Worst-case Procedure Research Articles

Distributionally Robust Losses for Latent Covariate Mixtures

Reliable Machine Learning via Structured Distributionally Robust OptimizationData sets used to train machine learning (ML) models often suffer from sampling biases and underrepresent marginalized groups. Standard machine learning models are trained to optimize average performance and perform poorly on tail subpopulations. In “Distributionally Robust Losses for Latent Covariate Mixtures,” John Duchi, Tatsunori Hashimoto, and Hongseok Namkoong formulate a DRO approach for training ML models to perform uniformly well over subpopulations. They design a worst case optimization procedure over structured distribution shifts salient in predictive applications: shifts in (a subset of) covariates. The authors propose a convex procedure that controls worst case subpopulation performance and provide finite-sample (nonparametric) convergence guarantees. Empirically, they demonstrate their worst case procedure on lexical similarity, wine quality, and recidivism prediction tasks and observe significantly improved performance across unseen subpopulations.

Active Vibration Control of Rotating Machines with Active Machine Foot Mounts on Soft Foundations Based on a 3D-Model

The paper presents a simplified 3D-model for active vibration control of rotating machines with active machine foot mounts on soft foundations, considering static and moment unbalance. After the model is mathematical described in the time domain, it is transferred into the Fourier domain, where the frequencies response functions regarding bearing housing vibrations, foundation vibrations and actuator forces are derived. Afterwards, the mathematical coherences are described in the Laplace domain and a worst case procedure is presented to analyze the vibration stability. For special controller structures in combination with certain feedback strategies, a calculation method is shown, where the controller parameters can be directly implemented into the stiffness matrix, damping matrix and mass matrix. Additionally a numerical example is presented, where the vibration stability and the frequency response functions are analyzed.

Attention-deficit/hyperactivity disorder and phonological working memory: Methodological variability affects clinical and experimental performance metrics.

Despite promising findings in extant research that suggest impaired working memory (WM) serves as a central neurocognitive deficit or candidate endophenotype of attention-deficit/hyperactivity disorder (ADHD), findings from translational research have been relatively underwhelming. This study aimed to explicate previous equivocal findings by systematically examining the effect of methodological variability on WM performance estimates across experimental and clinical WM measures. Age-matched boys (ages 8-12 years) with (n = 20) and without (n = 20) ADHD completed 1 experimental (phonological) and 2 clinical (digit span, letter-number sequencing) WM measures. The use of partial scoring procedures, administration of greater trial numbers, and high central executive demands yielded moderate-to-large between-groups effect sizes. Moreover, the combination of these best-case procedures, compared to worst-case procedures (i.e., absolute scoring, administration of few trials, use of discontinue rules, and low central executive demands), resulted in a 12.5% increase in correct group classification. Collectively, these findings explain inconsistent ADHD-related WM deficits in previous reports, and highlight the need for revised clinical measures that utilize best-case procedures. (PsycINFO Database Record

Intraocular pressure measurements during flap preparation using 2 femtosecond lasers and 1 microkeratome in human donor eyes

To evaluate and compare intraocular pressures (IOPs) during flap preparations performed using 2 femtosecond lasers and a mechanical microkeratome in human donor globes. University Medical Center Mainz, Mainz, and Euroeyes Clinic Stuttgart, Stuttgart, Germany. Experimental study. A cannula was inserted through the optic nerve in human globes. The IOP was obtained continuously during flap preparation using the 60 kHz Intralase femtosecond laser, the 200 kHz Visumax femtosecond laser, or the Amadeus II microkeratome. For each experiment, a normal lamellar flap preparation (regular procedure) and a worst-case procedure (femtosecond laser interface was pressed against globe until docking maneuver was aborted) were performed. During the regular procedure, the mean maximum IOP measured was 181.3 mm Hg (range 159.1 to 194.8 mm Hg) with the 60 kHz femtosecond laser, 77.6 mm Hg (range 58.1 to100.3 mm Hg) with the 200 kHz femtosecond laser, and 198.1 mm Hg (range 162.8 to 299.6mm Hg) with the microkeratome. During the worst-case procedure, the maximum measured IOP was 319.7 mm Hg (range 299.1 to 341.2 mm Hg) with the 60 kHz laser and 120.4 mm Hg (range 118.1 to 134.7 mm Hg) with the 200 kHz laser. Maximum IOPs during corneal flap preparations in human enucleated eyes were lower during performance of a regular procedure and a worst-case procedure with the 200 kHz femtosecond laser than with the 60 kHz femtosecond laser and the mechanical microkeratome. Dr. Sekundo is a member of the Scientific Advisory Board of Carl Zeiss Meditec AG, Jena, Germany. No author has a financial or proprietary interest in any material or method mentioned.

