Zirong Chen

Zirong Chen, Isaac Li, Haoxiang Zhang et al.

An increasing number of monitoring systems have been developed in smart cities to ensure that real-time operations of a city satisfy safety and performance requirements. However, many existing city requirements are written in English with missing, inaccurate, or ambiguous information. There is a high demand for assisting city policy makers in converting human-specified requirements to machine-understandable formal specifications for monitoring systems. To tackle this limitation, we build CitySpec, the first intelligent assistant system for requirement specification in smart cities. To create CitySpec, we first collect over 1,500 real-world city requirements across different domains from over 100 cities and extract city-specific knowledge to generate a dataset of city vocabulary with 3,061 words. We also build a translation model and enhance it through requirement synthesis and develop a novel online learning framework with validation under uncertainty. The evaluation results on real-world city requirements show that CitySpec increases the sentence-level accuracy of requirement specification from 59.02% to 86.64%, and has strong adaptability to a new city and a new domain (e.g., F1 score for requirements in Seattle increases from 77.6% to 93.75% with online learning).

Zirong Chen, Issa Li, Haoxiang Zhang et al.

An increasing number of monitoring systems have been developed in smart cities to ensure that the real-time operations of a city satisfy safety and performance requirements. However, many existing city requirements are written in English with missing, inaccurate, or ambiguous information. There is a high demand for assisting city policymakers in converting human-specified requirements to machine-understandable formal specifications for monitoring systems. To tackle this limitation, we build CitySpec, the first intelligent assistant system for requirement specification in smart cities. To create CitySpec, we first collect over 1,500 real-world city requirements across different domains (e.g., transportation and energy) from over 100 cities and extract city-specific knowledge to generate a dataset of city vocabulary with 3,061 words. We also build a translation model and enhance it through requirement synthesis and develop a novel online learning framework with shielded validation. The evaluation results on real-world city requirements show that CitySpec increases the sentence-level accuracy of requirement specification from 59.02% to 86.64%, and has strong adaptability to a new city and a new domain (e.g., the F1 score for requirements in Seattle increases from 77.6% to 93.75% with online learning). After the enhancement from the shield function, CitySpec is now immune to most known textual adversarial inputs (e.g., the attack success rate of DeepWordBug after the shield function is reduced to 0% from 82.73%). We test the CitySpec with 18 participants from different domains. CitySpec shows its strong usability and adaptability to different domains, and also its robustness to malicious inputs.

Zirong Chen, Isaac Li, Haoxiang Zhang et al.

As more and more monitoring systems have been deployed to smart cities, there comes a higher demand for converting new human-specified requirements to machine-understandable formal specifications automatically. However, these human-specific requirements are often written in English and bring missing, inaccurate, or ambiguous information. In this paper, we present CitySpec, an intelligent assistant system for requirement specification in smart cities. CitySpec not only helps overcome the language differences brought by English requirements and formal specifications, but also offers solutions to those missing, inaccurate, or ambiguous information. The goal of this paper is to demonstrate how CitySpec works. Specifically, we present three demos: (1) interactive completion of requirements in CitySpec; (2) human-in-the-loop correction while CitySepc encounters exceptions; (3) online learning in CitySpec.

Hao Liang, Linzhuang Sun, Minxuan Zhou et al.

With the rapid progress of Multimodal LLMs, evaluating their mathematical reasoning capabilities has become an increasingly important research direction. In particular, visual-textual mathematical reasoning serves as a key indicator of an MLLM's ability to comprehend and solve complex, multi-step quantitative problems. While existing benchmarks such as MathVista and MathVerse have advanced the evaluation of multimodal math proficiency, they primarily rely on digitally rendered content and fall short in capturing the complexity of real-world scenarios. To bridge this gap, we introduce MathScape, a novel benchmark focused on assessing MLLMs' reasoning ability in realistic mathematical contexts. MathScape comprises 1,369 high-quality math problems paired with human-captured real-world images, closely reflecting the challenges encountered in practical educational settings. We conduct a thorough multi-dimensional evaluation across nine leading closed-source MLLMs, three open-source MLLMs with over 20 billion parameters, and seven smaller-scale MLLMs. Our results show that even state-of-the-art models struggle with real-world math tasks, lagging behind human performance, highlighting critical limitations in current model capabilities. Moreover, we find that strong performance on synthetic or digitally rendered images does not guarantee similar effectiveness on real-world tasks. This underscores the necessity of MathScape in the next stage of multimodal mathematical reasoning.

