Jingyu Wu

Peng Ding, Jingyu Wu, Jun Kuang et al.

Multi-modal Large Language Models (MLLMs) have demonstrated remarkable performance on various visual-language understanding and generation tasks. However, MLLMs occasionally generate content inconsistent with the given images, which is known as "hallucination". Prior works primarily center on evaluating hallucination using standard, unperturbed benchmarks, which overlook the prevalent occurrence of perturbed inputs in real-world scenarios-such as image cropping or blurring-that are critical for a comprehensive assessment of MLLMs' hallucination. In this paper, to bridge this gap, we propose Hallu-PI, the first benchmark designed to evaluate Hallucination in MLLMs within Perturbed Inputs. Specifically, Hallu-PI consists of seven perturbed scenarios, containing 1,260 perturbed images from 11 object types. Each image is accompanied by detailed annotations, which include fine-grained hallucination types, such as existence, attribute, and relation. We equip these annotations with a rich set of questions, making Hallu-PI suitable for both discriminative and generative tasks. Extensive experiments on 12 mainstream MLLMs, such as GPT-4V and Gemini-Pro Vision, demonstrate that these models exhibit significant hallucinations on Hallu-PI, which is not observed in unperturbed scenarios. Furthermore, our research reveals a severe bias in MLLMs' ability to handle different types of hallucinations. We also design two baselines specifically for perturbed scenarios, namely Perturbed-Reminder and Perturbed-ICL. We hope that our study will bring researchers' attention to the limitations of MLLMs when dealing with perturbed inputs, and spur further investigations to address this issue. Our code and datasets are publicly available at https://github.com/NJUNLP/Hallu-PI.

Jing Dong, Jingyu Wu, Siwei Wang et al.

The congestion game is a powerful model that encompasses a range of engineering systems such as traffic networks and resource allocation. It describes the behavior of a group of agents who share a common set of $F$ facilities and take actions as subsets with $k$ facilities. In this work, we study the online formulation of congestion games, where agents participate in the game repeatedly and observe feedback with randomness. We propose CongestEXP, a decentralized algorithm that applies the classic exponential weights method. By maintaining weights on the facility level, the regret bound of CongestEXP avoids the exponential dependence on the size of possible facility sets, i.e., $\binom{F}{k} \approx F^k$, and scales only linearly with $F$. Specifically, we show that CongestEXP attains a regret upper bound of $O(kF\sqrt{T})$ for every individual player, where $T$ is the time horizon. On the other hand, exploiting the exponential growth of weights enables CongestEXP to achieve a fast convergence rate. If a strict Nash equilibrium exists, we show that CongestEXP can converge to the strict Nash policy almost exponentially fast in $O(F\exp(-t^{1-α}))$, where $t$ is the number of iterations and $α\in (1/2, 1)$.

Xuezhen Tu, Jingyu Wu, Fangyu Kang et al.

Spatio-Temporal Video Grounding requires jointly localizing target objects across both temporal and spatial dimensions based on natural language queries, posing fundamental challenges for existing Multimodal Large Language Models (MLLMs). We identify two core challenges: \textit{entangled spatio-temporal alignment}, arising from coupling two heterogeneous sub-tasks within the same autoregressive output space, and \textit{dual-domain visual token redundancy}, where target objects exhibit simultaneous temporal and spatial sparsity, rendering the overwhelming majority of visual tokens irrelevant to the grounding query. To address these, we propose \textbf{Bridge-STG}, an end-to-end framework that decouples temporal and spatial localization while maintaining semantic coherence. While decoupling is the natural solution to this entanglement, it risks creating a semantic gap between the temporal MLLM and the spatial decoder. Bridge-STG resolves this through two pivotal designs: the \textbf{Spatio-Temporal Semantic Bridging (STSB)} mechanism with Explicit Temporal Alignment (ETA) distills the MLLM's temporal reasoning context into enriched bridging queries as a robust semantic interface; and the \textbf{Query-Guided Spatial Localization (QGSL)} module leverages these queries to drive a purpose-built spatial decoder with multi-layer interactive queries and positive/negative frame sampling, jointly eliminating dual-domain visual token redundancy. Extensive experiments across multiple benchmarks demonstrate that Bridge-STG achieves state-of-the-art performance among MLLM-based methods. Bridge-STG improves average m\_vIoU from $26.4$ to $34.3$ on VidSTG and demonstrates strong cross-task transfer across various fine-grained video understanding tasks under a unified multi-task training regime.

Saadat Hasan Khan, Spencer Hong, Jingyu Wu et al.

