Suchergebnisse

Machine Learning (ML) and Artificial Intelligence (AI) impact many aspects of human life, from recommending a significant other to assist the search for extraterrestrial life. The area develops rapidly and exiting unexplored design spaces are constantly laid bare. The focus in this work is one of these areas; ML systems where decisions concerning ML model training, usage and selection of target domain lay in the hands of domain experts. This work is then on ML systems that function as a tool that augments and/or enhance human capabilities. The approach presented is denoted Human In Command ML (HIC-ML) systems. To enquire into this research domain design experiments of varying fidelity were used. Two of these experiments focus on augmenting human capabilities and targets the domains commuting and sorting batteries. One experiment focuses on enhancing human capabilities by identifying similar hand-painted plates. The experiments are used as illustrative examples to explore settings where domain experts potentially can: independently train an ML model and in an iterative fashion, interact with it and interpret and understand its decisions. HIC-ML should be seen as a governance principle that focuses on adding value and meaning to users. In this work, concrete application areas are presented and discussed. To open up for designing ML-based products for the area an abstract model for HIC-ML is constructed and design guidelines are proposed. In addition, terminology and abstractions useful when designing for explicability are presented by imposing structure and rigidity derived from scientific explanations. Together, this opens up for a contextual shift in ML and makes new application areas probable, areas that naturally couples the usage of AI technology to human virtues and potentially, as a consequence, can result in a democratisation of the usage and knowledge concerning this powerful technology.

A significant set of epistemic and political transformations are taking place as states and societies begin to understand themselves and their problems through the paradigm of deep neural network algorithms. A machine learning political order does not merely change the political technologies of governance, but is itself a reordering of politics, of what the political can be. When algorithmic systems reduce the pluridimensionality of politics to the output of a model, they simultaneously foreclose the potential for other political claims to be made and alternative political projects to be built. More than this foreclosure, a machine learning political order actively profits and learns from the fracturing of communities and the destabilising of democratic rights. The transformation from rules-based algorithms to deep learning models has paralleled the undoing of rules-based social and international orders – from the use of machine learning in the campaigns of the UK EU referendum, to the trialling of algorithmic immigration and welfare systems, and the use of deep learning in the COVID-19 pandemic – with political problems becoming reconfigured as machine learning problems. Machine learning political orders decouple their attributes, features and clusters from underlying social values, no longer tethered to notions of good governance or a good society, but searching instead for the optimal function of abstract representations of data.

Recent revelations concerning data firm Cambridge Analytica's illegitimate use of the data of millions of Facebook users highlights the ethical and, relatedly, legal issues arising from the use of machine learning techniques. Cambridge Analytica is, or was – the revelations brought about its demise - a firm that used machine learning processes to try to influence elections in the US and elsewhere by, for instance, targeting 'vulnerable' voters in marginal seats with political advertising. Of course, there is nothing new about political candidates and parties employing firms to engage in political advertising on their behalf, but if a data firm has access to the personal information of millions of voters, and is skilled in the use of machine learning techniques, then it can develop detailed, fine-grained voter profiles that enable political actors to reach a whole new level of manipulative influence over voters. My focus in this paper is not with the highly publicised ethical and legal issues arising from Cambridge Analytic's activities but rather with some important ethical issues arising from the use of machine learning techniques that have not received the attention and analysis that they deserve. I focus on three areas in which machine learning techniques are used or, it is claimed, should be used, and which give rise to problems at the interface of law and ethics (or law and morality, I use the terms "ethics" and "morality" interchangeably). The three areas are profiling and predictive policing (Saunders et al. 2016), legal adjudication (Zeleznikow, 2017), and machines' compliance with legally enshrined moral principles (Arkin 2010). I note that here, as elsewhere, new and emerging technologies are developing rapidly making it difficult to predict what might or might not be able to be achieved in the future. For this reason, I have adopted the conservative stance of restricting my ethical analysis to existing machine learning techniques and applications rather than those that are the object of speculation or even informed extrapolation (Mittelstadt et al. 2015). This has the consequence that what I might regard as a limitation of machine learning techniques, e.g. in respect of predicting novel outcomes or of accommodating moral principles, might be thought by others to be merely a limitation of currently available techniques. After all, has not the history of AI recently shown the naysayers to have been proved wrong? Certainly, AI has seen some impressive results, including the construction of computers that can defeat human experts in complex games, such as chess and Go (Silver et al. 2017), and others that can do a better job than human medical experts at identifying the malignancy of moles and the like (Esteva et al. 2017). However, since by definition future machine learning techniques and applications are not yet with us the general claim that current limitations will be overcome cannot at this time be confirmed or disconfirmed on the basis of empirical evidence.

