Ingredients for Responsible Machine Learning: A Commented Review of The Hitchhiker's Guide to Responsible Machine Learning.

In The hitchhiker's guide to responsible machine learning, Biecek, Kozak, and Zawada (here BKZ) provide an illustrated and engaging step-by-step guide on how to perform a machine learning (ML) analysis such that the algorithms, the software, and the entire process is interpretable and transparent for both the data scientist and the end user. This review summarises BKZ's book and elaborates on three elements key to ML analyses: inductive inference, causality, and interpretability.

Improved Accuracy in Predicting the Best Sensor Fusion Architecture for Multiple Domains.

Multi-sensor fusion intends to boost the general reliability of a decision-making procedure or allow one sensor to compensate for others' shortcomings. This field has been so prominent that authors have proposed many different fusion approaches, or "architectures" as we call them when they are structurally different, so it is now challenging to prescribe which one is better for a specific collection of sensors and a particular application environment, other than by trial and error. We propose an approach capable of predicting the best fusion architecture (from predefined options) for a given dataset. This method involves the construction of a meta-dataset where statistical characteristics from the original dataset are extracted. One challenge is that each dataset has a different number of variables (columns). Previous work took the principal component analysis's first k components to make the meta-dataset columns coherent and trained machine learning classifiers to predict the best fusion architecture. In this paper, we take a new route to build the meta-dataset. We use the Sequential Forward Floating Selection algorithm and a T transform to reduce the features and match them to a given number, respectively. Our findings indicate that our proposed method could improve the accuracy in predicting the best sensor fusion architecture for multiple domains.

Kvasir-Capsule, a video capsule endoscopy dataset.

Artificial intelligence (AI) is predicted to have profound effects on the future of video capsule endoscopy (VCE) technology. The potential lies in improving anomaly detection while reducing manual labour. Existing work demonstrates the promising benefits of AI-based computer-assisted diagnosis systems for VCE. They also show great potential for improvements to achieve even better results. Also, medical data is often sparse and unavailable to the research community, and qualified medical personnel rarely have time for the tedious labelling work. We present Kvasir-Capsule, a large VCE dataset collected from examinations at a Norwegian Hospital. Kvasir-Capsule consists of 117 videos which can be used to extract a total of 4,741,504 image frames. We have labelled and medically verified 47,238 frames with a bounding box around findings from 14 different classes. In addition to these labelled images, there are 4,694,266 unlabelled frames included in the dataset. The Kvasir-Capsule dataset can play a valuable role in developing better algorithms in order to reach true potential of VCE technology.

A Genetic Attack Against Machine Learning Classifiers to Steal Biometric Actigraphy Profiles from Health Related Sensor Data.

In this work, we propose the use of a genetic-algorithm-based attack against machine learning classifiers with the aim of 'stealing' users' biometric actigraphy profiles from health related sensor data. The target classification model uses daily actigraphy patterns for user identification. The biometric profiles are modeled as what we call impersonator examples which are generated based solely on the predictions' confidence score by repeatedly querying the target classifier. We conducted experiments in a black-box setting on a public dataset that contains actigraphy profiles from 55 individuals. The data consists of daily motion patterns recorded with an actigraphy device. These patterns can be used as biometric profiles to identify each individual. Our attack was able to generate examples capable of impersonating a target user with a success rate of 94.5%. Furthermore, we found that the impersonator examples have high transferability to other classifiers trained with the same training set. We also show that the generated biometric profiles have a close resemblance to the ground truth profiles which can lead to sensitive data exposure, like revealing the time of the day an individual wakes-up and goes to bed.

HyperKvasir, a comprehensive multi-class image and video dataset for gastrointestinal endoscopy.

Artificial intelligence is currently a hot topic in medicine. However, medical data is often sparse and hard to obtain due to legal restrictions and lack of medical personnel for the cumbersome and tedious process to manually label training data. These constraints make it difficult to develop systems for automatic analysis, like detecting disease or other lesions. In this respect, this article presents HyperKvasir, the largest image and video dataset of the gastrointestinal tract available today. The data is collected during real gastro- and colonoscopy examinations at Bærum Hospital in Norway and partly labeled by experienced gastrointestinal endoscopists. The dataset contains 110,079 images and 374 videos, and represents anatomical landmarks as well as pathological and normal findings. The total number of images and video frames together is around 1 million. Initial experiments demonstrate the potential benefits of artificial intelligence-based computer-assisted diagnosis systems. The HyperKvasir dataset can play a valuable role in developing better algorithms and computer-assisted examination systems not only for gastro- and colonoscopy, but also for other fields in medicine.

