Questions from a Contraceptive Pill Junkie: Applying Human Psychometrics to Investigate Gender Bias in Machine Learning

<p>With the rapid uptake of machine learning artificial intelligence in our daily lives, we are beginning to realise the risks involved in implementing this technology in high-stakes decision making. This risk is due to machine learning decisions being based in human-curated datasets, meaning these decisions are not bias-free. Machine learning datasets put women at a disadvantage due to factors including (but not limited to) historical exclusion of women in data collection, research, and design; as well as the low participation of women in artificial intelligence fields. These factors mean that applications of machine learning may fail to treat the needs and experiences of women as equal to those of men. Research into understanding gender biases in machine learning frequently occurs within the computer science field. This has frequently resulted in research where bias is inconsistently defined, and proposed techniques do not engage with relevant literature outside of the artificial intelligence field. This research proposes a novel, interdisciplinary approach to the measurement and validation of gender biases in machine learning. This approach translates methods of human-based gender bias measurement in psychology, forming a gender bias questionnaire for use on a machine rather than a human. The final output system of this research as a proof of concept demonstrates the potential for a new approach to gender bias investigation. This system takes advantage of the qualitative nature of language to provide a new way of understanding gender data biases by outputting both quantitative and qualitative results. These results can then be meaningfully translated into their real-world implications.</p>

Questions from a Contraceptive Pill Junkie: Applying Human Psychometrics to Investigate Gender Bias in Machine Learning

<p>With the rapid uptake of machine learning artificial intelligence in our daily lives, we are beginning to realise the risks involved in implementing this technology in high-stakes decision making. This risk is due to machine learning decisions being based in human-curated datasets, meaning these decisions are not bias-free. Machine learning datasets put women at a disadvantage due to factors including (but not limited to) historical exclusion of women in data collection, research, and design; as well as the low participation of women in artificial intelligence fields. These factors mean that applications of machine learning may fail to treat the needs and experiences of women as equal to those of men. Research into understanding gender biases in machine learning frequently occurs within the computer science field. This has frequently resulted in research where bias is inconsistently defined, and proposed techniques do not engage with relevant literature outside of the artificial intelligence field. This research proposes a novel, interdisciplinary approach to the measurement and validation of gender biases in machine learning. This approach translates methods of human-based gender bias measurement in psychology, forming a gender bias questionnaire for use on a machine rather than a human. The final output system of this research as a proof of concept demonstrates the potential for a new approach to gender bias investigation. This system takes advantage of the qualitative nature of language to provide a new way of understanding gender data biases by outputting both quantitative and qualitative results. These results can then be meaningfully translated into their real-world implications.</p>

Combining machine learning and conventional statistical approaches for risk factor discovery in a large cohort study

We present a simple and efficient hypothesis-free machine learning pipeline for risk factor discovery that accounts for non-linearity and interaction in large biomedical databases with minimal variable pre-processing. In this study, mortality models were built using gradient boosting decision trees (GBDT) and important predictors were identified using a Shapley values-based feature attribution method, SHAP values. Cox models controlled for false discovery rate were used for confounder adjustment, interpretability, and further validation. The pipeline was tested using information from 502,506 UK Biobank participants, aged 37–73 years at recruitment and followed over seven years for mortality registrations. From the 11,639 predictors included in GBDT, 193 potential risk factors had SHAP values ≥ 0.05, passed the correlation test, and were selected for further modelling. Of the total variable importance summed up, 60% was directly health related, and baseline characteristics, sociodemographics, and lifestyle factors each contributed about 10%. Cox models adjusted for baseline characteristics, showed evidence for an association with mortality for 166 out of the 193 predictors. These included mostly well-known risk factors (e.g., age, sex, ethnicity, education, material deprivation, smoking, physical activity, self-rated health, BMI, and many disease outcomes). For 19 predictors we saw evidence for an association in the unadjusted but not adjusted analyses, suggesting bias by confounding. Our GBDT-SHAP pipeline was able to identify relevant predictors ‘hidden’ within thousands of variables, providing an efficient and pragmatic solution for the first stage of hypothesis free risk factor identification.

A Segmentation-Free Machine Learning Architecture for Immune Landscape Phenotyping in Solid Tumors by Multichannel Imaging

Tissue specimens taken from primary tumors or metastases contain important information for diagnosis and treatment of cancer patients. Multispectral imaging allows in situ visualization of heterogeneous cell subsets, such as lymphocytes, in tissue samples. Many image processing pipelines first segment cell boundaries and then measure marker expression to assign cell phenotypes. In dense tissue environments such as solid tumors, segmentation-based phenotyping can be inaccurate due to segmentation errors or overlapping cell boundaries. Here we introduce a machine learning pipeline design called ImmuNet that directly identifies the positions and phenotypes of immune cells without determining their exact boundaries. ImmuNet is easy to train: human annotators only need to click on immune cells and rank their expression of each marker; full annotation of tissue regions is not necessary. We demonstrate that ImmuNet is a suitable approach for immune cell detection and phenotyping in multiplex immunohistochemistry: it compares favourably to segmentation-based methods, especially in dense tissues, and we externally validate ImmuNet results by comparing them to flow cytometric measurements from the same tissue. In summary, ImmuNet performs well on diverse tissue specimens, takes relatively little effort to train and implement, and is a simpler alternative to segmentation-based approaches when only cell positions and phenotypes, but not their shapes are required for downstream analyses. We hope that ImmuNet will help cancer researchers to analyze multichannel tissue images more easily and accurately.

Rapid 3D phenotypic analysis of neurons and organoids using data-driven cell segmentation-free machine learning

Phenotypic profiling of large three-dimensional microscopy data sets has not been widely adopted due to the challenges posed by cell segmentation and feature selection. The computational demands of automated processing further limit analysis of hard-to-segment images such as of neurons and organoids. Here we describe a comprehensive shallow-learning framework for automated quantitative phenotyping of three-dimensional (3D) image data; using unsupervised data-driven voxel-based feature learning, which enables computationally facile classification, clustering and advanced data visualization. We demonstrate the analysis potential on complex 3D images by investigating the phenotypic alterations of: neurons in response to apoptosis-inducing treatments and morphogenesis for oncogene-expressing human mammary gland acinar organoids. Our novel implementation of image analysis algorithms called Phindr3D allowed rapid implementation of data-driven voxel-based feature learning into 3D high content analysis (HCA) operations and constitutes a major practical advance as the computed assignments represent the biology while preserving the heterogeneity of the underlying data. Phindr3D is provided as Matlab code and as a stand-alone program (https://github.com/DWALab/Phindr3D).

Numerical Analysis of the Rotor of a 30 kW ORC Microturbine Considering Properties of Aerodynamic Gas Bearings

The paper focuses on the analysis of a 30 kW microturbine operating in the organic Rankine cycle (ORC) with a low-boiling working medium. The nominal speed of the rotor is 40,000 rpm. The investigated microturbine is an oil-free machine, which means that its bearings use the ORC working medium as a lubricant. We created a numerical model, which was used to assess the dynamic properties of the newly designed microturbine. The conducted analyses covered, inter alia, the optimization of some geometrical parameters of each bearing in order to cause the lubricating film to be created at a correspondingly low rotational speed as well as to obtain optimal dynamic properties of the system. The article provides a full dynamic picture of the rotor supported by two aerodynamic gas bearings. The included graphs demonstrate the vibration amplitude of the shaft as a function of the rotational speed as well as the results of the modal analysis in the form of natural vibration modes of the system and their corresponding natural frequencies.