Today, open data platforms host a wide and heterogeneous catalog of datasets. However, these datasets are often neglected in Machine Learning (ML) and other related tasks. This mainly happens because there are few available open data catalogs specialized in ML applications and because it is often unclear whether Machine Learning algorithms would be adequate and well performing on such datasets. Therefore, several open datasets go unused while they could be leveraged by the ML community to explain, evaluate, and challenge existing methods on real open data. For instance, these real-world data could be used by professors teaching ML courses, by students taking these courses, by researchers testing current and novel ML approaches, and possibly to promote the intersection of open data, ML and public policy. In this talk we will show you how we are tackling this issue working on datasets from data.gouv.fr (DGF), the French open data government platform. We aim to answer the question of what makes a dataset suitable and well performing for Machine Learning tasks by leveraging open source tools. Our goal is to establish a first small empirical assessment of the characteristics of a dataset (size, balance of its categorical variables and so on) that make it a "good fit" for Machine Learning algorithms. Specifically, we first manually select an adequate subset of datasets from DGF. Then we perform a statistic profiling on each of these datasets. Thirdly, we automatically train and validate a set of ML algorithms on them and we cluster the datasets according to their evaluation results. These steps help us to better understand the nature of each dataset and thus determine which ones seem suitable for ML applications. Based on these datasets, and inspired by existing resources, we build the first version of a catalog of open datasets for ML. We hope that this platform will be a first stepping stone towards the reuse of open datasets in Machine Learning contexts.

Nowadays, we are surrounded by electric appliances. Either at home by the washing machine, kettle, or oven, or work by the computer, cellphone, or printer. Such devices help us daily, but their popularization increased the energy consumption to concerning values. In an attempt to reduce energy consumption, governments started enforcing policies regarding energy education to teach homeowners how to reduce energy wastage on the demand side. One of those policies was the deployment of smart meters, which allow the consumer to know how much energy is being consumed at any given time through a display on the household energy meter. Even though this measure was well received, the studies show that the best results in energy conservation are obtained through real-time appliance level feedback. To get such feedback, one can either measure every outlet in a household, which is unviable for a broad deployment solution, or disaggregate the energy recorded by the smart meter. NILM or Non-Intrusive Load Monitoring is the name we give to the second option where we use the aggregated readings of a household to find the energy consumed by each appliance. There were many proposals to solve NILM ranging from HMMs to GSP, where deep learning models showed remarkable results, obtaining state-of-the-art results. With the intent to create a complete NILM solution, Withus partnered with the University of Aveiro and proposed this dissertation. The initial objective was to develop a machine learning model to solve NILM. Still, during the background analysis, we found the need to create a new dataset which led to the expansion of the initial proposal to include the dataset preprocessing and conversion. Regarding NILM, we proposed three new deep learning models: a convolutional neural network with residual blocks, a recurrent neural network, and a multilayer perceptron that uses discrete wavelet transforms as features. These models went through multiple iterations, being evaluated first in the simpler ON/OFF classification task and later ...