Retrieval-augmented generation (RAG) is a common technique for grounding language model outputs in domain-specific information. However, RAG is often challenged by reasoning-intensive question-answering (QA), since common retrieval methods like cosine similarity maximize relevance at the cost of introducing redundant content, which can reduce information recall. To address this, we introduce Diversity-Focused Retrieval-Augmented Generation (DF-RAG), which systematically incorporates diversity into the retrieval step to improve performance on complex, reasoning-intensive QA benchmarks. DF-RAG builds upon the Maximal Marginal Relevance framework to select information chunks that are both relevant to the query and maximally dissimilar from each other. A key innovation of DF-RAG is its ability to optimize the level of diversity for each query dynamically at test time without requiring any additional fine-tuning or prior information. We show that DF-RAG improves F1 performance on reasoning-intensive QA benchmarks by 4-10 percent over vanilla RAG using cosine similarity and also outperforms other established baselines. Furthermore, we estimate an Oracle ceiling of up to 18 percent absolute F1 gains over vanilla RAG, of which DF-RAG captures up to 91.3 percent.

Hongyi Cai, Mohammad Mahdinur Rahman, Mohammad Shahid Akhtar et al.

Image Transformers show a magnificent success in Image Restoration tasks. Nevertheless, most of transformer-based models are strictly bounded by exorbitant memory occupancy. Our goal is to reduce the memory consumption of Swin Transformer and at the same time speed up the model during training process. Thus, we introduce AgileIR, group shifted attention mechanism along with window attention, which sparsely simplifies the model in architecture. We propose Group Shifted Window Attention (GSWA) to decompose Shift Window Multi-head Self Attention (SW-MSA) and Window Multi-head Self Attention (W-MSA) into groups across their attention heads, contributing to shrinking memory usage in back propagation. In addition to that, we keep shifted window masking and its shifted learnable biases during training, in order to induce the model interacting across windows within the channel. We also re-allocate projection parameters to accelerate attention matrix calculation, which we found a negligible decrease in performance. As a result of experiment, compared with our baseline SwinIR and other efficient quantization models, AgileIR keeps the performance still at 32.20 dB on Set5 evaluation dataset, exceeding other methods with tailor-made efficient methods and saves over 50% memory while a large batch size is employed.

Jingyu Wu, Aditya Shrivastava, Jing Zhu et al.

Efficiently reranking documents retrieved from information retrieval (IR) pipelines to enhance overall quality of Retrieval-Augmented Generation (RAG) system remains an important yet challenging problem. Recent studies have highlighted the importance of Large Language Models (LLMs) in reranking tasks. In particular, Pairwise Reranking Prompting (PRP) has emerged as a promising plug-and-play approach due to its usability and effectiveness. However, the inherent complexity of the algorithm, coupled with the high computational demands and latency incurred due to LLMs, raises concerns about its feasibility in real-time applications. To address these challenges, this paper presents a focused study on pairwise reranking, demonstrating that carefully applied optimization methods can significantly mitigate these issues. By implementing these methods, we achieve a remarkable latency reduction of up to 166 times, from 61.36 seconds to 0.37 seconds per query, with an insignificant drop in performance measured by Recall@k. Our study highlights the importance of design choices that were previously overlooked, such as using smaller models, limiting the reranked set, using lower precision, reducing positional bias with one-directional order inference, and restricting output tokens. These optimizations make LLM-based reranking substantially more efficient and feasible for latency-sensitive, real-world deployments.

Pan Peng, Jingyu Wu

We study quantum property testing for directed graphs with maximum in-degree and out-degree bounded by some universal constant $d$. For a proximity parameter $\varepsilon$, we show that any property that can be tested with $O_{\varepsilon,d}(1)$ queries in the classical bidirectional model, where both incoming and outgoing edges are accessible, can also be tested in the quantum unidirectional model, where only outgoing edges are accessible, using $n^{1/2 - Ω_{\varepsilon,d}(1)}$ queries. This yields an almost quadratic quantum speedup over the best known classical algorithms in the unidirectional model. Moreover, we prove that our transformation is almost tight by giving an explicit property $P_\varepsilon$ that is $\varepsilon$-testable within $O_\varepsilon(1)$ classical queries in the bidirectional model, but requires $\widetildeΩ(n^{1/2-f'(\varepsilon)})$ quantum queries in the unidirectional model, where $f'(\varepsilon)$ is a function that approaches $0$ as $\varepsilon$ approaches $0$. As a byproduct, we show that in the unidirectional model, the number of occurrences of any constant-size subgraph $H$ can be approximated up to additive error $δn$ using $o(\sqrt{n})$ quantum queries.

Hong Gao, Jingyu Wu, Xiangkai Xu et al.

