Mingyuan Wu

Banghao Chi, Yining Xie, Mingyuan Wu et al.

Spreadsheet systems (e.g., Microsoft Excel, Google Sheets) play a central role in modern data-centric workflows. As AI agents grow increasingly capable of automating complex tasks, such as controlling computers and generating presentations, building an AI-driven spreadsheet agent has emerged as a promising research direction. Most existing spreadsheet agents rely on specialized prompting over general-purpose LLMs; while this design has potentials on simple spreadsheet operations, it struggles to manage the complex, multi-step workflows typical of real-world applications. We introduce Spreadsheet-RL, a reinforcement learning (RL) fine-tuning framework designed to train specialized spreadsheet agents within a realistic Microsoft Excel environment. Spreadsheet-RL features an automated pipeline for scalable collection of paired start-goal spreadsheets from online forums, as well as domain-specific evaluation tasks in areas such as finance and supply chain management, which we compile into the new Domain-Spreadsheet benchmark dataset. It also includes a Spreadsheet Gym environment designed for multi-turn RL: Spreadsheet Gym exposes extensive Excel functionality through a Python sandbox, along with a refined harness that incorporates a comprehensive tool set and carefully designed tool-routing rules for spreadsheet tasks. Through comprehensive experiments, we show that Spreadsheet-RL substantially enhances AI agent's performance on both general and domain-specific spreadsheet tasks: it improves Qwen3-4B-Thinking-2507's Pass@1 on SpreadsheetBench from 12.0% to 23.4%, and raises Pass@1 from 8.4% to 17.2% on our curated Domain-Spreadsheet dataset. These results highlight Spreadsheet-RL's strong potential for generalization and real-world adoption in spreadsheet automation, and broadly, its promise for advancing LLM-based interactions with data interfaces in everyday work.

Jingcheng Yang, Tianhu Xiong, Shengyi Qian et al.

Vision-language models (VLMs) are powerful but remain opaque black boxes. We introduce the first framework for transparent circuit tracing in VLMs to systematically analyze multimodal reasoning. By utilizing transcoders, attribution graphs, and attention-based methods, we uncover how VLMs hierarchically integrate visual and semantic concepts. We reveal that distinct visual feature circuits can handle mathematical reasoning and support cross-modal associations. Validated through feature steering and circuit patching, our framework proves these circuits are causal and controllable, laying the groundwork for more explainable and reliable VLMs.

Xinyu Pi, Mingyuan Wu, Jize Jiang et al.

Smaller-scale Vision-Langauge Models (VLMs) often claim to perform on par with larger models in general-domain visual grounding and question-answering benchmarks while offering advantages in computational efficiency and storage. However, their ability to handle rare objects, which fall into the long tail of data distributions, is less understood. To rigorously evaluate this aspect, we introduce the "Uncontextualized Uncommon Objects" (UOUO) benchmark. This benchmark focuses on systematically testing VLMs with both large and small parameter counts on rare and specialized objects. Our comprehensive analysis reveals that while smaller VLMs maintain competitive performance on common datasets, they significantly underperform on tasks involving uncommon objects. We also propose an advanced, scalable pipeline for data collection and cleaning, ensuring the UOUO benchmark provides high-quality, challenging instances. These findings highlight the need to consider long-tail distributions when assessing the true capabilities of VLMs.

Yingjie Yu, Mingyuan Wu, Ahmadreza Eslaminia et al.

QoS-QoE translation is a fundamental problem in multimedia systems because it characterizes how measurable system and network conditions affect user-perceived experience. Although many prior studies have examined this relationship, their findings are often developed for specific setups and remain scattered across papers, experimental settings, and reporting formats, limiting systematic reuse, cross-scenario generalization, and large-scale analysis. To address this gap, we first introduce QoS-QoE Translation dataset, a source-grounded dataset of structured QoS-QoE relationships from the multimedia literature, with a focus on video streaming related tasks. We construct the dataset through an automated pipeline that combines paper curation, QoS-QoE relationship extraction, and iterative data evaluation. Each record preserves the extracted relationship together with parameter definitions, supporting evidence, and contextual metadata. We further evaluate the capability of large language models (LLMs) on QoS-QoE translation, both before and after supervised fine-tuning on our dataset, and show strong performance on both continuous-value and discrete-label prediction in bidirectional translation, from QoS-QoE and QoE-QoS. Our dataset provides a foundation for benchmarking LLMs in QoS-QoE translation and for supporting future LLM-based reasoning for multimedia quality prediction and optimization. The complete dataset and code are publicly available at https://yyu6969.github.io/qos-qoe-translation-page/, for full reproducibility and open access.

Mingyuan Wu, Jize Jiang, Haozhen Zheng et al.