This dissertation is comprised of three essays that apply machine learning and high-dimensional statistics to causal inference. The first essay proposes a parametric alternative to the synthetic control method (Abadie and Gardeazabal, 2003; Abadie et al., 2010) that relies on a Lasso-type first-step. We show that the resulting estimator is doubly robust, asymptotically Gaussian and ``immunized'' against first-step selection mistakes. The second essay studies a penalized version of the synthetic control method especially useful in the presence of micro-economic data. The penalization parameter trades off pairwise matching discrepancies with respect to the characteristics of each unit in the synthetic control against matching discrepancies with respect to the characteristics of the synthetic control unit as a whole. We study the properties of the resulting estimator, propose data-driven choices of the penalization parameter and discuss randomization-based inference procedures. The last essay applies the Generic Machine Learning framework (Chernozhukov et al., 2018) to study heterogeneity of the treatment in a randomized experiment designed to compare public and private provision of job counselling. From a methodological perspective, we discuss the extension of the Generic Machine Learning framework to experiments with imperfect compliance. ; Cette thèse regroupe trois travaux d'économétrie liés par l'application du machine learning et de la statistique en grande dimension à l'évaluation de politiques publiques. La première partie propose une alternative paramétrique au contrôle synthétique (Abadie and Gardeazabal, 2003; Abadie et al., 2010) sous la forme d'un estimateur reposant sur une première étape de type Lasso, dont on montre qu'il est doublement robuste, asymptotiquement Normal et ``immunisé'' contre les erreurs de première étape. La seconde partie étudie une version pénalisée du contrôle synthétique en présence de données de nature micro-économique. La pénalisation permet d'obtenir une unité synthétique qui ...

This dissertation is comprised of three essays that apply machine learning and high-dimensional statistics to causal inference. The first essay proposes a parametric alternative to the synthetic control method (Abadie and Gardeazabal, 2003; Abadie et al., 2010) that relies on a Lasso-type first-step. We show that the resulting estimator is doubly robust, asymptotically Gaussian and ``immunized'' against first-step selection mistakes. The second essay studies a penalized version of the synthetic control method especially useful in the presence of micro-economic data. The penalization parameter trades off pairwise matching discrepancies with respect to the characteristics of each unit in the synthetic control against matching discrepancies with respect to the characteristics of the synthetic control unit as a whole. We study the properties of the resulting estimator, propose data-driven choices of the penalization parameter and discuss randomization-based inference procedures. The last essay applies the Generic Machine Learning framework (Chernozhukov et al., 2018) to study heterogeneity of the treatment in a randomized experiment designed to compare public and private provision of job counselling. From a methodological perspective, we discuss the extension of the Generic Machine Learning framework to experiments with imperfect compliance. ; Cette thèse regroupe trois travaux d'économétrie liés par l'application du machine learning et de la statistique en grande dimension à l'évaluation de politiques publiques. La première partie propose une alternative paramétrique au contrôle synthétique (Abadie and Gardeazabal, 2003; Abadie et al., 2010) sous la forme d'un estimateur reposant sur une première étape de type Lasso, dont on montre qu'il est doublement robuste, asymptotiquement Normal et ``immunisé'' contre les erreurs de première étape. La seconde partie étudie une version pénalisée du contrôle synthétique en présence de données de nature micro-économique. La pénalisation permet d'obtenir une unité synthétique qui ...

This dissertation is comprised of three essays that apply machine learning and high-dimensional statistics to causal inference. The first essay proposes a parametric alternative to the synthetic control method (Abadie and Gardeazabal, 2003; Abadie et al., 2010) that relies on a Lasso-type first-step. We show that the resulting estimator is doubly robust, asymptotically Gaussian and ``immunized'' against first-step selection mistakes. The second essay studies a penalized version of the synthetic control method especially useful in the presence of micro-economic data. The penalization parameter trades off pairwise matching discrepancies with respect to the characteristics of each unit in the synthetic control against matching discrepancies with respect to the characteristics of the synthetic control unit as a whole. We study the properties of the resulting estimator, propose data-driven choices of the penalization parameter and discuss randomization-based inference procedures. The last essay applies the Generic Machine Learning framework (Chernozhukov et al., 2018) to study heterogeneity of the treatment in a randomized experiment designed to compare public and private provision of job counselling. From a methodological perspective, we discuss the extension of the Generic Machine Learning framework to experiments with imperfect compliance. ; Cette thèse regroupe trois travaux d'économétrie liés par l'application du machine learning et de la statistique en grande dimension à l'évaluation de politiques publiques. La première partie propose une alternative paramétrique au contrôle synthétique (Abadie and Gardeazabal, 2003; Abadie et al., 2010) sous la forme d'un estimateur reposant sur une première étape de type Lasso, dont on montre qu'il est doublement robuste, asymptotiquement Normal et ``immunisé'' contre les erreurs de première étape. La seconde partie étudie une version pénalisée du contrôle synthétique en présence de données de nature micro-économique. La pénalisation permet d'obtenir une unité synthétique qui ...

