About

Yiping Lu is an Assistant Professor of Industrial Engineering and Management Sciences at Northwestern University, with courtesy appointments in Engineering Sciences and Applied Mathematics. He holds a Ph.D. in Applied and Computational Mathematics from Stanford University and a B.S. in Computational Mathematics from Peking University. His research focuses on scaling laws in machine learning, aiming to understand when, why, and how machine-learning systems improve predictably as resources such as data, model size, and computation are increased. Lu investigates the reliable principles of scaling, particularly in large language models, and seeks to develop a general theory and algorithmic framework for scalable learning that ensures increased resources lead to better performance. His work addresses fundamental questions about the limits of model scaling, the geometric aspects of optimization as models grow larger, and inference-time scaling techniques that improve model performance by allocating additional computation after training. Lu's research aims to move beyond empirical observations, providing a principled understanding of optimization geometry, statistical complexity, and resource allocation in machine learning systems. His contributions include studying the interactions of approximation error, optimization difficulty, and statistical uncertainty, as well as developing scaling-aware optimization methods and inference strategies that enhance the reliability and predictability of machine learning performance at scale.

Research topics

• Computer Science
• Materials science
• Artificial Intelligence
• Engineering
• Mathematics
• Algorithm
• Geometry
• Mathematical optimization
• Mechanical engineering
• Structural engineering

Selected publications

• An Assisted Numerical Simulation Diagnosis Method for Atherosclerosis Based on Hemodynamics

  Applied Sciences · 2025-04-07

  articleOpen access

  The mechanism of atherosclerosis lesions was investigated based on a fluid–structure interaction method according to the geometric reconstruction of human arteries by medical imaging. Numerical simulation, mechanical analysis, and dynamic simulation were used to establish a model of the mechanical characteristics of arterial blood transport, analyze the fluid properties of arterial blood flow under the influence of vascular lesion process, and study the mechanism of arterial lesion formation and development. The results indicated that the clinically important areas of secondary flow were generated at stenosis and bifurcation sites, which were prone to lesions during a cardiac cycle. The low flow rate and shear stress levels of blood in this region led to the adhesion and precipitation of lesion-inducing factors on the intimal tissue, creating a hydrodynamic environment suitable for lesion development. According to the research reported here in, early clinical detection and follow-up of atherosclerosis can be performed by collecting data on wall shear stress and blood flow pressure difference.

  PublisherOA PDFDOI

• Tensor-decomposition-based A Priori Surrogate (TAPS) modeling for ultra large-scale simulations

  Computer Methods in Applied Mechanics and Engineering · 2025-06-09 · 12 citations

  articleOpen access
  PublisherOA PDFDOI

• Bivariate Reliability Evaluation Based on Semi-parametric Method Considering Degradation Dependencies and Random Effect

  2025-11-26

  article1st authorCorresponding

  Accurate reliability evaluation is a crucial prerequisite for the development of maintenance strategies and the reduction of maintenance costs. However, the traditional parametric degradation modeling methods are difficult to address the challenges of unknown, complex and time-varying distribution of degradation increments, and most of the studies have not fully considered the bivariate reliability evaluation with degradation dependencies and unit-to-unit variability. To address above issues, this paper proposes a novel semi-parametric degradation model with degradation dependencies and random effect. Firstly, an improved log-change adaptive kernel density estimation (AKDE) method is developed to address the uncertainty of the degradation increment distribution and the boundary estimation bias. Secondly, the extrapolation prediction of degradation states is constructed based on two-dimensional AKDE. Furthermore, a semi-parametric bivariate reliability evaluation model is established based on the copula function, and the random effects are introduced into the copula function. In addition, a two-stage statistical inference method for unknown parameters estimation is developed. Finally, the wet-heat aging experiment of O-ring seals is used to verify the validity and accuracy of the proposed model. The results show that the proposed method outperforms the existing models. The proposed method provides a new idea for component reliability prediction with complex degradation mechanisms.

  PublisherDOI

• FuXi-$γ$: Efficient Sequential Recommendation with Exponential-Power Temporal Encoder and Diagonal-Sparse Positional Mechanism

  ArXiv.org · 2025-12-14

  preprintOpen accessSenior author

  Sequential recommendation aims to model users' evolving preferences based on their historical interactions. Recent advances leverage Transformer-based architectures to capture global dependencies, but existing methods often suffer from high computational overhead, primarily due to discontinuous memory access in temporal encoding and dense attention over long sequences. To address these limitations, we propose FuXi-$γ$, a novel sequential recommendation framework that improves both effectiveness and efficiency through principled architectural design. FuXi-$γ$ adopts a decoder-only Transformer structure and introduces two key innovations: (1) An exponential-power temporal encoder that encodes relative temporal intervals using a tunable exponential decay function inspired by the Ebbinghaus forgetting curve. This encoder enables flexible modeling of both short-term and long-term preferences while maintaining high efficiency through continuous memory access and pure matrix operations. (2) A diagonal-sparse positional mechanism that prunes low-contribution attention blocks using a diagonal-sliding strategy guided by the persymmetry of Toeplitz matrix. Extensive experiments on four real-world datasets demonstrate that FuXi-$γ$ achieves state-of-the-art performance in recommendation quality, while accelerating training by up to 4.74$\times$ and inference by up to 6.18$\times$, making it a practical and scalable solution for long-sequence recommendation. Our code is available at https://github.com/Yeedzhi/FuXi-gamma.