Vision Language Models (VLMs) have achieved remarkable success in a wide range of vision applications of increasing complexity and scales, yet choosing the right VLM model size involves a trade-off between response quality and cost. While smaller VLMs are cheaper to run, they typically produce responses only marginally better than random guessing on benchmarks such as MMMU. In this paper, we propose Cache of Thought (CoT), a master apprentice framework for collaborative inference between large and small VLMs. CoT manages high quality query results from large VLMs (master) in a cache, which are then selected via a novel multi modal retrieval and in-context learning to aid the performance of small VLMs (apprentice). We extensively evaluate CoT on various widely recognized and challenging general reasoning benchmarks, and show that CoT increases overall reasoning performance by up to 7.7% under the same budget, and specifically boosts the performance of apprentice VLMs by up to 36.6%. Our code is available at https://github.com/UIUC-MONET/Cache-of-Thoughts

Jiachen Liu, Jiaxin Pei, Jintao Huang et al.

Scientific publication compresses a branching, iterative research process into a linear narrative, discarding the majority of what was discovered along the way. This compilation imposes two structural costs: a Storytelling Tax, where failed experiments, rejected hypotheses, and the branching exploration process are discarded to fit a linear narrative; and an Engineering Tax, where the gap between reviewer-sufficient prose and agent-sufficient specification leaves critical implementation details unwritten. Tolerable for human readers, these costs become critical when AI agents must understand, reproduce, and extend published work. We introduce the Agent-Native Research Artifact (Ara), a protocol that replaces the narrative paper with a machine-executable research package structured around four layers: scientific logic, executable code with full specifications, an exploration graph that preserves the failures compilation discards, and evidence grounding every claim in raw outputs. Three mechanisms support the ecosystem: a Live Research Manager that captures decisions and dead ends during ordinary development; an Ara Compiler that translates legacy PDFs and repos into Aras; and an Ara-native review system that automates objective checks so human reviewers can focus on significance, novelty, and taste. On PaperBench and RE-Bench, Ara raises question-answering accuracy from 72.4% to 93.7% and reproduction success from 57.4% to 64.4%. On RE-Bench's five open-ended extension tasks, preserved failure traces in Ara accelerate progress, but can also constrain a capable agent from stepping outside the prior-run box depending on the agent's capabilities.

Mingyuan Wu, Jingcheng Yang, Jize Jiang et al.

Reinforcement Learning Finetuning (RFT) has significantly advanced the reasoning capabilities of large language models (LLMs) by enabling long chains of thought, self-correction, and effective tool use. While recent works attempt to extend RFT to vision-language models (VLMs), these efforts largely produce text-only reasoning conditioned on static image inputs, falling short of true multimodal reasoning in the response. In contrast, test-time methods like Visual Sketchpad incorporate visual steps but lack training mechanisms. We introduce VTool-R1, the first framework that trains VLMs to generate multimodal chains of thought by interleaving text and intermediate visual reasoning steps. VTool-R1 integrates Python-based visual editing tools into the RFT process, enabling VLMs to learn when and how to generate visual reasoning steps that benefit final reasoning. Trained with outcome-based rewards tied to task accuracy, our approach elicits strategic visual tool use for reasoning without relying on process-based supervision. Experiments on structured visual question answering over charts and tables show that VTool-R1 enhances reasoning performance by teaching VLMs to "think with images" and generate multimodal chain of thoughts with tools.

Haozhen Zheng, Beitong Tian, Mingyuan Wu et al.

Despite significant recent progress of Multimodal Large Language Models (MLLMs), current MLLMs are challenged by "spatio-temporal" prompts, i.e., prompts that refer to 1) the entirety of an environment encoded in a point cloud that the MLLM should consider; and simultaneously also refer to 2) actions that happened in part of the environment and are encoded in a short ego-centric video clip. However, such a holistic spatio-temporal understanding is important for agents operating in the real world. To address this challenge, we first develop a framework to collect a large-scale dataset. Using the collected "Reasoning about Environments and Actions" (REA) dataset, we show that recent MLLMs indeed struggle to correctly answer "spatio-temporal" prompts. Building on this dataset, we study two spatio-temporal LLM (STLLM) baselines: 1) STLLM-3D, which directly fuses point cloud, video, and text representations as inputs to the LLM; and 2) STLLM-Aligner, which aligns spatial context with video and text before LLM decoding. Both baselines aim to enhance spatial understanding of environments and temporal grounding of egocentric observations. On REA, the STLLM baselines outperform existing models, demonstrating the effectiveness of our designs. Code and data are available at https://zoezheng126.github.io/STLLM-website/.

Beitong Tian, Lingzhi Zhao, Bo Chen et al.

Underwater activities like scuba diving enable millions annually to explore marine environments for recreation and scientific research. Maintaining situational awareness and effective communication are essential for diver safety. Traditional underwater communication systems are often bulky and expensive, limiting their accessibility to divers of all levels. While recent systems leverage lightweight smartphones and support text messaging, the messages are predefined and thus restrict context-specific communication. In this paper, we present AquaVLM, a tap-and-send underwater communication system that automatically generates context-aware messages and transmits them using ubiquitous smartphones. Our system features a mobile vision-language model (VLM) fine-tuned on an auto-generated underwater conversation dataset and employs a hierarchical message generation pipeline. We co-design the VLM and transmission, incorporating error-resilient fine-tuning to improve the system's robustness to transmission errors. We develop a VR simulator to enable users to experience AquaVLM in a realistic underwater environment and create a fully functional prototype on the iOS platform for real-world experiments. Both subjective and objective evaluations validate the effectiveness of AquaVLM and highlight its potential for personal underwater communication as well as broader mobile VLM applications.