Weichu Xie, Haozhe Zhao, Wenpu Liu et al.

Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve reasoning in large language models, but rewards only final-answer correctness with no supervision over intermediate steps. Rubric-based methods such as Rubrics as Rewards (RaR) introduce finer-grained supervision by scoring rollouts against structured criteria, yet the rubric scores are still aggregated into a single scalar applied to the entire response, causing three weaknesses: loss of multi-criterion structure, uniform supervision of correct and incorrect steps, and reward hacking through unbounded self-correction. On 1,000 problems, we find 18.2% of steps in correct-answer responses are wrong yet positively rewarded, while 49.9% of steps in incorrect-answer responses are correct yet penalized. We introduce Step-wise Rubrics as Rewards (SRaR), an RLVR framework that (i) uses an LLM judge to attribute each rubric item to a specific reasoning step, (ii) normalizes per-step rubric scores across rollouts so only steps whose quality varies produce a learning signal, and (iii) combines the per-step reward with the outcome reward through a decoupled advantage estimator that keeps the outcome baseline stable. We further build a 16K-problem rubric dataset by contrastively distilling rubric items from correct and flawed reasoning paths sampled from a strong model. Across six mathematical reasoning benchmarks, SRaR improves average accuracy over RaR by 3.57 points on Qwen3-8B and 2.75 points on Qwen3-32B, raises the Faithful Reasoning Rate on AIME 2025 from 34.5% to 46.7%, and reduces self-correction looping from 48.1% to 26.5%.

Zirong Chen, Meiyi Ma

Emergency call-takers form the first operational link in public safety response, handling over 240 million calls annually while facing a sustained training crisis: staffing shortages exceed 25\% in many centers, and preparing a single new hire can require up to 720 hours of one-on-one instruction that removes experienced personnel from active duty. Traditional training approaches struggle to scale under these constraints, limiting both coverage and feedback timeliness. In partnership with Metro Nashville Department of Emergency Communications (MNDEC), we designed, developed, and deployed a GenAI-powered call-taking training system under real-world constraints. Over six months, deployment scaled from initial pilot to 190 operational users across 1,120 training sessions, exposing systematic challenges around system delivery, rigor, resilience, and human factors that remain largely invisible in controlled or purely simulated evaluations. By analyzing deployment logs capturing 98,429 user interactions, organizational processes, and stakeholder engagement patterns, we distill four key lessons, each coupled with concrete design and governance practices. These lessons provide grounded guidance for researchers and practitioners seeking to embed AI-driven training systems in safety-critical public sector environments where embedded constraints fundamentally shape socio-technical design.

Jiacheng Wang, Ping Jian, Zhen Yang et al.

Large language models (LLMs) have made rapid progress, yet adapting them to downstream scenarios still commonly relies on supervised fine-tuning (SFT). When downstream data exhibit a substantial distribution shift from the model's prior training distribution, SFT can induce catastrophic forgetting. To narrow this gap, data rewriting has been proposed as a data-centric approach that rewrites downstream training data prior to SFT. However, existing methods typically sample rewrites from a prompt-induced conditional distribution, so the resulting targets are not necessarily aligned with the model's natural QA-style generation distribution. Moreover, reliance on fixed templates can lead to diversity collapse. To address these issues, we cast data rewriting as a policy learning problem and learn a rewriting policy that better matches the backbone's QA-style generation distribution while preserving diversity. Since distributional alignment, diversity and task consistency are automatically evaluable but difficult to optimize end-to-end with differentiable objectives, we leverage reinforcement learning to optimize the rewrite distribution under reward feedback and propose an RL-based data-rewriting agent. The agent jointly optimizes QA-style distributional alignment and diversity under a hard task-consistency gate, thereby constructing a higher-quality rewritten dataset for downstream SFT. Extensive experiments show that our method achieves downstream gains comparable to standard SFT while reducing forgetting on non-downstream benchmarks by 12.34% on average. Our code is available at https://anonymous.4open.science/r/Patch-the-Prompt-Gap-4112 .

Kuan Zhang, Dongchen Liu, Qiyue Zhao et al.

