About

Haoqi Zhang is an associate professor in Computer Science and Design at Northwestern University, with a courtesy appointment in Learning Sciences. His research advances the design of integrated socio-technical models that address complex problems and promote human values. His work bridges multiple disciplines including Computer Science, Design, Learning Science, Psychology, and Philosophy, and is supported by the National Science Foundation and the Center for Advancing Safety in Machine Intelligence. Zhang received his PhD in Computer Science and a BA in Computer Science and Economics from Harvard University. At Northwestern, he founded and directs the Design, Technology, and Research (DTR) program, which offers an innovative model for research-based learning and growth for over 180 students. Additionally, he co-directs the Delta Lab, an interdisciplinary research lab and design studio that spans computer science, learning sciences, and design, working on projects that explore complex socio-technical systems and human-centered design.

Research topics

• Natural Language Processing
• Computer Science
• Information Retrieval
• Artificial Intelligence
• World Wide Web
• General surgery
• Internal medicine
• Surgery
• Medicine

Selected publications

• Multi-Token Residual Prediction

  arXiv (Cornell University) · 2026-05-12

  preprintOpen access

  Diffusion Language Models (DLMs) generate text by iteratively denoising masked token sequences, offering a tradeoff between parallelism and quality compared to autoregressive models. In current practice, the number of tokens decoded per step is controlled by a confidence threshold, and quality degrades monotonically as more tokens are denoised per step. We introduce Multi-token Residual Prediction (MRP), a lightweight module that enables dependency-aware multi-token denoising within a single backbone forward pass. MRP exploits a key property of the denoising process: the logit distributions at adjacent denoising steps are remarkably similar. Rather than running the backbone a second time to obtain the next-step logits, MRP predicts the residual between steps from the backbone's hidden states, effectively denoising more tokens per backbone forward at a fraction of the cost. We deploy MRP in two inference modes: direct decoding, which uses the corrected logits without verification for a tunable quality--speed tradeoff; and speculative decoding, which verifies MRP's proposals against the backbone for lossless acceleration. Experiments on SDAR models at the 1.7B, 4B, and 8B scales across reasoning and code generation benchmarks demonstrate up to $1.42\times$ lossless speedup in SGLang.

  PublisherDOI

• Multi-Token Residual Prediction

  ArXiv.org · 2026-05-12

  articleOpen access

  Diffusion Language Models (DLMs) generate text by iteratively denoising masked token sequences, offering a tradeoff between parallelism and quality compared to autoregressive models. In current practice, the number of tokens decoded per step is controlled by a confidence threshold, and quality degrades monotonically as more tokens are denoised per step. We introduce Multi-token Residual Prediction (MRP), a lightweight module that enables dependency-aware multi-token denoising within a single backbone forward pass. MRP exploits a key property of the denoising process: the logit distributions at adjacent denoising steps are remarkably similar. Rather than running the backbone a second time to obtain the next-step logits, MRP predicts the residual between steps from the backbone's hidden states, effectively denoising more tokens per backbone forward at a fraction of the cost. We deploy MRP in two inference modes: direct decoding, which uses the corrected logits without verification for a tunable quality--speed tradeoff; and speculative decoding, which verifies MRP's proposals against the backbone for lossless acceleration. Experiments on SDAR models at the 1.7B, 4B, and 8B scales across reasoning and code generation benchmarks demonstrate up to $1.42\times$ lossless speedup in SGLang.

  PublisherOA PDF

• YAP promotes fibrosis by regulating macrophage to myofibroblast transdifferentiation and M2 polarization in chronic pancreatitis

  International Immunopharmacology · 2025-01-15 · 7 citations

  article
  PublisherDOI

• Research on Dynamic Reconfiguration Method of New Distribution System Network to Enhance New Energy Carrying Capacity

  2025-02-28 · 1 citations

  article

  With the continuous development of clean energy, a large number of distributed generation (DG) sources are connected to the distribution network, and the compatibility of the distribution network with renewable distributed energy sources has led to the diversification of optimization objectives. This paper first analyzes the impact of DG connection location and capacity on the distribution network based on a typical distribution network system. Then, a distribution network reconfiguration model is constructed. Considering the reliability and economy of the distribution network, a genetic algorithm is used to optimize the network reconfiguration structure and DG output, achieving a balanced distribution of energy across all branches. Finally, based on the IEEE33-node system, three typical scenarios with different DG connection capacities are considered to verify the effectiveness of the model.

