Significant advancements have been achieved in both fields of Natural Language Processing (NLP) and Computer Vision (CV) with the advent of Multimodal Large Language Models (MLLMs), sometimes referred to as large vision-language models (LVMs). MLLMs show promising ability in multimodal tasks, such as image captioning, visual question answering, etc. However, there is a concerning trend associated with the advancement in MLLMs. These models exhibit an inclination to generate hallucinations and misleading facts, resulting in seemingly plausible yet factually spurious content. To address these challenges, our team, DeepSIX, leverages recent advances in MLLMs to enhance the ability to detect hallucination and verify factual information within the scope of the ACM MM 2025 grand challenge 8: Truthful and Responsible Multimodal Learning (ResMM). We participated in both tasks: Multimodal Hallucination Detection (Task 1) and Multimodal Fact Checking (Task 2). Our approach leverages the interpretive power of the vision and language components of vision language models (VLMs) to analyze and summarize insights from text and images. It performs contextual reasoning by uncovering semantic relationships among entities in the text and objects in the images. By employing diverse prompting techniques, our method deconstructs critical entities in the text, effectively uncovers implicit relationships between text and images, and identifies hallucinations and false facts. Experimental results demonstrate the strength of our approach: it achieved second place in the Hallucination Detection task and third place in the Fact Verification task, confirming the potential of LLM-based methods in MLLMs. We open-source our code at https://github.com/JAIST-DeepSIX/ACMMM25

Online service failures in production computing environments are notoriously difficult to debug. One of the key challenges is to allow the developer to replay the failure execution within an interactive debugging tool such as GDB. Previous work has proposed in-situ approaches to inferring the production-run failure path within the production environment. However, those tools may sometimes suggest failure execution paths that are infeasible to reach by any program inputs. Moreover, production site often does not record or provide failure-triggering inputs due to the user privacy concern. In this paper, we present RDE, a Replay DEbug system that can replay a production-site failure at the development site within an interactive debugging environment without requiring user inputs. RDE takes an inferred production failure path as input and performs execution synthesis using a new guided symbolic execution technique. RDE can tolerate imprecise or inaccurate failure path information by navigating the symbolic execution along a set of selected paths. RDE synthesizes an input from the selected symbolic execution path which can be fed to a debugging tool to replay the failure. We have implemented an initial prototype of RDE and tested it with a set of coreutils bugs. The results show that RDE can successfully replay all the tested bugs within GDB.

Infrastructure-as-a-service clouds are becoming widely adopted. However, resource sharing and multi-tenancy have made performance anomalies a top concern for users. Timely debugging those anomalies is paramount for minimizing the performance penalty for users. Unfortunately, this debugging often takes a long time due to the inherent complexity and sharing nature of cloud infrastructures. When an application experiences a performance anomaly, it is important to distinguish between faults with a global impact and faults with a local impact as the diagnosis and recovery steps forfaults with a global impact or local impact are quite different. In this paper, we present PerfCompass, an online performance anomaly fault debugging tool that can quantify whether a production-run performance anomaly has a global impact or local impact. PerfCompass can use this information to suggest the root cause as either an external fault (e.g., environment-based) or an internal fault (e.g., software bugs). Furthermore, PerfCompass can identify top affected system calls to provide useful diagnostic hints for detailed performance debugging. PerfCompass does not require source code or runtime application instrumentation, which makes it practical for production systems. We have tested PerfCompass by running five common open source systems (e.g., Apache, MySQL, Tomcat, Hadoop, Cassandra) inside a virtualized cloud testbed. Our experiments use a range of common infrastructure sharing issues and real software bugs. The results show that PerfCompass accurately classifies 23 out of the 24 tested cases without calibration and achieves 100 percent accuracy with calibration. PerfCompass provides useful diagnosis hints within several minutes and imposes negligible runtime overhead to the production system during normal execution time.