Spatio-Temporal Video Grounding (STVG) aims to localize target objects in videos based on natural language descriptions. Despite recent advances in Multimodal Large Language Models, a significant gap remains between current models and real-world demands involving diverse objects and complex queries. We attribute this to limited benchmark scope, causing models to exhibit category bias, oversimplified reasoning, and poor linguistic robustness. To address these limitations, we introduce OmniGround, a comprehensive benchmark with 3,475 videos spanning 81 categories and complex real-world queries. We propose the Forward-Backward-Refinement annotation pipeline that combines multi-directional tracking with intelligent error correction for high-quality labels. We further introduce DeepSTG, a systematic evaluation framework quantifying dataset quality across four complementary dimensions beyond superficial statistics. Evaluations reveal performance average drop of 10.4% on complex real-world scenes, particularly with small/occluded objects and intricate spatial relations. Motivated by these, we propose PG-TAF, a training-free two-stage framework decomposing STVG into high-level temporal grounding and fine-grained spatio-temporal propagation. Experiments demonstrate PG-TAF achieves 25.6% and 35.6% improvements in m\_tIoU and m\_vIoU on OmniGround with consistent gains across four benchmarks.

Xujun Peng, Anoop Kumar, Jingyu Wu et al.

Retrieval-Augmented Generation (RAG) systems leverage Large Language Models (LLMs) to generate accurate and reliable responses that are grounded in retrieved context. However, LLMs often generate inconsistent outputs for semantically equivalent inputs, a problem compounded by the scarcity of consistency-focused training data and the limitations of current fine-tuning techniques in enhancing output consistency. We propose a new approach combining systematic synthetic data generation, triplet loss for better embeddings, and a novel layer-wise model merging approach. Using consistency-aware weights derived from intermediate layer activations, our method effectively integrates knowledge from specialized models. Experimental results how that our merged model significantly enhances output consistency, achieving a ~47.5\% improvement in response similarity over the baseline, thus offering a practical solution for increasing the reliability of an industrial RAG system.

Chenyang Zhu, Spencer Hong, Jingyu Wu et al.

We have reached a critical roadblock in the development and enhancement of long-horizon, multi-component LLM agentic systems: it is incredibly tricky to identify where these systems break down and why. Evaluation capabilities that currently exist today (e.g., single pass LLM-as-a-judge) are limited in that they often focus on individual metrics or capabilities, end-to-end outcomes, and are narrowly grounded on the preferences of humans. We argue that to match the agentic capabilities, evaluation frameworks must also be able to reason, probe, iterate, and understand the complex logic passing through these systems over long horizons. In this paper, we present RAFFLES - an evaluation architecture that incorporates reasoning and iterative refinement. Specifically, RAFFLES operates as an iterative, multi-component pipeline, using a central Judge to systematically investigate faults and a set of specialized Evaluators to assess not only the system's components but also the quality of the reasoning by the Judge itself, thereby building a history of hypotheses. We tested RAFFLES against several baselines on the Who&When dataset, a benchmark designed to diagnose the "who" (agent) and "when" (step) of a system's failure. RAFFLES outperforms these baselines, achieving an agent-step fault pair accuracy of over 43% on the Algorithmically-Generated dataset (a substantial increase from the previously published best of 16.6%) and over 20% on the Hand-Crafted dataset (surpassing the previously published best of 8.8%). These results demonstrate a key step towards introducing automated fault detection for autonomous systems over labor-intensive manual human review.

Hongyi Cai, Mohammad Mahdinur Rahman, Wenzhen Dong et al.

Feature pyramids have been widely adopted in convolutional neural networks and transformers for tasks in medical image segmentation. However, existing models generally focus on the Encoder-side Transformer for feature extraction. We further explore the potential in improving the feature decoder with a well-designed architecture. We propose Cross Feature Pyramid Transformer decoder (CFPFormer), a novel decoder block that integrates feature pyramids and transformers. Even though transformer-like architecture impress with outstanding performance in segmentation, the concerns to reduce the redundancy and training costs still exist. Specifically, by leveraging patch embedding, cross-layer feature concatenation mechanisms, CFPFormer enhances feature extraction capabilities while complexity issue is mitigated by our Gaussian Attention. Benefiting from Transformer structure and U-shaped connections, our work is capable of capturing long-range dependencies and effectively up-sample feature maps. Experimental results are provided to evaluate CFPFormer on medical image segmentation datasets, demonstrating the efficacy and effectiveness. With a ResNet50 backbone, our method achieves 92.02\% Dice Score, highlighting the efficacy of our methods. Notably, our VGG-based model outperformed baselines with more complex ViT and Swin Transformer backbone.