Mingyuan Wu, Meitang Li, Jingcheng Yang et al.

Inference-time techniques such as decoding-time scaling and self-refinement have been shown to substantially improve reasoning in large language models (LLMs), driven by emergent self-correction and self-verification behaviors often elicited through reinforcement learning (RL). In this work, we investigate whether these inference-time scaling methods similarly benefit vision-language models (VLMs), especially those fine-tuned with RL. Through extensive evaluation, we find that while strategies like majority vote and best-of-N with self-verification enhance VLM performance, majority vote significantly outperforms verification-centric ones. Furthermore, inference time scaling behaviors commonly associated with RL-tuned models, such as the 'A-ha moment,' do not yield consistent performance gains. Our analysis identifies a key limitation: current RL-trained VLMs exhibit weak self-verification across both visual and textual modalities, limiting the effectiveness of inference-time scaling.

Ruiyang Xu, Jialun Cao, Mingyuan Wu et al.

Large Language Models (LLMs) have shown promise in various tasks, yet few benchmarks assess their capabilities in embedded system development.In this paper, we introduce EmbedAgent, a paradigm designed to simulate real-world roles in embedded system development, such as Embedded System Programmer, Architect, and Integrator. This paradigm enables LLMs to be tested in tasks that bridge the gap between digital and physical systems, allowing for a more comprehensive assessment of their capabilities. To evaluate LLMs on these tasks, we propose Embedbench, the first comprehensive benchmark for embedded system programming, circuit design, and cross-platform migration.Embedbench consists of 126 cases, covering 9 electronic components across 3 hardware platforms. Through extensive experiments on 10 mainstream LLMs, we uncover several key findings. Surprisingly, despite the simplicity of the cases, DeepSeek-R1 achieves only a 55.6% pass@1 rate when provided with schematic information, and 50.0% when tasked with generating the schematics itself. In the cross-platform migration tasks, LLMs show relatively strong performance with MicroPython on the Raspberry Pi Pico (with the top model achieving 73.8% pass@1), but perform poorly on ESP-IDF, where the best model reaches only 29.4% pass@1.Interestingly, we observe that general-purpose chat LLMs like DeepSeek-V3 often fail to utilize relevant pre-trained knowledge in this domain, while reasoning LLMs tend to overthink and overlook efficient knowledge during pretraining. Based on these insights, we propose two strategies: retrieval augmented generation and compiler feedback-to enhance LLM performance. These strategies result in significant improvements, with Deepseek-R1 reaching a 65.1% pass@1 with correct schematics, and 53.1% without. Additionally, the accuracy of the Arduino to ESP32 migration task improves from 21.4% to 27.8%.

Mingyuan Wu, Zichuan Liu, Haozhen Zheng et al.

Efficient single instance segmentation is essential for unlocking features in the mobile imaging applications, such as capture or editing. Existing on-the-fly mobile imaging applications scope the segmentation task to portraits or the salient subject due to the computational constraints. Instance segmentation, despite its recent developments towards efficient networks, is still heavy due to the cost of computation on the entire image to identify all instances. To address this, we propose and formulate a one tap driven single instance segmentation task that segments a single instance selected by a user via a positive tap. This task, in contrast to the broader task of segmenting anything as suggested in the Segment Anything Model \cite{sam}, focuses on efficient segmentation of a single instance specified by the user. To solve this problem, we present TraceNet, which explicitly locates the selected instance by way of receptive field tracing. TraceNet identifies image regions that are related to the user tap and heavy computations are only performed on selected regions of the image. Therefore overall computation cost and memory consumption are reduced during inference. We evaluate the performance of TraceNet on instance IoU average over taps and the proportion of the region that a user tap can fall into for a high-quality single-instance mask. Experimental results on MS-COCO and LVIS demonstrate the effectiveness and efficiency of the proposed approach. TraceNet can jointly achieve the efficiency and interactivity, filling in the gap between needs for efficient mobile inference and recent research trend towards multimodal and interactive segmentation models.

Mingyuan Wu, Husheng Zhou, Lingming Zhang et al.

While CUDA has become a major parallel computing platform and programming model for general-purpose GPU computing, CUDA-induced bug patterns have not yet been well explored. In this paper, we conduct the first empirical study to reveal important categories of CUDA program bug patterns based on 319 bugs identified within 5 popular CUDA projects in GitHub. Our findings demonstrate that CUDA-specific characteristics may cause program bugs such as synchronization bugs that are rather difficult to detect. To efficiently detect such synchronization bugs, we establish the first lightweight general CUDA bug detection framework, namely Simulee, to simulate CUDA program execution by interpreting the corresponding llvm bytecode and collecting the memory-access information to automatically detect CUDA synchronization bugs. To evaluate the effectiveness and efficiency of Simulee, we conduct a set of experiments and the experimental results suggest that Simulee can detect 20 out of the 27 studied synchronization bugs and successfully detects 26 previously unknown synchronization bugs, 10 of which have been confirmed by the developers.