Zihe Song

Shanghao Li, Jinda Han, Yibo Wang et al.

In many reasoning tasks, large language models (LLMs) rely on structured external knowledge, such as graphs and tables, which is typically linearized into sequential token representations. However, even when sufficient knowledge is available, LLMs can still produce hallucinated outputs, and the underlying mechanisms behind such failures remain poorly understood. We investigate these mechanisms and find that hallucinations arise from systematic internal dynamics rather than random noise. First, attention disproportionately concentrates toward shortcut-like structural cues rather than distributing across the full context. Second, feed-forward representations fail to ground the provided knowledge, causing the model to revert to parametric memory. Moreover, our results indicate that hallucination is consistently associated with failures in semantic grounding within feed-forward layers, while attention allocation exhibits greater task-dependent variability. Finally, we show that these mechanistic patterns generalize beyond single-hop graphs to multi-hop and tabular settings, enabling effective hallucination detection across structured knowledge formats.

Weizhi Zhang, Liangwei Yang, Zihe Song et al.

The efficiency and scalability of graph convolution networks (GCNs) in training recommender systems (RecSys) have been persistent concerns, hindering their deployment in real-world applications. This paper presents a critical examination of the necessity of graph convolutions during the training phase and introduces an innovative alternative: the Light Post-Training Graph Ordinary-Differential-Equation (LightGODE). Our investigation reveals that the benefits of GCNs are more pronounced during testing rather than training. Motivated by this, LightGODE utilizes a novel post-training graph convolution method that bypasses the computation-intensive message passing of GCNs and employs a non-parametric continuous graph ordinary-differential-equation (ODE) to dynamically model node representations. This approach drastically reduces training time while achieving fine-grained post-training graph convolution to avoid the distortion of the original training embedding space, termed the embedding discrepancy issue. We validate our model across several real-world datasets of different scales, demonstrating that LightGODE not only outperforms GCN-based models in terms of efficiency and effectiveness but also significantly mitigates the embedding discrepancy commonly associated with deeper graph convolution layers. Our LightGODE challenges the prevailing paradigms in RecSys training and suggests re-evaluating the role of graph convolutions, potentially guiding future developments of efficient large-scale graph-based RecSys.

Henry Peng Zou, Vinay Samuel, Yue Zhou et al.

Existing datasets for attribute value extraction (AVE) predominantly focus on explicit attribute values while neglecting the implicit ones, lack product images, are often not publicly available, and lack an in-depth human inspection across diverse domains. To address these limitations, we present ImplicitAVE, the first, publicly available multimodal dataset for implicit attribute value extraction. ImplicitAVE, sourced from the MAVE dataset, is carefully curated and expanded to include implicit AVE and multimodality, resulting in a refined dataset of 68k training and 1.6k testing data across five domains. We also explore the application of multimodal large language models (MLLMs) to implicit AVE, establishing a comprehensive benchmark for MLLMs on the ImplicitAVE dataset. Six recent MLLMs with eleven variants are evaluated across diverse settings, revealing that implicit value extraction remains a challenging task for MLLMs. The contributions of this work include the development and release of ImplicitAVE, and the exploration and benchmarking of various MLLMs for implicit AVE, providing valuable insights and potential future research directions. Dataset and code are available at https://github.com/HenryPengZou/ImplicitAVE

Ravishka Rathnasuriya, Zihe Song, Nidhi Majoju et al.

Tile-based programming frameworks are increasingly adopted to write high-performance GPU kernels in domains such as deep learning and scientific computing. While these frameworks enhance productivity and hardware utilization, their multi-stage compilation pipelines introduce distinct code generation bugs that are tightly coupled to input shapes, data types, and backend targets. These bugs often manifest as silent correctness or performance issues, making them difficult to detect using existing compiler testing tools. Additionally, the unique programming conventions of tile domain-specific languages complicate root cause identification, while fixing such bugs demands specialized knowledge of tile abstractions and compilation pipelines. Despite the growing adoption of tile-based systems, their code generation bugs remain largely unexplored. This paper presents the first systematic study of tile-program code generation bugs. We curate 401 bug reports from GitHub and identify 301 tile-program codegen bugs for analysis, categorizing the root causes, symptoms, input patterns, test oracles that trigger these bugs, and the strategies used to fix bugs. Our study provides foundational insights for building debugging, testing, and repair tools tailored to tile-based compiler infrastructures.

Shanghao Li, Jinda Han, Yibo Wang et al.