This dissertation is comprised of three essays that apply machine learning and high-dimensional statistics to causal inference. The first essay proposes a parametric alternative to the synthetic control method (Abadie and Gardeazabal, 2003; Abadie et al., 2010) that relies on a Lasso-type first-step. We show that the resulting estimator is doubly robust, asymptotically Gaussian and ``immunized'' against first-step selection mistakes. The second essay studies a penalized version of the synthetic control method especially useful in the presence of micro-economic data. The penalization parameter trades off pairwise matching discrepancies with respect to the characteristics of each unit in the synthetic control against matching discrepancies with respect to the characteristics of the synthetic control unit as a whole. We study the properties of the resulting estimator, propose data-driven choices of the penalization parameter and discuss randomization-based inference procedures. The last essay applies the Generic Machine Learning framework (Chernozhukov et al., 2018) to study heterogeneity of the treatment in a randomized experiment designed to compare public and private provision of job counselling. From a methodological perspective, we discuss the extension of the Generic Machine Learning framework to experiments with imperfect compliance. ; Cette thèse regroupe trois travaux d'économétrie liés par l'application du machine learning et de la statistique en grande dimension à l'évaluation de politiques publiques. La première partie propose une alternative paramétrique au contrôle synthétique (Abadie and Gardeazabal, 2003; Abadie et al., 2010) sous la forme d'un estimateur reposant sur une première étape de type Lasso, dont on montre qu'il est doublement robuste, asymptotiquement Normal et ``immunisé'' contre les erreurs de première étape. La seconde partie étudie une version pénalisée du contrôle synthétique en présence de données de nature micro-économique. La pénalisation permet d'obtenir une unité synthétique qui ...

This dissertation is comprised of three essays that apply machine learning and high-dimensional statistics to causal inference. The first essay proposes a parametric alternative to the synthetic control method (Abadie and Gardeazabal, 2003; Abadie et al., 2010) that relies on a Lasso-type first-step. We show that the resulting estimator is doubly robust, asymptotically Gaussian and ``immunized'' against first-step selection mistakes. The second essay studies a penalized version of the synthetic control method especially useful in the presence of micro-economic data. The penalization parameter trades off pairwise matching discrepancies with respect to the characteristics of each unit in the synthetic control against matching discrepancies with respect to the characteristics of the synthetic control unit as a whole. We study the properties of the resulting estimator, propose data-driven choices of the penalization parameter and discuss randomization-based inference procedures. The last essay applies the Generic Machine Learning framework (Chernozhukov et al., 2018) to study heterogeneity of the treatment in a randomized experiment designed to compare public and private provision of job counselling. From a methodological perspective, we discuss the extension of the Generic Machine Learning framework to experiments with imperfect compliance. ; Cette thèse regroupe trois travaux d'économétrie liés par l'application du machine learning et de la statistique en grande dimension à l'évaluation de politiques publiques. La première partie propose une alternative paramétrique au contrôle synthétique (Abadie and Gardeazabal, 2003; Abadie et al., 2010) sous la forme d'un estimateur reposant sur une première étape de type Lasso, dont on montre qu'il est doublement robuste, asymptotiquement Normal et ``immunisé'' contre les erreurs de première étape. La seconde partie étudie une version pénalisée du contrôle synthétique en présence de données de nature micro-économique. La pénalisation permet d'obtenir une unité synthétique qui ...

Users are increasingly interacting with machine learning (ML)-based curation systems. YouTube and Facebook, two of the most visited websites worldwide, utilize such systems to curate content for billions of users. Contemporary challenges such as fake news, filter bubbles, and biased predictions make the understanding of ML-based curation systems an important and timely concern. Despite their political, social, and cultural importance, practitioners' framing of machine learning and users' understanding of ML-based curation systems have not been investigated systematically. This is problematic since machine learning - as a novel programming paradigm in which a mapping between input and output is inferred from data - poses a variety of open research questions regarding users' understanding. The first part of this thesis provides the first in-depth investigation of ML-based curation systems as socio-technical systems. The second part of the thesis contributes recommendations on how ML-based curation systems can and should be explained and audited. The first part analyses practitioners' framing of ML by examining how the term machine learning, ML applications, and ML algorithms are framed in tutorials. The thesis also investigates the beliefs that users have about YouTube and introduces a user belief framework of ML-based curation systems. Furthermore, it demonstrates how limited users' capabilities for providing input data for ML-based curation systems are. The second part evaluates different explanations of ML-based systems. This evaluation uncovered an explanatory gap between what is available to explain ML-based curation systems and what users need to understand such systems. Informed by this explanatory gap, the second part of this thesis demonstrates that audits of ML systems can be an important alternative to explanations. This demonstration of audits also uncovers a popularity bias enacted by YouTube's ML-based curation system. Based on these findings, the thesis recommends performing audits to ensure that ML-based systems act in the public's interest.