The real world unfolds along a single set of physics laws, yet human intelligence demonstrates a remarkable capacity to generalize experiences from this singular physical existence into a multiverse of games, each governed by entirely different rules, aesthetics, physics, and objectives. This omni-reality adaptability is a hallmark of general intelligence. As Artificial Intelligence progresses towards Artificial General Intelligence, the multiverse of games has evolved from mere entertainment into the ultimate ground for training and evaluating AGI. The pursuit of this generality has unfolded across four eras: from environment-specific symbolic and reinforcement learning agents, to current large foundation models acting as generalist players, and toward a future creator stage where agent both creates new game worlds and continually evolves within them. We trace the full lifecycle of a generalist game player along four interdependent pillars: Dataset, Model, Harness, and Benchmark. Every advance across these pillars can be read as an attempt to break one of five fundamental trade-offs that currently bound the whole system. Building on this end-to-end view, we chart a five-level roadmap, progressing from single-game mastery to the ultimate creator stage in which the agent simultaneously creates and evolves within theoretical game multiverse. Taken together, our work offers a unified lens onto a rapidly shifting field,and a principled path toward the omnipotent generalist agent capable of seamlessly mastering any challenge within the multiverse of games, thereby paving the way for AGI.

Hanchen David Wang, Siwoo Bae, Zirong Chen et al.

Artificial intelligence systems in critical fields like autonomous driving and medical imaging analysis often continually learn new tasks using a shared stream of input data. For instance, after learning to detect traffic signs, a model may later need to learn to classify traffic lights or different types of vehicles using the same camera feed. This scenario introduces a challenging setting we term Continual Multitask Learning (CMTL), where a model sequentially learns new tasks on an underlying data distribution without forgetting previously learned abilities. Existing continual learning methods often fail in this setting because they learn fragmented, task-specific features that interfere with one another. To address this, we introduce Learning with Preserving (LwP), a novel framework that shifts the focus from preserving task outputs to maintaining the geometric structure of the shared representation space. The core of LwP is a Dynamically Weighted Distance Preservation (DWDP) loss that prevents representation drift by regularizing the pairwise distances between latent data representations. This mechanism of preserving the underlying geometric structure allows the model to retain implicit knowledge and support diverse tasks without requiring a replay buffer, making it suitable for privacy-conscious applications. Extensive evaluations on time-series and image benchmarks show that LwP not only mitigates catastrophic forgetting but also consistently outperforms state-of-the-art baselines in CMTL tasks. Notably, our method shows superior robustness to distribution shifts and is the only approach to surpass the strong single-task learning baseline, underscoring its effectiveness for real-world dynamic environments.

Zirong Chen, Haotian Xue

Due to the unfamiliarity to particular words(or proper nouns) for ingredients, non-native English speakers can be extremely confused about the ordering process in restaurants like Subway. Thus, We developed a dialogue system, which supports Chinese(Mandarin)1 and English2 at the same time. In other words, users can switch arbitrarily between Chinese(Mandarin) and English as the conversation is being conducted. This system is specifically designed for Subway ordering3. In BilDOS, we designed a Discriminator module to tell the language is being used in inputted user utterance, a Translator module to translate used language into English if it is not English, and a Dialogue Manager module to detect the intention within inputted user utterances, handle outlier inputs by throwing clarification requests, map detected Intention and detailed Keyword4 into a particular intention class, locate the current ordering process, continue to give queries to finish the order, conclude the order details once the order is completed, activate the evaluation process when the conversation is done.

Zirong Chen

FrameNet( Fillmore and Baker [2009] ) is well-known for its wide use for knowledge representation in the form of inheritance-based ontologies and lexica( Trott et al. [2020] ). Although FrameNet is usually applied to languages like English, Spanish and Italian, there are still plenty of FrameNet data sets available for other languages like Chinese, which differs significantly from those languages based on Latin alphabets. In this paper, the translation from ancient Chinese Poetry to modern Chinese will be first conducted to further apply the Chinese FrameNet(CFN, provided by Shanxi University). Afterwards, the translation from modern Chinese will be conducted as well for the comparison between the applications of CFN and English FrameNet. Finally, the overall comparison will be draw between CFN to modern Chinese and English FrameNet.

Zirong Chen, Elizabeth Chason, Noah Mladenovski et al.