Graph-based Retrieval-Augmented Generation (GraphRAG) enhances Large Language Models (LLMs) by incorporating external knowledge from linearized subgraphs retrieved from knowledge graphs. However, LLMs struggle to interpret the relational and topological information in these inputs, resulting in hallucinations that are inconsistent with the retrieved knowledge. To analyze how LLMs attend to and retain structured knowledge during generation, we propose two lightweight interpretability metrics: Path Reliance Degree (PRD), which measures over-reliance on shortest-path triples, and Semantic Alignment Score (SAS), which assesses how well the model's internal representations align with the retrieved knowledge. Through empirical analysis on a knowledge-based QA task, we identify failure patterns associated with over-reliance on salient paths and weak semantic grounding, as indicated by high PRD and low SAS scores. We further develop a lightweight post-hoc hallucination detector, Graph Grounding and Alignment (GGA), which outperforms strong semantic and confidence-based baselines across AUC and F1. By grounding hallucination analysis in mechanistic interpretability, our work offers insights into how structural limitations in LLMs contribute to hallucinations, informing the design of more reliable GraphRAG systems in the future.

Weizhi Zhang, Liangwei Yang, Zihe Song et al.

Recommender systems (RecSys) play a vital role in online platforms, offering users personalized suggestions amidst vast information. Graph contrastive learning aims to learn from high-order collaborative filtering signals with unsupervised augmentation on the user-item bipartite graph, which predominantly relies on the multi-task learning framework involving both the pair-wise recommendation loss and the contrastive loss. This decoupled design can cause inconsistent optimization direction from different losses, which leads to longer convergence time and even sub-optimal performance. Besides, the self-supervised contrastive loss falls short in alleviating the data sparsity issue in RecSys as it learns to differentiate users/items from different views without providing extra supervised collaborative filtering signals during augmentations. In this paper, we propose Mixed Supervised Graph Contrastive Learning for Recommendation (MixSGCL) to address these concerns. MixSGCL originally integrates the training of recommendation and unsupervised contrastive losses into a supervised contrastive learning loss to align the two tasks within one optimization direction. To cope with the data sparsity issue, instead unsupervised augmentation, we further propose node-wise and edge-wise mixup to mine more direct supervised collaborative filtering signals based on existing user-item interactions. Extensive experiments on three real-world datasets demonstrate that MixSGCL surpasses state-of-the-art methods, achieving top performance on both accuracy and efficiency. It validates the effectiveness of MixSGCL with our coupled design on supervised graph contrastive learning.

Jifan Zhang, Fangxin Wang, Zihe Song et al.

Stream networks, a unique class of spatiotemporal graphs, exhibit complex directional flow constraints and evolving dependencies, making uncertainty quantification a critical yet challenging task. Traditional conformal prediction methods struggle in this setting due to the need for joint predictions across multiple interdependent locations and the intricate spatio-temporal dependencies inherent in stream networks. Existing approaches either neglect dependencies, leading to overly conservative predictions, or rely solely on data-driven estimations, failing to capture the rich topological structure of the network. To address these challenges, we propose Spatio-Temporal Adaptive Conformal Inference (\texttt{STACI}), a novel framework that integrates network topology and temporal dynamics into the conformal prediction framework. \texttt{STACI} introduces a topology-aware nonconformity score that respects directional flow constraints and dynamically adjusts prediction sets to account for temporal distributional shifts. We provide theoretical guarantees on the validity of our approach and demonstrate its superior performance on both synthetic and real-world datasets. Our results show that \texttt{STACI} effectively balances prediction efficiency and coverage, outperforming existing conformal prediction methods for stream networks.

Ravishka Rathnasuriya, Tingxi Li, Zexin Xu et al.

Deep Learning Systems (DLSs) are increasingly deployed in real-time applications, including those in resourceconstrained environments such as mobile and IoT devices. To address efficiency challenges, Dynamic Deep Learning Systems (DDLSs) adapt inference computation based on input complexity, reducing overhead. While this dynamic behavior improves efficiency, such behavior introduces new attack surfaces. In particular, efficiency adversarial attacks exploit these dynamic mechanisms to degrade system performance. This paper systematically explores efficiency robustness of DDLSs, presenting the first comprehensive taxonomy of efficiency attacks. We categorize these attacks based on three dynamic behaviors: (i) attacks on dynamic computations per inference, (ii) attacks on dynamic inference iterations, and (iii) attacks on dynamic output production for downstream tasks. Through an in-depth evaluation, we analyze adversarial strategies that target DDLSs efficiency and identify key challenges in securing these systems. In addition, we investigate existing defense mechanisms, demonstrating their limitations against increasingly popular efficiency attacks and the necessity for novel mitigation strategies to secure future adaptive DDLSs.

Yuanjie Zhu, Liangwei Yang, Ke Xu et al.