  PublisherDOI

• Ultra-precise sensing in PVDF piezoelectric nanogenerator with self-oriented nanocrystals via multi-level counter-chain relaxation strategy

  Nano Energy · 2025-03-05 · 13 citations

  article
  PublisherDOI

• Noncoding RNAs in Atopic Dermatitis: Insight Into Inflammation and Immune Regulation

  Dermatologic Therapy · 2025-01-01 · 15 citations

  articleOpen access

  Atopic dermatitis (AD) is a chronic inflammatory skin disorder affecting approximately 20% of children and 10% of adults. While previous studies have linked AD to allergen exposure, disruption of the skin barrier, and Type 2 immune responses, the precise pathophysiology of AD remains elusive, significantly limiting the effectiveness of current treatments. Noncoding RNAs (ncRNAs), a diverse group of transcripts that do not encode proteins and account for at least 98% of the human genome, are implicated in numerous physiological and pathological processes. A growing body of evidence underscores the pivotal role of ncRNAs in the pathogenesis and progression of AD. This review offers a detailed synthesis of the latest insights into the involvement of ncRNAs in AD, as well as their potential as diagnostic biomarkers and therapeutic targets.

  PublisherOA PDFDOI

• Maternal obesity impairs fetal brown adipogenesis by attenuating fetal FGF21 signaling

  Cell Reports · 2025-07-01 · 2 citations

  articleOpen access

  In mammals, maternal obesity typically impairs brown adipose tissue (BAT) formation in fetuses, increasing their risk of metabolic disorders in adulthood. However, the mechanisms behind this phenomenon are not well understood. Our single-nucleus transcriptomic analysis revealed dynamic changes in cell heterogeneity within the fetal interscapular BAT (iBAT) from obese dams, leading to compromised thermogenesis in their offspring. Obese dams displayed elevated levels of circulating fibroblast growth factor 21 (FGF21), while their fetuses exhibited lower circulating FGF21 due to reduced trans-placental transfer. Maternal FGF21, significantly increased during late gestation, was the primary source of fetal FGF21, played a crucial role in regulating fetal brown adipogenesis, and likely prevented metabolic dysfunction in offspring. Additionally, the impaired iBAT development in utero due to maternal obesity could be mitigated by postnatal FGF21 supplementation. This study suggests that FGF21 signaling is a promising target for addressing impaired BAT development in fetuses resulting from maternal obesity.

  PublisherDOI

• Liver ERα mediates sex differences in metabolic pattern changes in response to time-restricted feeding

  Life Metabolism · 2025-03-25 · 5 citations

  articleOpen access1st authorCorresponding

  Time-restricted feeding (TRF) is a dietary strategy used to prevent and treat obesity in both sexes. However, TRF affects liver metabolism differently in males and females, and the mechanisms behind these differences remain unclear. Our study reveals that during TRF, female livers are more likely to break down amino acids (AAs) to synthesize fats, while male livers significantly reduce fatty acid synthesis. The changes in the liver's AA metabolic profile after gonadectomy suggest that estrogen signaling is crucial for regulating AA metabolism in females during TRF. Additionally, we demonstrate that hepatic estrogen receptor α (ERα)-mediated AA metabolism contributes to the sex-specific effects of TRF on liver metabolism. These findings offer new insights into the molecular mechanisms of TRF and its potential clinical application for treating fatty liver and other metabolic disorders. They also emphasize the need to consider sex differences when developing nutritional and pharmacological treatments for metabolic diseases in females.

  PublisherOA PDFDOI

• UVB radiation and amphibian resilience: Analyzing skin color, immune suppression and oxidative stress in Rana kukunoris from different elevations

