomes for the selected subset. We rigorously explore various subset selection policies and introduce a novel RL-based policy that leverages the captured dependencies. Empirical results demonstrate that our approach significantly reduces the number of evaluation prompts required while maintaining accurate performance estimates compared to previous methods.

Physical neural networks (PNNs) are emerging paradigms for neural network acceleration due to their high-bandwidth, in-propagation analogue processing. Despite the advantages of PNN for inference, training remains a challenge. The imperfect information of the physical transformation means the failure of conventional gradient-based updates from backpropagation (BP). Here, we present the asymmetrical training (AT) method, which treats the PNN structure as a grey box. AT performs training while only knowing the last layer output and neuron topological connectivity of a deep neural network structure, not requiring information about the physical control-transformation mapping. We experimentally demonstrated the AT method on deep grey-box PNNs implemented by uncalibrated photonic integrated circuits (PICs), improving the classification accuracy of Iris flower and modified MNIST hand-written digits from random guessing to near theoretical maximum. We also showcased the consistently enhanced performance of AT over BP for different datasets, including MNIST, fashion-MNIST, and Kuzushiji-MNIST. The AT method demonstrated successful training with minimal hardware overhead and reduced computational overhead, serving as a robust light-weight training alternative to fully explore the advantages of physical computation.

The primary challenge in Federated Learning (FL) is to model non-IID distributions across clients, whose fine-grained structure is important to improve knowledge sharing. For example, some knowledge is globally shared across all clients, some is only transferable within a subgroup of clients, and some are client-specific. To capture and exploit this structure, we train models organized in a multi-level structure, called ``Multi-level Additive Models (MAM)'', for better knowledge-sharing across heterogeneous clients and their personalization. In federated MAM (FeMAM), each client is assigned to at most one model per level and its personalized prediction sums up the outputs of models assigned to it across all levels. For the top level, FeMAM trains one global model shared by all clients as FedAvg. For every mid-level, it learns multiple models each assigned to a subgroup of clients, as clustered FL. Every bottom-level model is trained for one client only. In the training objective, each model aims to minimize the residual of the additive predictions by the other models assigned to each client. To approximate the arbitrary structure of non-IID across clients, FeMAM introduces more flexibility and adaptivity to FL by incrementally adding new models to the prediction of each client and reassigning another if necessary, automatically optimizing the knowledge-sharing structure. Extensive experiments show that FeMAM surpasses existing clustered FL and personalized FL methods in various non-IID settings. Our code is available at https://github.com/shutong043/FeMAM.

Remote sensing image change captioning (RSICC) aims at generating human-like language to describe the semantic changes between bi-temporal remote sensing image pairs. It provides valuable insights into environmental dynamics and land management. Unlike conventional change captioning task, RSICC involves not only retrieving relevant information across different modalities and generating fluent captions, but also mitigating the impact of pixel-level differences on terrain change localization. The pixel problem due to long time span decreases the accuracy of generated caption. Inspired by the remarkable generative power of diffusion model, we propose a probabilistic diffusion model for RSICC to solve the aforementioned problems. In training process, we construct a noise predictor conditioned on cross modal features to learn the distribution from the real caption distribution to the standard Gaussian distribution under the Markov chain. Meanwhile, a cross-mode fusion and a stacking self-attention module are designed for noise predictor in the reverse process. In testing phase, the well-trained noise predictor helps to estimate the mean value of the distribution and generate change captions step by step. Extensive experiments on the LEVIR-CC dataset demonstrate the effectiveness of our Diffusion-RSCC and its individual components. The quantitative results showcase superior performance over existing methods across both traditional and newly augmented metrics. The code and materials will be available online at https://github.com/Fay-Y/Diffusion-RSCC.

Many of the recent capabilities demonstrated by Large Language Models (LLMs) arise primarily from their ability to exploit contextual information. In this paper, we explore ways to improve reasoning capabilities of LLMs through (1) exploration of different chains of thought and (2) validation of the individual steps of the reasoning process. We propose three general principles that a model should adhere to while reasoning: (i) Relevance, (ii) Mathematical Accuracy, and (iii) Logical Consistency. We apply these constraints to the reasoning steps generated by the LLM to improve the accuracy of the final generation. The constraints are applied in the form of verifiers: the model itself is asked to verify if the generated steps satisfy each constraint. To further steer the generations towards high-quality solutions, we use the perplexity of the reasoning steps as an additional verifier. We evaluate our method on 4 distinct types of reasoning tasks, spanning a total of 9 different datasets. Experiments show that our method is always better than vanilla generation, and, in 6 out of the 9 datasets, it is better than best-of N sampling which samples N reasoning chains and picks the lowest perplexity generation.