About

Zhaoran Wang is an Associate Professor of Industrial Engineering and Management Sciences at Northwestern University, with a courtesy appointment in Computer Science. His research focuses on developing a new generation of data-driven decision-making methods, theories, and systems that leverage artificial intelligence to address pressing societal challenges. His work aims to make autonomous learning agents more efficient both computationally and statistically, enabling their application in critical domains. Additionally, he is dedicated to scaling autonomous learning agents to design and optimize societal-scale multi-agent systems involving cooperation and competition among humans and robots. His research interests span across machine learning, optimization, statistics, game theory, and information theory. Wang has contributed to advancing the understanding and development of reinforcement learning, representation learning, and control systems, with a particular emphasis on provable efficiency and sample complexity. His work has been published in leading conferences such as ICML, NeurIPS, ICLR, and COLT, reflecting his active engagement in cutting-edge research in artificial intelligence and decision-making systems.

Research topics

• Computer Science
• Artificial Intelligence
• Machine Learning
• Information Retrieval
• Mathematics
• Combinatorics
• Mathematical optimization
• Statistics
• Discrete mathematics
• Engineering
• Psychology
• Econometrics

Selected publications

• An Investigation into the Application of Gamification and Socialization Design in International Chinese Language Educational Software

  Innovation Series Advanced Science · 2026-01-01

  articleOpen access1st authorCorresponding

  This paper focuses on the application of gamification and socialization design in international Chinese educational software.First of all, we adopt the Literature Review Method to sort out the principle of interest motivation, Behaviorism Learning Theory, Maslow's hierarchy of needs theory, self-efficacy theory and input-output hypothesis, and introduce the theoretical basis of gamification and socialization design and the necessity of integration of the two.Next, we analyze the typical cases in depth to reveal the problems that exist in the current products.The problems of the current products are revealed through in-depth analyses of typical cases.On this basis, the design concept of the deep integration of gamification and socialization is proposed.In the end, it is stressed that the deep integration of gamification and socialization is the key path to solve the current difficulties of international Chinese language education software, but it is also necessary to return to the essence of learning, and naturally integrate with the knowledge, in order to truly stimulate the motivation of learning.

  PublisherOA PDFDOI

• Evidentially Driven Uncertainty Decomposition for Weakly Supervised Point Cloud Semantic Segmentation

  ISPRS International Journal of Geo-Information · 2026-04-12

  articleOpen access

  Point cloud semantic segmentation is a core component in indoor scene understanding and autonomous driving. Under weak point-level supervision, only a small subset of points is annotated, making effective use of unlabeled points critical yet non-trivial. Many existing approaches rely on prediction confidence to filter pseudo labels or enforce consistency, which can bias training toward easy points and amplify early mistakes. Consequently, confidently wrong predictions may be reinforced, while uncertain points around class boundaries or in geometrically complex regions are less utilized, limiting further gains. An evidential uncertainty decomposition framework is introduced for weakly supervised point cloud semantic segmentation. Network outputs are interpreted as evidential distributions, and uncertainty is decomposed to separate lack-of-knowledge uncertainty from boundary-related ambiguity, providing a more informative reliability signal for unlabeled points. Based on this signal, different constraints are applied to different subsets: reliable points are trained with pseudo labels together with prototype-based regularization to encourage intra-class compactness; boundary-ambiguous points are guided by evidential consistency to improve boundary learning; and points with high epistemic uncertainty are excluded from pseudo-label-based supervision to mitigate error reinforcement. In addition, an uncertainty calibration term on sparsely labeled points helps stabilize training. Experiments on S3DIS, ScanNet-V2, and SemanticKITTI yield 67.7%, 59.7%, and 53.3% mIoU, respectively, with only 0.1% labeled points, comparing favorably with prior weakly supervised point cloud segmentation methods.

  PublisherDOI

• How Does Dual Openness Asymmetrically Shape Bank Systemic Risk? Bringing In versus Going Global in China's Banking Industry

  SSRN Electronic Journal · 2026-01-01

  preprintOpen access
  PublisherDOI

• Digital Evocation in the Posthuman: The Hauntological Production of Video Games as Cultural Industry

  GBP Proceedings Series · 2026-05-02

  article1st authorCorresponding

  Since the emergence of the posthuman condition, digital technologies and artificial intelligence have profoundly restructured contemporary cultural industries. These developments have not only transformed modes of production but also engendered a novel production logic predicated on the permanence of data storage and the virtualization of material production. Within this context, video games, by virtue of their intrinsic capacity for data persistence, algorithmic modulation, and material virtualization, have evolved from paradigmatic expressions of digital culture into distinct ontological sites of production. Their outputs are not discrete physical objects, but rather pervasive, durative, and hauntological presences: spectral formations that linger across temporal, spatial, and ontological registers and can be endlessly updated, reconfigured, and replayed. This hauntological condition, suspended between the virtual and the real and operating at the interstices of human agency and machine operation, disrupts the materially grounded epistemological boundaries that have long defined traditional cultural industries. In doing so, it compels a fundamental reconfiguration of how we conceptualize memory, history, and identity in the posthuman era, particularly as archives, narratives, and subjectivities become increasingly encoded in software and networked infrastructures. From the dual vantage points of digital cultural transformation and hauntological theory, it is therefore imperative to examine the medium-specific affordances of video games as a pivotal sector within contemporary cultural industries and to clarify their distinctive production logic and broader implications.