Comparison of Intraocular Pressure During Corneal Flap Preparation Between a Femtosecond Laser and a Mechanical Microkeratome in Porcine Eyes

To compare the increase in intraocular pressure (IOP) during corneal flap preparation in porcine eyes when using a femtosecond laser or a mechanical microkeratome. The present study was conducted at a university hospital and a private clinic. The vitreous IOP was directly measured with a cannula through the optic nerve in 24 porcine globes (n = 12 for each device). In the first procedure (worst-case procedure), the eye interface was lowered against the globe until abortion of the docking maneuver when using the IntraLase femtosecond laser (Abbott Medical Optics, Santa Ana, CA) or the suction ring was pressed very firmly against the globe when using the Amadeus microkeratome (Ziemer Ophthalmic Systems AG, Port, Switzerland). In the second procedure (regular procedure), a normal lamellar flap creation was performed using each of these devices. During the worst-case procedure, a maximum IOP of 260 ± 53 mm Hg (mean ± SD) was reached with the IntraLase and 318 ± 59 mm Hg was reached with the Amadeus. During the regular procedure, the maximum IOP increased to 135 ± 16 mm Hg when using the IntraLase femtosecond laser and to 152 ± 24 mm Hg with the Amadeus microkeratome. In the worst-case procedure, the maximum IOP levels were lower in the IntraLase group. No significant difference in maximum IOP levels could be obtained during the regular procedure. The duration of the suction phase is longer when using the IntraLase femtosecond laser.

Energization of Wind Turbine Transformers With an Auxiliary Generator in a Large Offshore Wind Farm During Islanded Operation

The prospective inrush currents during the energization of wind turbine transformers in a large offshore wind farm during islanded operation were estimated by time-domain simulations covering different energizing scenarios. In this way, worst case procedures could be identified. The energization of the farm was carried out by an auxiliary diesel-driven synchronous generator placed on the offshore platform. The available information from all of the components and excitation control was implemented in PSCAD-EMTDC. The results from the simulations were compared with the capability curves and current limits of the synchronous generator. Based on the worst possible transformer energization scenario, an optimized procedure to reduce the loading of the auxiliary generator during transient conditions has been proposed based on the reduced generator terminal voltage, automatic voltage regulator parameters, and reactive power compensation element switching at the offshore platform.

Intraocular Pressure During Corneal Flap Preparation: Comparison Among Four Femtosecond Lasers in Porcine Eyes

To compare the course of intraocular pressure (IOP) during corneal flap preparation using four different femtosecond lasers in porcine globes. Forty-eight (12 in each group) enucleated globes were successfully cannulated through the optic nerve. Intraocular pressure was measured continuously through the cannula during a normal lamellar flap creation (regular procedure) using four femtosecond lasers (IntraLase, Abbott Medical Optics; VisuMax, Carl Zeiss Meditec AG; Femtec, Technolas Perfect Vision; and Femto LDV, Ziemer Ophthalmic Systems AG). In an additional measurement (worst-case procedure), the patient interface was pressed against the globe with increasing force until the applanation maneuver was automatically aborted by those devices capable of doing so. During the regular procedure, the maximum IOP reached was 135±16 mmHg when using the Intra-Lase, 65±20 mmHg with the VisuMax, 205±32 mmHg with the Femtec, and 184±28 mmHg with the Femto LDV. During the worst-case procedure, a maximum IOP of 260±53 mmHg was reached with the IntraLase, 105±13 mmHg with the VisuMax, and 248±51 mmHg with the Femtec. There is considerable variation in IOP among the tested femtosecond lasers during a regular lamellar flap creation and during the worst-case procedure. The VisuMax femtosecond laser seems to cause the lowest IOP rise in both settings.

On minimisation of word length of multidimensional filter realisations

In this paper, we present a statistical word length minimisation procedure based on the dynamical model of stable-unstable transitions in N-dimensional digital filters. This dynamical model [1] was originally used to determine a priori, for a worst case design, the minimum word length of the multipliers needed to retain stability. In this contribution, the dynamical model of stable-unstable transitions in N-dimensional filters is used to estimate each multiplier's minimum word length sufficient, in some statistical sense, to retain stability. The new procedure has been compared with the worst case procedure. It has been found that the present procedure results in a significant saving of the word length in comparison with the worst case analysis. The method is therefore an attractive alternative to the worst case procedure. Numerical examples are solved at the end to illustrate the procedure presented.