Applying machine learning in motor activity time series of depressed bipolar and unipolar patients compared to healthy controls.

Current practice of assessing mood episodes in affective disorders largely depends on subjective observations combined with semi-structured clinical rating scales. Motor activity is an objective observation of the inner physiological state expressed in behavior patterns. Alterations of motor activity are essential features of bipolar and unipolar depression. The aim was to investigate if objective measures of motor activity can aid existing diagnostic practice, by applying machine-learning techniques to analyze activity patterns in depressed patients and healthy controls. Random Forrest, Deep Neural Network and Convolutional Neural Network algorithms were used to analyze 14 days of actigraph recorded motor activity from 23 depressed patients and 32 healthy controls. Statistical features analyzed in the dataset were mean activity, standard deviation of mean activity and proportion of zero activity. Various techniques to handle data imbalance were applied, and to ensure generalizability and avoid overfitting a Leave-One-User-Out validation strategy was utilized. All outcomes reports as measures of accuracy for binary tests. A Deep Neural Network combined with SMOTE class balancing technique performed a cut above the rest with a true positive rate of 0.82 (sensitivity) and a true negative rate of 0.84 (specificity). Accuracy was 0.84 and the Matthews Correlation Coefficient 0.65. Misclassifications appear related to data overlapping among the classes, so an appropriate future approach will be to compare mood states intra-individualistically. In summary, machine-learning techniques present promising abilities in discriminating between depressed patients and healthy controls in motor activity time series.

Activity Recognition Using Community Data to Complement Small Amounts of Labeled Instances.

Human Activity Recognition (HAR) is an important part of ambient intelligence systems since it can provide user-context information, thus allowing a greater personalization of services. One of the problems with HAR systems is that the labeling process for the training data is costly, which has hindered its practical application. A common approach is to train a general model with the aggregated data from all users. The problem is that for a new target user, this model can perform poorly because it is biased towards the majority type of users and does not take into account the particular characteristics of the target user. To overcome this limitation, a user-dependent model can be trained with data only from the target user that will be optimal for this particular user; however, this requires a considerable amount of labeled data, which is cumbersome to obtain. In this work, we propose a method to build a personalized model for a given target user that does not require large amounts of labeled data. Our method uses data already labeled by a community of users to complement the scarce labeled data of the target user. Our results showed that the personalized model outperformed the general and the user-dependent models when labeled data is scarce.

Automatic Stress Detection in Working Environments From Smartphones' Accelerometer Data: A First Step.

Increase in workload across many organizations and consequent increase in occupational stress are negatively affecting the health of the workforce. Measuring stress and other human psychological dynamics is difficult due to subjective nature of selfreporting and variability between and within individuals. With the advent of smartphones, it is now possible to monitor diverse aspects of human behavior, including objectively measured behavior related to psychological state and consequently stress. We have used data from the smartphone's built-in accelerometer to detect behavior that correlates with subjects stress levels. Accelerometer sensor was chosen because it raises fewer privacy concerns (e.g., in comparison to location, video, or audio recording), and because its low-power consumption makes it suitable to be embedded in smaller wearable devices, such as fitness trackers. About 30 subjects from two different organizations were provided with smartphones. The study lasted for eight weeks and was conducted in real working environments, with no constraints whatsoever placed upon smartphone usage. The subjects reported their perceived stress levels three times during their working hours. Using combination of statistical models to classify selfreported stress levels, we achieved a maximum overall accuracy of 71% for user-specific models and an accuracy of 60% for the use of similar-users models, relying solely on data from a single accelerometer.

Long-term activity recognition from wristwatch accelerometer data.

With the development of wearable devices that have several embedded sensors, it is possible to collect data that can be analyzed in order to understand the user's needs and provide personalized services. Examples of these types of devices are smartphones, fitness-bracelets, smartwatches, just to mention a few. In the last years, several works have used these devices to recognize simple activities like running, walking, sleeping, and other physical activities. There has also been research on recognizing complex activities like cooking, sporting, and taking medication, but these generally require the installation of external sensors that may become obtrusive to the user. In this work we used acceleration data from a wristwatch in order to identify long-term activities. We compare the use of Hidden Markov Models and Conditional Random Fields for the segmentation task. We also added prior knowledge into the models regarding the duration of the activities by coding them as constraints and sequence patterns were added in the form of feature functions. We also performed subclassing in order to deal with the problem of intra-class fragmentation, which arises when the same label is applied to activities that are conceptually the same but very different from the acceleration point of view.