Article
Mathematical & Computational Biology
Qing Li, Jianchang Lin, Mengya Liu, Liwen Wu, Yingying Liu
Summary: This article proposes using a surrogate endpoint to improve conditional power calculation in adaptive designs, demonstrating practical feasibility and benefits for studies with delayed treatment effects. The method shows significantly higher overall power compared to conventional approaches, particularly in cases of delayed treatment effects in primary survival endpoints. The approach is also demonstrated in a Phase III non-small cell lung cancer trial, providing recommendations for implementation in confirmatory clinical trials.
STATISTICS IN BIOPHARMACEUTICAL RESEARCH
(2022)
Article
Mathematical & Computational Biology
Roland G. G. Gera, Tim Friede
Summary: The increasing interest in subpopulation analysis in personalized medicine and targeted therapies has led to the development of various new trial designs and analysis methods. This paper proposes a trial design applicable to any set of composite populations and considers normally distributed endpoints and random baseline covariates. The study uses p-values calculated on subset levels and the inverse normal combination function to test treatment effects for composite populations, while also accounting for multiple testing using the closed testing procedure. The paper also derives critical boundaries for intersection hypothesis tests and provides simulations demonstrating the absence of practical relevant inflation of the type I error rate. The target power after sample size recalculation is typically met or close to being met.
BIOMETRICAL JOURNAL
(2023)
Article
Computer Science, Interdisciplinary Applications
Jixiang Qing, Nicolas Knudde, Federico Garbuglia, Domenico Spina, Ivo Couckuyt, Tom Dhaene
Summary: This paper presents a new design space exploration methodology using Gaussian process metamodels and adaptive sampling techniques to discover feasible regions in the design space and address multiple constraints problems. The proposed approach is compared with state-of-art techniques on benchmark problems and practical engineering examples to evaluate its efficiency, accuracy, and robustness.
ENGINEERING WITH COMPUTERS
(2022)
Article
Medicine, General & Internal
Bander Ali Saleh Al-rimy, Faisal Saeed, Mohammed Al-Sarem, Abdullah M. Albarrak, Sultan Noman Qasem
Summary: In this research, DenseNet169, a deep convolutional neural network architecture, was investigated for knee osteoarthritis detection using X-ray images. An adaptive early stopping technique and gradual cross-entropy loss estimation were proposed to prevent overfitting and optimize the model. The proposed model outperformed existing solutions in terms of accuracy, precision, recall, and loss performance.
Article
Mathematical & Computational Biology
Qi Zhang, Yuqian Shen, Hui Quan, Pascal Minini, Lin Wang
Summary: Due to the need to accelerate the drug development process in rare disease areas, a two-stage adaptive design option is being evaluated for a placebo-controlled registration study. The study involves participants from an ongoing phase 2 study for stage 1 and newly enrolled participants for stage 2, with different treatment periods. The primary endpoint is the rate of change for a continuous measurement, which will be evaluated using a mixed model. Interim analyses will be conducted to adjust sample size and assess early efficacy.
STATISTICS IN BIOPHARMACEUTICAL RESEARCH
(2023)
Article
Engineering, Electrical & Electronic
Qinghua Hu, Yucong Gao, Bing Cao
Summary: Modern AI systems often suffer from catastrophic forgetting in real-world applications. Previous methods using knowledge distillation and bias correction are not flexible enough, leading to forgetting of crucial knowledge. This study proposes a curiosity-driven class-incremental learning approach that selectively learns informative samples, effectively preventing catastrophic forgetting.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Yan Li, Chang Liu, Suyun Zhao, Qiang Hua
Summary: Partial label learning is a weak supervised learning method that uses samples with candidate label sets to train a classifier. This paper proposes a partial label learning method based on active learning mechanism, which uses a small number of partially labeled samples and a large number of unlabeled samples to construct an effective classifier. Experimental results show that the proposed method achieves higher classification accuracy than representative similar methods and requires labeling only a small number of samples for stable performance.
INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS
(2022)
Article
Medicine, Research & Experimental
Ruitao Lin, Zhao Yang, Ying Yuan, Guosheng Yin
Summary: The heterogeneity of clinical trial participants poses a fundamental challenge in the field of precision medicine, but adaptive enrichment designs offer a flexible and intuitive solution. By enriching the subgroup of trial participants with a higher likelihood of benefit from a new treatment, these designs can control type I error rate and improve statistical power and expected sample size.