  PublisherOA PDFDOI

• Understanding the Effect of Build Direction and Scanning Strategy on the Tensile Response of Additively Manufactured in 625 with Innovative Calibration Strategy

  SSRN Electronic Journal · 2025-01-01

  preprintOpen access
  PublisherDOI

• Hybrid Encoder–Decoder Model for Ultra-Short-Term Prediction of Wind Farm Power

  Smart Energy System Research · 2025-01-01 · 1 citations

  articleOpen access

  The ability to ensure safe and economic operation of power grids is challenging because of the large-scale integration of wind power as a result of its intermittent and fluctuating nature. Accurate wind power prediction is critical to overcome these concerns. This study proposed a novel hybrid encoder–decoder model by combining bidirectional gated recurrent unit, multi-head attention mechanism, and ensemble technique for multi-step ultra-short-term power prediction of wind farms. The bidirectional gated recurrent unit accurately details the complex temporal dependency of input sequence information in the encoder and outputs the encoded vector. To focus on features that contribute more to the output, two types of multi-head attention mechanism, including self-attention and cross-attention, were used in the decoder to decode the encoded vector and obtain the forecast wind power sequence. Furthermore, an ensemble technique was used to integrate forecast results from various individual predictors, which reduced the uncertainty of individual prediction results and improved predictive accuracy. The input data included historical information from the wind farm and future information from numerical weather prediction. The forecast model was validated using actual data, and results showed that it achieved superior accuracy and stability compared with other existing models in four multi-step prediction scenarios (1-, 2-, 3-, and 4-h prediction).

  PublisherOA PDFDOI

• Aggressive Infection by K1/ST1265 Klebsiella pneumoniae Leading to Multiple Abscesses: Case Report and Literature Review

  Infection and Drug Resistance · 2025-01-01 · 3 citations

  articleOpen access

  Abstract: Hypervirulent Klebsiella pneumoniae (hvKp) has attracted increasing attention in recent years. Diabetes and serotype K1 or K2 are risk factors for invasive liver abscess syndrome including liver abscesses and the metastatic complications such as bacteremia, meningitis, endophthalmitis, and necrotizing fasciitis. Simultaneous infections of the liver, lungs, prostate, brain, and eyes are exceedingly rare. In this paper, a 41-year-old male patient who presented with a 4-day history of fever with polydipsia and polyuria and untreated diabetes deteriorated dramatically with sepsis, prostate abscess, lung abscess, liver abscess and intracranial infection as well as endophthalmitis. He was diagnosed with infection by K1/ST1265 hypervirulent Klebsiella pneumoniae and after treatment with antibiotics and abscess drainage, while the patient still passed away. K1/ST1265 hvKp exhibits exceptionally high virulence and invasiveness, necessitating broad awareness and vigilant monitoring. Keywords: Klebsiella pneumoniae , liver abscess, prostate abscess, brain abscess, serotype K1, sequence type 1265

  PublisherOA PDFDOI

• Predictive value of NT-pro BNP on outcomes of children with ventricular septal defect surgery

  Frontiers in Cardiovascular Medicine · 2025-01-15 · 1 citations

  articleOpen access

  Background Limited study has shown whether NT-proBNP is related to the prognosis of children wth ventricular septal defect (VSD) surgery. The study was conducted to determine the predictive value of NT-proBNP on outcomes of children with VSD surgery. Methods A total of 798 children with VSD surgery were enrolled, with NT-proBNP measured at preoperatively and 24-h postoperatively. The short- and mid-term clinical outcomes were recorded. Propensity scores (PS) was performed to acquire pre-op and post NT-proBNP 24-h PS-matched cohorts for comparisons between groups. Results In the pre NT-proBNP PS-matched cohort, the higher NT pro-BNP group had longer hospitalization time and lower post-op 1-month EF value compared with low NT pro-BNP group (all P < 0.05), and there wasn't significant difference of mechanical ventilation time, cardiopulmonary bypass (CPB) time, intensive care unit (CCU) stay, and ejection fraction (EF) values of 3 month to 12 months after surgery (all P > 0.05). In the post NT-proBNP PS-matched cohort, there wasn't significant difference of mechanical ventilation time, CPB time, CCU stay, hospitalization time, and EF values of 1 month to 12 months after surgery between two groups (all P > 0.05). Conclusions VSD children with higher pre NT-proBNP level had longer hospital stays after surgery than those with lower level. Pre NT-proBNP level had no effect on mechanical ventilation time, CPB time, ACC time and CCU stay and cardiac function after 3 months postoperatively. Post-op 24-h NT pro-BNP level wasn't associated with clinical outcomes.