Emergency response services are vital for enhancing public safety by safeguarding the environment, property, and human lives. As frontline members of these services, 9-1-1 dispatchers have a direct impact on response times and the overall effectiveness of emergency operations. However, traditional dispatcher training methods, which rely on role-playing by experienced personnel, are labor-intensive, time-consuming, and often neglect the specific needs of underserved communities. To address these challenges, we introduce Sim911, the first training simulation for 9-1-1 dispatchers powered by Large Language Models (LLMs). Sim911 enhances training through three key technical innovations: (1) knowledge construction, which utilizes archived 9-1-1 call data to generate simulations that closely mirror real-world scenarios; (2) context-aware controlled generation, which employs dynamic prompts and vector bases to ensure that LLM behavior aligns with training objectives; and (3) validation with looped correction, which filters out low-quality responses and refines the system performance.

Hao Liang, Zirong Chen, Hejun Dong et al.

Video question-answering (QA) is a core task in video understanding. Evaluating the quality of video QA and video caption data quality for training video large language models (VideoLLMs) is an essential challenge. Although various methods have been proposed for assessing video caption quality, there remains a lack of dedicated evaluation methods for Video QA. To address this gap, we introduce EVQAScore, a reference-free method that leverages keyword extraction to assess both video caption and video QA data quality. Additionally, we incorporate frame sampling and rescaling techniques to enhance the efficiency and robustness of our evaluation, this enables our score to evaluate the quality of extremely long videos. Our approach achieves state-of-the-art (SOTA) performance (32.8 for Kendall correlation and 42.3 for Spearman correlation, 4.7 and 5.9 higher than the previous method PAC-S++) on the VATEX-EVAL benchmark for video caption evaluation. Furthermore, by using EVQAScore for data selection, we achieved SOTA results with only 12.5\% of the original data volume, outperforming the previous SOTA method PAC-S and 100\% of data.

Ziyan An, Xia Wang, Hendrik Baier et al.

In response to the lack of trust in Artificial Intelligence (AI) for sequential planning, we design a Computational Tree Logic-guided large language model (LLM)-based natural language explanation framework designed for the Monte Carlo Tree Search (MCTS) algorithm. MCTS is often considered challenging to interpret due to the complexity of its search trees, but our framework is flexible enough to handle a wide range of free-form post-hoc queries and knowledge-based inquiries centered around MCTS and the Markov Decision Process (MDP) of the application domain. By transforming user queries into logic and variable statements, our framework ensures that the evidence obtained from the search tree remains factually consistent with the underlying environmental dynamics and any constraints in the actual stochastic control process. We evaluate the framework rigorously through quantitative assessments, where it demonstrates strong performance in terms of accuracy and factual consistency.

Zirong Chen, Ziyan An, Jennifer Reynolds et al.

Emergency response services are critical to public safety, with 9-1-1 call-takers playing a key role in ensuring timely and effective emergency operations. To ensure call-taking performance consistency, quality assurance is implemented to evaluate and refine call-takers' skillsets. However, traditional human-led evaluations struggle with high call volumes, leading to low coverage and delayed assessments. We introduce LogiDebrief, an AI-driven framework that automates traditional 9-1-1 call debriefing by integrating Signal-Temporal Logic (STL) with Large Language Models (LLMs) for fully-covered rigorous performance evaluation. LogiDebrief formalizes call-taking requirements as logical specifications, enabling systematic assessment of 9-1-1 calls against procedural guidelines. It employs a three-step verification process: (1) contextual understanding to identify responder types, incident classifications, and critical conditions; (2) STL-based runtime checking with LLM integration to ensure compliance; and (3) automated aggregation of results into quality assurance reports. Beyond its technical contributions, LogiDebrief has demonstrated real-world impact. Successfully deployed at Metro Nashville Department of Emergency Communications, it has assisted in debriefing 1,701 real-world calls, saving 311.85 hours of active engagement. Empirical evaluation with real-world data confirms its accuracy, while a case study and extensive user study highlight its effectiveness in enhancing call-taking performance.

Zirong Chen, Xutong Sun, Yuanhe Li et al.