  Ecotoxicology and Environmental Safety · 2025-03-20 · 10 citations

  articleOpen access

  Ultraviolet-B radiation (UVBR), intensified by ozone depletion and climate change, poses a growing ecological threat to amphibians, particularly in high-elevation regions such as the Qinghai-Tibet Plateau. Endemic to this region, Rana kukunoris spans a wide range of elevations, where distinct populations may have evolved unique strategies and regulatory mechanisms to cope with UVBR. However, specific adaptive responses in adult frogs remain underexplored. This study compared the physiological responses of high- and low-altitude Rana kukunoris populations to UVBR exposure, focusing on dorsal color, immune function, antioxidant capacity, and DNA repair gene expression. High-altitude populations exhibited stable, dark pigmentation—potentially reducing the need for rapid melanin synthesis—alongside a robust immune profile and enhanced antioxidant enzyme activity, collectively conferring resilience against oxidative and immune stress under chronic UVBR exposure. Conversely, low-altitude populations exhibited pronounced UVBR-induced responses, including significant skin darkening, heightened immune activation evidenced by increased white blood cell counts, and increased oxidative damage marked by higher malondialdehyde (MDA) levels, coupled with reduced superoxide dismutase (SOD) and catalase (CAT) activities. Furthermore, tissue-specific upregulation of DNA repair genes in high-altitude populations suggested a stable DNA repair capacity adapted to high-UVBR environments. These findings reveal distinct physiological strategies within the same species for coping with UVBR across altitudinal gradients. Amid global increases in UVBR, this study offers novel insights into amphibian resilience in high-UVBR habitats and informs conservation strategies for populations across varying elevations. • High- and low-altitude Rana kukunoris show distinct UVB adaptation strategies on the Qinghai-Tibet Plateau. • High-altitude frogs have stable dark skin, strong immunity, and antioxidant defenses under UVBR. • Low-altitude frogs exhibit UVBR-induced skin darkening, immune activation, and oxidative damage. • High-altitude populations upregulate DNA repair genes, aiding genomic stability in high-UVBR environments. • Study guides conservation strategies for amphibians facing rising UVBR.

  PublisherDOI

• Nomogram to predict late extraluminal postpancreatectomy hemorrhage in patients with postoperative pancreatic fistula after pancreaticoduodenectomy

  Gland Surgery · 2025-03-01 · 1 citations

  articleOpen access

  Background: Late extraluminal postpancreatectomy hemorrhage (LEPPH) is a rare but severe complication of pancreaticoduodenectomy (PD). Current predictors of LEPPH are limited and cannot quantify bleeding risk. As a consequence, establishment of a prediction model of LEPPH is important. This study aims to construct a nomogram combining perioperative factors to predict LEPPH. Methods: A total of 2,924 retrospective and 467 prospective cases undergoing PD, 420 retrospective cases and 131 prospective cases with postoperative pancreatic fistula (POPF) after PD from three centers were included. Three hundred and seventy-one retrospective cases from West China Hospital were divided randomly into the development cohort (n=259) and the internal validation cohort (n=112). Another 180 patients consisting of 49 retrospective and 131 prospective cases from three pancreatic centers were enrolled as the external validation set. A nomogram was established based on the independent risk factors. Results: Multivariable analysis identified pancreaticoenteric anastomotic dorsal fluid accumulation, bubble sign, pancreaticoenteric anastomotic cracking (PEAC), surgery-related acute pancreatitis (AP), and positive culture in intra-abdominal drainage fluid as independent risk factors of LEPPH. Combined with those variables, the nomogram showed reliable C-index of 0.932, 0.924 and 0.954 in predicting LEPPH in the three cohorts respectively. Conclusions: The nomogram exhibited excellent predictive capabilities for LEPPH after PD. It could aid surgeons in early identification of patients prone to LEPPH following PD, enabling timely interventions and improving patient survival.

  PublisherDOI

Recent grants

• CHS: Small: Coordination of Opportunistic Actions to Produce Globally Effective Behaviors for Physical Crowdsourcing

  NSF · $496k · 2016–2021

• EXP: Readily Available Learning Experiences: Turning the Entire Web into Progressive Examples to Bridge Conceptual Knowledge Gaps for Novice Web Developers

  NSF · $550k · 2017–2022

• EXP: Agile Research Studios: Scaling Cognitive Apprenticeship to Advance Undergraduate and Graduate Research Training in STEM

  NSF · $550k · 2016–2021

Frequent coauthors

• David C. Parkes

  14 shared

• Chunlu Tan

  12 shared

• Xubao Liu

  Sichuan University

  12 shared

• Darren Gergle

  Northwestern University

  11 shared

• Robert C. Miller

  Mayo Clinic in Florida

  11 shared

• Steven P. Dow

  University of California, San Diego

  10 shared

• Juho Kim

  Korea Advanced Institute of Science and Technology

  9 shared

• Elizabeth M. Gerber

  8 shared

Labs

• Design, Technology, and Research (DTR) programPI