Performance bugs which manifest in a production cloud computing infrastructure are notoriously difficult to diagnose because of both the difficulty of reproducing those bugs and the lack of debugging information. In this paper, we present PerfScope, a practical online performance bug inference tool to help the developer understand how a performance bug happened during the production run. PerfScope achieves online bug inference to obviate the need for offline bug reproduction. PerfScope does not require application source code or any runtime instrumentation to the production system. PerfScope is application-agnostic, which can support both interpreted and compiled programs running inside a cloud infrastructure.

Virtual Arc Consistency (VAC) is a recent local consistency for processing cost function networks (aka weighted constraint networks) that exploits a simple but powerful connection with standard constraint networks. It has allowed to close hard frequency assignment benchmarks and is capable of directly solving networks of sub modular functions. The algorithm enforcing VAC is an iterative algorithm that solves a sequence of standard constraint networks. This algorithm has been improved by exploiting the idea of dynamic arc consistency between each iteration, leading to the dynamic VAC algorithm. When VAC is maintained during search, the difference between two adjacent nodes in the search tree is also limited. In this paper, we show that the incrementality of Dynamic VAC can also be useful when maintaining VAC during search and we present results showing that maintaining dynamic VAC during search can effectively accelerate search.

Distributed applications running inside cloud systems are prone to performance anomalies due to various reasons such as resource contentions, software bugs, and hardware failures. One big challenge for diagnosing an abnormal distributed application is to pinpoint the faulty components. In this paper, we present a black-box online fault localization system called FChain that can pinpoint faulty components immediately after a performance anomaly is detected. FChain first discovers the onset time of abnormal behaviors at different components by distinguishing the abnormal change point from many change points caused by normal workload fluctuations. Faulty components are then pinpointed based on the abnormal change propagation patterns and inter-component dependency relationships. FChain performs runtime validation to further filter out false alarms. We have implemented FChain on top of the Xen platform and tested it using several benchmark applications (RUBiS, Hadoop, and IBM System S). Our experimental results show that FChain can quickly pinpoint the faulty components with high accuracy within a few seconds. FChain can achieve up to 90% higher precision and 20% higher recall than existing schemes. FChain is non-intrusive and light-weight, which imposes less than 1% overhead to the cloud system.

Virtualized cloud systems are prone to performance anomalies due to various reasons such as resource contentions, software bugs, and hardware failures. In this paper, we present a novel Predictive Performance Anomaly Prevention (PREPARE) system that provides automatic performance anomaly prevention for virtualized cloud computing infrastructures. PREPARE integrates online anomaly prediction, learning-based cause inference, and predictive prevention actuation to minimize the performance anomaly penalty without human intervention. We have implemented PREPARE on top of the Xen platform and tested it on the NCSU's Virtual Computing Lab using a commercial data stream processing system (IBM System S) and an online auction benchmark (RUBiS). The experimental results show that PREPARE can effectively prevent performance anomalies while imposing low overhead to the cloud infrastructure.

Commercial telerobotic surgery systems for soft tissue surgery are generally limited to visual imaging, though it is possible to simultaneously view picture-in-picture visualization of another workstation. However, it is not easy to correlate such information with the primary endoscopic view since it may not relate to the surface visible in the visual endoscopic images. As critical surfaces as well as surgical targets often lie subsurface, a range of techniques (e.g. ultrasound and near-infrared imaging) and registration methods have been investigated as robotic surgery gains popularity. While investigation of nerves, blood vessels, and tumors has received prior attention, we present a new prototype system for real-time multimodal image registration that focuses on the visualization of the urinary tract. By providing an accurate registration between stereo video images and a near infrared imager, we aim to enhance surgical awareness and make critical uretary tasks such as mobilization of the ureters easier. Engineering validation experiments with a prototype imager, and in-vivo experiments using a prototype Hamamatsu Photodynamic Eye (PDE) imager together with the da Vinci surgical system demonstrating feasibility are presented. These initial experiments have also shown encouraging response from the clinicians.