  PublisherDOI

• Multimodal large language models in brain tumor imaging: clinical applications and future perspectives

  Die Radiologie · 2026-04-14

  article1st author
  PublisherDOI

• An underlying mechanism of bovine mastitis: PGE ₂ regulates <i>Staphylococcus aureus</i> -induced inflammatory response through TLR2, TLR4, and NLRP3 in macrophages

  Veterinary Quarterly · 2026-01-17 · 2 citations

  articleOpen access

  -induced mastitis, suggesting that targeting this interaction may provide novel therapeutic strategies.

  PublisherDOI

• Energy-Efficient and Perturbation-Aware Dwell-Recharge Integrated Strategy With Deep Reinforcement Learning for Catenary-Free Tramway

  IEEE Transactions on Intelligent Transportation Systems · 2026-01-01

  article1st authorCorresponding

  Mounting global energy challenges necessitate a transition toward green and lean operational strategies across contemporary industries to bring eco-economic benefits about. This imperative extends to transportation systems, where emerging catenary-free tramways with novel onboard-offboard power supply architectures exhibit sustainable mobility while enhancing urban aesthetics. However, such systems still face critical challenges: simultaneous power demands during recharging cycles at stations risk destabilizing the tram traction power network, while mixed-traffic urban environments introduce operational vulnerabilities, manifesting as delays and congestion due to shared rights-of-way. To address these challenges, this study presents an integrated dwell time regulation and recharging scheduling method for catenary-free tramway, leveraging deep reinforcement learning (DRL) to balance dynamic energy demands with operational efficiency. The system dynamics are formalized through discrete event simulation (DES), and the decision-making process is formulated as an event-driven Markov decision process (MDP) to optimize real-time actions. The case study on a real-world catenary-free tramline in China demonstrates that our method can effectively diminish the peak power superimposition on local power network and the energy costs. Compared with representative heuristic and online optimization methods, DRL approach delivers a captivating solution for agile decision-making of intelligent tramway in the dynamic urban environments.

  PublisherDOI

• Estimating attributable risk functions for censored time-to-event in disease prevention research

  Lifetime Data Analysis · 2026-03-27

  article
  PublisherDOI

• Learning to Reason as Action Abstractions with Scalable Mid-Training RL

  ArXiv.org · 2025-09-30

  preprintOpen access

  Large language models excel with reinforcement learning (RL), but fully unlocking this potential requires a mid-training stage. An effective mid-training phase should identify a compact set of useful actions and enable fast selection among them through online RL. We formalize this intuition by presenting the first theoretical result on how mid-training shapes post-training: it characterizes an action subspace that minimizes both the value approximation error from pruning and the RL error during subsequent planning. Our analysis reveals two key determinants of mid-training effectiveness: pruning efficiency, which shapes the prior of the initial RL policy, and its impact on RL convergence, which governs the extent to which that policy can be improved via online interactions. These results suggest that mid-training is most effective when the decision space is compact and the effective horizon is short, highlighting the importance of operating in the space of action abstractions rather than primitive actions. Building on these insights, we propose Reasoning as Action Abstractions (RA3), a scalable mid-training algorithm. Specifically, we derive a sequential variational lower bound and optimize it by iteratively discovering temporally-consistent latent structures via RL, followed by fine-tuning on the bootstrapped data. Experiments on code generation tasks demonstrate the effectiveness of our approach. Across multiple base models, RA3 improves the average performance on HumanEval and MBPP by 8 and 4 points over the base model and the next-token prediction baseline. Furthermore, RA3 achieves faster convergence and higher asymptotic performance in RLVR on HumanEval+, MBPP+, LiveCodeBench, and Codeforces.

  PublisherOA PDFDOI

• The Application and Challenges of Artificial Intelligence in Contemporary Art Curation

  Journal of Education Humanities and Social Sciences · 2025-12-09 · 1 citations

  articleOpen access

  The integration of Artificial Intelligence (AI) into contemporary art curation is revolutionizing how exhibitions are designed, artworks are selected, and audiences interact with art. Through technologies such as machine learning, natural language processing (NLP), and computer vision, AI enables curators to analyze vast collections, predict visitor preferences, and construct dynamic exhibition experiences. This paper examines the dual nature of AI in art curation—its potential to enhance creativity and efficiency, and the ethical, technical, and philosophical challenges it introduces. Drawing upon case studies from global museums and digital art platforms, the research identifies key application domains, evaluates curatorial outcomes, and analyzes stakeholders’ perceptions of AI-assisted practices. Quantitative data from 30 institutions are supplemented by qualitative analysis to explore how AI reshapes curatorial authority, audience engagement, and the meaning-making process. Findings reveal that while AI fosters innovation and accessibility, it also risks algorithmic bias, cultural homogenization, and diminished human interpretive agency. The paper concludes that sustainable AI curation requires a human–machine collaborative framework emphasizing transparency, inclusivity, and cultural sensitivity.

  PublisherOA PDFDOI

Frequent coauthors

• Zhuoran Yang

  125 shared

• David M. Blei

  39 shared

• Daniel Shu Wei Ting

  Duke-NUS Medical School

  31 shared

• Tien Yin Wong

  Singapore National Eye Center

  28 shared

• Gilbert Lim

  Singapore Eye Research Institute

  22 shared

• Michael I. Jordan

  22 shared

• Lingxiao Wang

  Beijing University of Chinese Medicine

  21 shared

• Huiguang He

  Institute of Automation

  16 shared