Bernd Malle

The Right to Be Forgotten: Towards Machine Learning on Perturbed Knowledge Bases

Today’s increasingly complex information infrastructures represent the basis of any data-driven industries which are rapidly becoming the 21st century’s economic backbone. The sensitivity of those infrastructures to disturbances in their knowledge bases is therefore of crucial interest for companies, organizations, customers and regulating bodies. This holds true with respect to the direct provisioning of such information in crucial applications like clinical settings or the energy industry, but also when considering additional insights, predictions and personalized services that are enabled by the automatic processing of those data. In the light of new EU Data Protection regulations applying from 2018 onwards which give customers the right to have their data deleted on request, information processing bodies will have to react to these changing jurisdictional (and therefore economic) conditions. Their choices include a re-design of their data infrastructure as well as preventive actions like anonymization of databases per default. Therefore, insights into the effects of perturbed/anonymized knowledge bases on the quality of machine learning results are a crucial basis for successfully facing those future challenges. In this paper we introduce a series of experiments we conducted on applying four different classifiers to an established dataset, as well as several distorted versions of it and present our initial results.

On the Generation of Point Cloud Data Sets: Step One in the Knowledge Discovery Process

Computational geometry and topology are areas which have much potential for the analysis of arbitrarily high-dimensional data sets. In order to apply geometric or topological methods one must first generate a representative point cloud data set from the original data source, or at least a metric or distance function, which defines a distance between the elements of a given data set. Consequently, the first question is: How to get point cloud data sets? Or more precise: What is the optimal way of generating such data sets? The solution to these questions is not trivial. If a natural image is taken as an example, we are concerned more with the content, with the shape of the relevant data represented by this image than its mere matrix of pixels. Once a point cloud has been generated from a data source, it can be used as input for the application of graph theory and computational topology. In this paper we first describe the case for natural point clouds, i.e. where the data already are represented by points; we then provide some fundamentals of medical images, particularly dermoscopy, confocal laser scanning microscopy, and total-body photography; we describe the use of graph theoretic concepts for image analysis, give some medical background on skin cancer and concentrate on the challenges when dealing with lesion images. We discuss some relevant algorithms, including the watershed algorithm, region splitting (graph cuts), region merging (minimum spanning tree) and finally describe some open problems and future challenges.

On Graph Extraction from Image Data

Hot topics in knowledge discovery and interactive data mining from natural images include the application of topological methods and machine learning algorithms. For any such approach one needs at first a relevant and robust digital content representation from the image data. However, traditional pixel-based image analysis techniques do not effectively extract, hence represent the content. A very promising approach is to extract graphs from images, which is not an easy task. In this paper we present a novel approach for knowledge discovery by extracting graph structures from natural image data. For this purpose, we created a framework built upon modern Web technologies, utilizing HTML canvas and pure Javascript inside a Web-browser, which is a very promising engineering approach. Following on a short description of some popular image classification and segmentation methodologies, we outline a specific data processing pipeline suitable for carrying out future scientific research. A demonstration of our implementation, compared to the results of a traditional watershed transformation performed in Matlab showed very promising results in both quality and runtime, despite some open problems. Finally, we provide a short discussion of a few open problems and outline some of our future research routes.

Machine Learning and Knowledge Extraction in Digital Pathology Needs an Integrative Approach

During the last decade pathology has benefited from the rapid progress of image digitizing technologies, which led to the development of scanners, capable to produce so-called Whole Slide images (WSI) which can be explored by a pathologist on a computer screen comparable to the conventional microscope and can be used for diagnostics, research, archiving and also education and training. Digital pathology is not just the transformation of the classical microscopic analysis of histological slides by pathologists to just a digital visualization. It is a disruptive innovation that will dramatically change medical work-flows in the coming years and help to foster personalized medicine. Really powerful gets a pathologist if she/he is augmented by machine learning, e.g. by support vector machines, random forests and deep learning. The ultimate benefit of digital pathology is to enable to learn, to extract knowledge and to make predictions from a combination of heterogenous data, i.e. the histological image, the patient history and the *omics data. These challenges call for integrated/integrative machine learning approach fostering transparency, trust, acceptance and the ability to explain step-by-step why a decision has been made.

Machine Learning and Data Mining Methods for Managing Parkinson’s Disease

Parkinson’s disease (PD) results primarily from dying of dopaminergic neurons in the Substantia Nigra, a part of the Mesencephalon (midbrain), which is not curable to date. PD medications treat symptoms only, none halt or retard dopaminergic neuron degeneration. Here machine learning methods can be of help since one of the crucial roles in the management and treatment of PD patients is detection and classification of tremors. In the clinical practice, this is one of the most common movement disorders and is typically classified using behavioral or etiological factors. Another important issue is to detect and evaluate PD related gait patterns, gait initiation and freezing of gait, which are typical symptoms of PD. Medical studies have shown that 90% of people with PD suffer from vocal impairment, consequently the analysis of voice data to discriminate healthy people from PD is relevant. This paper provides a quick overview of the state-of-the-art and some directions for future research, motivated by the ongoing PD_manager project.

The More the Merrier - Federated Learning from Local Sphere Recommendations.

With Google’s Federated Learning & Facebook’s introduction of client-side NLP into their chat service, the era of client-side Machine Learning is upon us. While interesting ML approaches beyond the realm of toy examples were hitherto confined to large data-centers and powerful GPU’s, exponential trends in technology and the introduction of billions of smartphones enable sophisticated processing swarms of even hand-held devices. Such approaches hold several promises: 1. Without the need for powerful server infrastructures, even small companies could be scalable to millions of users easily and cost-efficiently; 2. Since data only used in the learning process never need to leave the client, personal information can be used free of privacy and data security concerns; 3. Since privacy is preserved automatically, the full range of personal information on the client device can be utilized for learning; and 4. without round-trips to the server, results like recommendations can be made available to users much faster, resulting in enhanced user experience. In this paper we propose an architecture for federated learning from personalized, graph based recommendations computed on client devices, collectively creating & enhancing a global knowledge graph. In this network, individual users will ‘train’ their local recommender engines, while a server-based voting mechanism aggregates the developing client-side models, preventing over-fitting on highly subjective data from tarnishing the global model.

DO NOT DISTURB? Classifier Behavior on Perturbed Datasets

Exponential trends in data generation are presenting today’s organizations, economies and governments with challenges never encountered before, especially in the field of privacy and data security. One crucial trade-off regulators are facing regards the simultaneous need for publishing personal information for the sake of statistical analysis and Machine Learning in order to increase quality levels in areas like medical services, while at the same time protecting the identity of individuals. A key European measure will be the introduction of the General Data Protection Regulation (GDPR) in 2018, giving customers the ‘right to be forgotten’, i.e. having their data deleted on request. As this could lead to a competitive disadvantage for European companies, it is important to understand which effects deletion of significant data points has on the performance of ML techniques. In a previous paper we introduced a series of experiments applying different algorithms to a binary classification problem under anonymization as well as perturbation. In this paper we extend those experiments by multi-class classification and introduce outlier-removal as an additional scenario. While the results of our previous work were mostly in-line with our expectations, our current experiments revealed unexpected behavior over a range of different scenarios. A surprising conclusion of those experiments is the fact that classification on an anonymized dataset with outliers removed in beforehand can almost compete with classification on the original, un-anonymized dataset. This could soon lead to competitive Machine Learning pipelines on anonymized datasets for real-world usage in the marketplace.