  PublisherDOI

• Multi-objective bi-level programming algorithm to design transition mechanism for wind farm participation in electricity markets

  International Journal of Electrical Power & Energy Systems · 2025-08-01

  articleOpen access

  • Government-driven MLT contract for wind farms, including pricing, settlement, and integration with green certificate market. • Bi-level model: upper level optimizes MLT coverage; lower level conducts stochastic clearing of energy and reserve markets. • Adaptive weighted-sum algorithm applied to generate a uniformly distributed set of Pareto-optimal solutions. • Validation of the proposed mechanism under two MLT forms: energy contracts and time-based trades. To address the participation of wind farms with different construction costs in the electricity market, designing a reasonable transition mechanism is crucial. A single leader multi-follower bi-level programming model is proposed. The upper-level employs multi-objective optimization to determine the medium- and long-term contract coverages of subsidized and unsubsidized wind farms to minimize the subsidy expenditures and revenue disparity. The lower level model jointly clears day-ahead and real-time energy and reserve markets considering the uncertainty of wind farm’s outputs via stochastic optimization problem to maximize social welfare. The bi-level model is transformed into a multi-objective mixed-integer nonlinear programming model using the Karush-Kuhn-Tucker condition and the big M method. Linearization strategies are proposed to handle the product term of price and continuous variables and the absolute value term in the upper-level objective. The adaptive weighted-sum algorithm is adopted to obtain uniformly distributed Pareto optimal solutions. Simulations on a 44-unit 1560-bus system under two forms of medium- and long-term trading are carried out to verify the effectiveness of the proposed method. Compared to the constraint method, the adaptive weighted sum algorithm reduces the maximum-minimum solution distance difference by 42.356 %, average distance by 5.2%, and standard deviation by 3.499 %. The Pareto solution closest to the utopia point reduces subsidies by 66.67% and profit disparity by 55.82 %. By optimizing contract coverage, government subsidies can be significantly reduced, and the unit profit disparities among wind farms built in different periods can be minimized, facilitating a smooth policy transition.

  PublisherDOI

• Testosterone suppression with or without testicular Preservation: Impact on mouse behavior, including emotional state, locomotor activity, social interaction, cognitive function, and sexual behavior

  Biochemical and Biophysical Research Communications · 2025-06-16 · 1 citations

  articleOpen access

  The deposition of androgens significantly influences livestock meat quality, yet traditional surgical castration methods involving testicular removal raise concerns about animal welfare due to their association with stress and behavioral changes. To address these issues, this study compared two approaches—surgical castration (testicular removal) and immunocastration (retention of testicular tissue)—to evaluate their psychological and behavioral impacts in mice across domains such as sexual behavior, emotional regulation, locomotor activity, social interaction, and memory performance. Results demonstrated that surgical castration induced robust depressive-like symptoms, including reduced motivation and impaired social engagement, whereas immunocastrated mice exhibited normal emotional and social behaviors without signs of depression or distress. These findings highlight the psychological burden of surgical castration and underscore the ethical implications of its use. Immunocastration emerges as a more humane alternative, effectively controlling sexual behavior without inducing depressive-like symptoms, likely by preserving the animals' sense of normalcy and dignity. This study emphasizes the importance of balancing productivity with ethical considerations in livestock management and provides valuable insights for advancing welfare-friendly castration methods. By advocating for immunocastration, this research contributes to improving animal welfare standards while maintaining meat quality, offering a promising alternative to traditional practices. • 1.The androgen blockade model affects mouse behavior and psychological state. • 2.Surgical castration leads to depressive-like symptoms, including reduced motivation and impaired social behavior. • Immunocastration offers a more humane alternative to traditional surgical castration, minimizing psychological harm. • Immunocastrated mice maintain normal emotional, social, and cognitive functions without signs of distress. • 5.Immunocastration improves welfare while preserving meat quality, offering a viable alternative to conventional castration practices.

  PublisherDOI

Frequent coauthors

• Anthony Gravouil

  40 shared

• Nawfal Blal

  Institut National des Sciences Appliquées de Lyon

  40 shared

• Wing Kam Liu

  20 shared

• Ted Belytschko

  17 shared

• Satyajit Mojumder

  13 shared

• Abdullah Al Amin

  Northwestern University

  10 shared

• Benoı̂t Bary

  Commissariat à l'Énergie Atomique et aux Énergies Alternatives

  9 shared

• Gregory J. Wagner

  Northwestern University

  9 shared

Education

• PhD, Mechanical Engineering

  Institut National des Sciences Appliquées de Lyon

  2017

Awards & honors

• 40th Conference on Uncertainty in Artificial Intelligence, 2…
• International Conference on Learning Representations (ICLR)…
• Thirty-seventh Conference on Neural Information Processing S…
• Eleventh International Conference on Learning Representation…