Emergency and non-emergency response systems are essential services provided by local governments and critical to protecting lives, the environment, and property. The effective handling of (non-)emergency calls is critical for public safety and well-being. By reducing the burden through non-emergency callers, residents in critical need of assistance through 911 will receive a fast and effective response. Collaborating with the Department of Emergency Communications (DEC) in Nashville, we analyzed 11,796 non-emergency call recordings and developed Auto311, the first automated system to handle 311 non-emergency calls, which (1) effectively and dynamically predicts ongoing non-emergency incident types to generate tailored case reports during the call; (2) itemizes essential information from dialogue contexts to complete the generated reports; and (3) strategically structures system-caller dialogues with optimized confidence. We used real-world data to evaluate the system's effectiveness and deployability. The experimental results indicate that the system effectively predicts incident type with an average F-1 score of 92.54%. Moreover, the system successfully itemizes critical information from relevant contexts to complete reports, evincing a 0.93 average consistency score compared to the ground truth. Additionally, emulations demonstrate that the system effectively decreases conversation turns as the utterance size gets more extensive and categorizes the ongoing call with 94.49% mean accuracy.

Zirong Chen, Hongchao Zhang, Meiyi Ma

9-1-1 call-taking training requires mastery of over a thousand interdependent skills, covering diverse incident types and protocol-specific nuances. A nationwide labor shortage is already straining training capacity, but effective instruction still demands that trainers tailor objectives to each trainee's evolving competencies. This personalization burden is one that current practice cannot scale. Partnering with Metro Nashville Department of Emergency Communications (MNDEC), we propose PACE (Personalized Adaptive Curriculum Engine), a co-pilot system that augments trainer decision-making by (1) maintaining probabilistic beliefs over trainee skill states, (2) modeling individual learning and forgetting dynamics, and (3) recommending training scenarios that balance acquisition of new competencies with retention of existing ones. PACE propagates evidence over a structured skill graph to accelerate diagnostic coverage and applies contextual bandits to select scenarios that target gaps the trainee is prepared to address. Empirical results show that PACE achieves 19.50% faster time-to-competence and 10.95% higher terminal mastery compared to state-of-the-art frameworks. Co-pilot studies with practicing training officers further demonstrate a 95.45% alignment rate between PACE's and experts' pedagogical judgments on real-world cases. Under estimation, PACE cuts turnaround time to merely 34 seconds from 11.58 minutes, up to 95.08% reduction.

Zhen Hao Wong, Jingwen Deng, Runming He et al.

Large language models (LLMs) excel at many supervised tasks but often struggle with structured reasoning in unfamiliar settings. This discrepancy suggests that standard fine-tuning pipelines may instill narrow, domain-specific heuristics rather than fostering general-purpose thinking strategies. In this work, we propose a "play to learn" framework that fine-tunes LLMs through reinforcement learning on a suite of seven custom logic puzzles, each designed to cultivate distinct reasoning skills such as constraint propagation, spatial consistency, and symbolic deduction. Using a reinforcement learning setup with verifiable rewards, models receive binary feedback based on puzzle correctness, encouraging iterative, hypothesis-driven problem solving. We demonstrate that this training approach significantly improves out-of-distribution performance on a range of mathematical benchmarks, especially for mid-difficulty problems that require multi-step reasoning. Analyses across problem categories and difficulty levels reveal that puzzle training promotes transferable reasoning routines, strengthening algebraic manipulation, geometric inference, and combinatorial logic, while offering limited gains on rote or highly specialized tasks. These findings show that reinforcement learning over logic puzzles reshapes the internal reasoning of LLMs, enabling more robust and compositional generalization without relying on task-specific symbolic tools.

Dezhi Peng, Zikai Sun, Zirong Chen et al.

This paper presents a method that can accurately detect heads especially small heads under the indoor scene. To achieve this, we propose a novel method, Feature Refine Net (FRN), and a cascaded multi-scale architecture. FRN exploits the multi-scale hierarchical features created by deep convolutional neural networks. The proposed channel weighting method enables FRN to make use of features alternatively and effectively. To improve the performance of small head detection, we propose a cascaded multi-scale architecture which has two detectors. One called global detector is responsible for detecting large objects and acquiring the global distribution information. The other called local detector is designed for small objects detection and makes use of the information provided by global detector. Due to the lack of head detection datasets, we have collected and labeled a new large dataset named SCUT-HEAD which includes 4405 images with 111251 heads annotated. Experiments show that our method has achieved state-of-the-art performance on SCUT-HEAD.