Large Language Models (LLMs) are reshaping unsupervised learning by offering an unprecedented ability to perform text clustering based on their deep semantic understanding. However, their direct application is fundamentally limited by a lack of stateful memory for iterative refinement and the difficulty of managing cluster granularity. As a result, existing methods often rely on complex pipelines with external modules, sacrificing a truly end-to-end approach. We introduce LLM-MemCluster, a novel framework that reconceptualizes clustering as a fully LLM-native task. It leverages a Dynamic Memory to instill state awareness and a Dual-Prompt Strategy to enable the model to reason about and determine the number of clusters. Evaluated on several benchmark datasets, our tuning-free framework significantly and consistently outperforms strong baselines. LLM-MemCluster presents an effective, interpretable, and truly end-to-end paradigm for LLM-based text clustering.

Dezhi Ran, Yuan Cao, Mengzhou Wu et al.

Large language models (LLMs) have demonstrated remarkable capability in function-level code generation tasks. Unlike isolated functions, real-world applications demand reasoning over the entire software system: developers must orchestrate how different components interact, maintain consistency across states over time, and ensure the application behaves correctly within the lifecycle and framework constraints. Yet, no existing benchmark adequately evaluates whether LLMs can bridge this gap and construct entire software systems from scratch. To address this gap, we propose APPFORGE, a benchmark consisting of 101 software development problems drawn from real-world Android apps. Given a natural language specification detailing the app functionality, a language model is tasked with implementing the functionality into an Android app from scratch. Developing an Android app from scratch requires understanding and coordinating app states, lifecycle management, and asynchronous operations, calling for LLMs to generate context-aware, robust, and maintainable code. To construct APPFORGE, we design a multi-agent system to automatically summarize the main functionalities from app documents and navigate the app to synthesize test cases validating the functional correctness of app implementation. Following rigorous manual verification by Android development experts, APPFORGE incorporates the test cases within an automated evaluation framework that enables reproducible assessment without human intervention, making it easily adoptable for future research. Our evaluation on 12 flagship LLMs show that all evaluated models achieve low effectiveness, with the best-performing model (GPT-5) developing only 18.8% functionally correct applications, highlighting fundamental limitations in current models' ability to handle complex, multi-component software engineering challenges.

Liangwei Yang, Hengrui Zhang, Zihe Song et al.

This paper answers a fundamental question in artificial neural network (ANN) design: We do not need to build ANNs layer-by-layer sequentially to guarantee the Directed Acyclic Graph (DAG) property. Drawing inspiration from biological intelligence (BI), where neurons form a complex, graph-structured network, we introduce the groundbreaking Cyclic Neural Networks (Cyclic NNs). It emulates the flexible and dynamic graph nature of biological neural systems, allowing neuron connections in any graph-like structure, including cycles. This offers greater adaptability compared to the DAG structure of current ANNs. We further develop the Graph Over Multi-layer Perceptron, which is the first detailed model based on this new design paradigm. Experimental validation of the Cyclic NN's advantages on widely tested datasets in most generalized cases, demonstrating its superiority over current BP training methods through the use of a forward-forward (FF) training algorithm. This research illustrates a totally new ANN design paradigm, which is a significant departure from current ANN designs, potentially leading to more biologically plausible AI systems.

Simin Chen, Zihe Song, Mirazul Haque et al.

Neural image caption generation (NICG) models have received massive attention from the research community due to their excellent performance in visual understanding. Existing work focuses on improving NICG model accuracy while efficiency is less explored. However, many real-world applications require real-time feedback, which highly relies on the efficiency of NICG models. Recent research observed that the efficiency of NICG models could vary for different inputs. This observation brings in a new attack surface of NICG models, i.e., An adversary might be able to slightly change inputs to cause the NICG models to consume more computational resources. To further understand such efficiency-oriented threats, we propose a new attack approach, NICGSlowDown, to evaluate the efficiency robustness of NICG models. Our experimental results show that NICGSlowDown can generate images with human-unnoticeable perturbations that will increase the NICG model latency up to 483.86%. We hope this research could raise the community's concern about the efficiency robustness of NICG models.

Alan Romano, Zihe Song, Sampath Grandhi et al.

Flaky tests have gained attention from the research community in recent years and with good reason. These tests lead to wasted time and resources, and they reduce the reliability of the test suites and build systems they affect. However, most of the existing work on flaky tests focus exclusively on traditional unit tests. This work ignores UI tests that have larger input spaces and more diverse running conditions than traditional unit tests. In addition, UI tests tend to be more complex and resource-heavy, making them unsuited for detection techniques involving rerunning test suites multiple times. In this paper, we perform a study on flaky UI tests. We analyze 235 flaky UI test samples found in 62 projects from both web and Android environments. We identify the common underlying root causes of flakiness in the UI tests, the strategies used to manifest the flaky behavior, and the fixing strategies used to remedy flaky UI tests. The findings made in this work can provide a foundation for the development of detection and prevention techniques for flakiness arising in UI tests.