CONTEMPORARY CLINICAL TRIALS
(2021)
Article
Management
Zhenxia Cheng, Jun Luo, Ruijing Wu
Summary: We study the simulation optimization problem of selecting the best system design, known as ranking and selection (R&S). We propose fully sequential procedures that incorporate adaptive sampling rules while preserving finite-sample statistical guarantees. Specifically, we introduce an adaptive sampling rule that utilizes consecutively updated sample mean and variance information by solving a minimization problem of the approximated total sample size. Extensive simulation experiments demonstrate the efficiency of the proposed procedures, and we apply them to solve an ambulance dispatching problem.
EUROPEAN JOURNAL OF OPERATIONAL RESEARCH
(2023)
Article
Mathematics, Applied
Pranab Jyoti Deka, Lukas Einkemmer
Summary: Traditional step-size controllers are not effective in exponential integrators as they assume that the cost of a time step is independent of the step size. This manuscript presents an adaptive step-size controller for exponential Rosenbrock methods that determines the step size based on minimizing computational cost. Experimental results show significant improvements in computational cost compared to traditional step-size controllers for a range of nonlinear partial differential equations.
COMPUTERS & MATHEMATICS WITH APPLICATIONS
(2022)
Article
Chemistry, Physical
Bowen Lei, Tanner Quinn Kirk, Anirban Bhattacharya, Debdeep Pati, Xiaoning Qian, Raymundo Arroyave, Bani K. Mallick
Summary: Bayesian optimization is essential for optimizing objective functions without known functional forms or expensive evaluations. Optimal experimental design within BO workflow leads to more efficient exploration of design space. Using adaptive and flexible Bayesian surrogate models enhances search efficiency and robustness in experimental design.
NPJ COMPUTATIONAL MATERIALS
(2021)
Article
Ecology
Garrett M. Street, Jonathan R. Potts, Luca Borger, James C. Beasley, Stephen Demarais, John M. Fryxell, Philip D. McLoughlin, Kevin L. Monteith, Christina M. Prokopenko, Miltinho C. Ribeiro, Arthur R. Rodgers, Bronson K. Strickland, Floris M. van Beest, David A. Bernasconi, Larissa T. Beumer, Guha Dharmarajan, Samantha P. Dwinnell, David A. Keiter, Alexine Keuroghlian, Levi J. Newediuk, Julia Emi F. Oshima, Olin Rhodes, Peter E. Schlichting, Niels M. Schmidt, Eric Vander Wal
Summary: Sample size sufficiency is crucial for estimating resource selection functions (RSFs) from GPS-based animal telemetry, with thresholds such as M >= 30 captured animals and maximum relocations per animal N recommended. This study provides a comprehensive solution by deriving mathematical expressions for necessary M and N based on biologically meaningful quantities, showing the decline in required sample sizes with increasing selection strength and landscape complexity. Analytical solutions demonstrate that the most relevant effects on utilization distribution can often be estimated with fewer than M=30 animals, regardless of availability definition, and should be a mandatory component for all future RSF studies.
METHODS IN ECOLOGY AND EVOLUTION
(2021)
Article
Computer Science, Interdisciplinary Applications
Leonardo Galvis, Tim Offermans, Carlo G. Bertinetto, Andrea Carnoli, Emina Karamujic, Weiwei Li, Ewa Szymanska, Lutgarde M. C. Buydens, Jeroen J. Jansen
Summary: Quality by Design (QbD) is a popular approach for designing and optimizing industrial production facilities. In retrospective QbD (rQbD) studies, historical production data are used for optimization. However, current rQbD literature has limited discussion on the identification of critical process parameters (CPPs) and does not cover situations with limited process knowledge or the use of parallel equipment. This work presents a new rQbD strategy that balances statistical analysis of historical data with plant-scale experiments and addresses these challenges.
COMPUTERS IN INDUSTRY
(2022)
Article
Engineering, Multidisciplinary
Serdar Carbas, Osman Tunca
Summary: Real size complex steel space frame structures are numerically designed to achieve optimal design weight using a rectified cuckoo search algorithm with adaptive method and boosted with greedy selection for better algorithmic performance.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
(2023)
Article
Multidisciplinary Sciences
Katharina T. Schmid, Barbara Hoellbacher, Cristiana Cruceanu, Anika Boettcher, Heiko Lickert, Elisabeth B. Binder, Fabian J. Theis, Matthias Heinig
Summary: The authors present a statistical framework for informed multi-sample experimental design in scRNASeq data to reduce unnecessary costs and maximize data utility. The study shows that shallow sequencing of high numbers of cells leads to higher overall power than deep sequencing of fewer cells. The model is implemented as an R package and accessible as a web tool.
NATURE COMMUNICATIONS
(2021)