Within the framework of the project "Uncertainty Management for Robust Industrial Design in Aeronautics" (UMRIDA), funded by the European Union, several machine learning-based predictive models were compared in terms of their efficiency in estimating statistics of aerodynamic performance of aerofoils. The results show that the models based on both samples and gradients achieve better accuracy than those based solely on samples at the same computational costs.

Today's conflicts are becoming increasingly complex, fluid and fragmented, often involving a host of national and international actors with multiple and often divergent interests. This development poses significant challenges for conflict mediation, as mediators struggle to make sense of conflict dynamics, such as the range of conflict parties and the evolution of their political positions, the distinction between relevant and less relevant actors in peace making, or the identification of key conflict issues and their interdependence. International peace efforts appear increasingly ill-equipped to successfully address these challenges. While technology is being increasingly used in a range of conflict related fields, such as conflict predicting or information gathering, less attention has been given to how technology can contribute to conflict mediation. This case study is the first to apply state-of-the-art machine learning technologies to data from an ongoing mediation process. Using dialogue transcripts from peace negotiations in Yemen, this study shows how machine-learning tools can effectively support international mediators by managing knowledge and offering additional conflict analysis tools to assess complex information. Apart from illustrating the potential of machine learning tools in conflict mediation, the paper also emphasises the importance of interdisciplinary and participatory research design for the development of context-sensitive and targeted tools and to ensure meaningful and responsible implementation.

Dissertação de mestrado em Matemática e Computação ; Portuguese legislation states the compulsory reporting of the addition of amenities, such as swimming pools, to the Portuguese tax authority. The purpose is to update the property tax value, to be charged annually to the owner of each real estate. According to Technavio and Market- Watch, this decade will bring a global rise to the number of swimming pools due to certain factors such as: cost reduction, increasing health consciousness, and others. The need for inspections to ensure that all new constructions are communicated to the competent authorities is therefore rapidly increasing and new solutions are needed to address this problem. Typically, supervision is done by sending human resources to the field, involving huge time and resource consumption, and preventing the catalogue from updating at a rate close to the speed of construction. Automation is rapidly becoming an absolute requirement to improve task efficiency and affordability. Recently, Deep Learning algorithms have shown incredible performance results when used for object detection tasks. Based on the above, the objective of this thesis is to study the various existing object detection algorithms and implement a Deep Learning model capable of recognising swimming pools from satellite images. To achieve the best results for this specific task, the RetinaNet algorithm was chosen. To provide a smooth user experience with the developed model, a simple graphical user interface was also created. ; A legislação Portuguesa declara a obrigatoriedade da comunicação de novas construções, como piscinas, à Autoridade Tributária e Aduaneira. Esta comunição permite o ajustamento do Imposto Municipal sobre Imóveis a pagar anualmente pelo proprietário. De acordo com o Technavio e o MarketWatch, irá ocorrer um aumento significativo do número de piscinas devido a vários fatores como a redução do custo da construção, o aumento da consciência para a adoção de um estilo de vida saudável, entre outros. Isto leva à ...

Plasmodium is one of India's biggest public health problems. Early prediction of a malaria epidemic is that the secret to malaria morbidity management, mortality as well as reducing the risk of malaria transmission in the community will benefit politicians, health care providers, medical officers, health ministry and other health organizations to better target medical resources to areas of greatest need. In this project, we acquire data sets from hospital databases, which have the information about the causes of malaria, and the images of cells infected with malaria. We then analyze these data sets and feed them to our machine-learning model. Here we are using contour detection and random forest algorithms for training the model and predicting the output