Search

Citizen science (CS) projects have started to utilize Machine Learning (ML) to sort through large datasets generated in fields like astronomy, ecology and biodiversity, biology, and neuroimaging. Human–machine systems have been created to take advantage of the complementary strengths of humans and machines and have been optimized for efficiency and speed. We conducted qualitative content analysis on meta-summaries of documents reporting the results of 12 citizen science projects that used machine learning to optimize classification tasks. We examined the distribution of tasks between citizen scientists, experts, and algorithms, and how epistemic agency was enacted in terms of whose knowledge shapes the distribution of tasks, who decides what knowledge is relevant to the classification, and who validates it. In our descriptive results, we found that experts, who include professional scientists and algorithm developers, are involved in every aspect of a project, from annotating or labelling data to giving data to algorithms to train them to make decisions from predictions. Experts also test and validate models to improve their accuracy by scoring their outputs when algorithms fail to make correct decisions. Experts are mostly the humans involved in a loop, but when algorithms encounter problems, citizens are also involved at several stages. In this paper, we present three main examples of citizens-in-the-loop: (a) when algorithms provide incorrect suggestions; (b) when algorithms fail to know how to perform classification; and (c) when algorithms pose queries. We consider the implications of the emphasis on optimization on the ideal of science and the role of citizen scientists from a perspective informed by Science and Technology Studies (STS) and Information Systems (IS). Based on our findings, we conclude that ML in CS classification projects, far from being deterministic in its nature and effects, may be open to question. There is no guarantee that these technologies can replace citizen scientists, nor any guarantee that they can provide citizens with opportunities for more interesting tasks. [ABSTRACT FROM AUTHOR]

In recent years, the gastronomy industry has increased the demand for rigorous and reliable tools to evaluate culinary creativity; but conceptually, creativity is difficult to define and even more difficult to measure. In this paper, we propose an AI-based method for assessing culinary product creativity by using the renowned high cuisine restaurant elBulli as a case study to understand the proliferation and scale of an entity's creativity and innovation. To achieve so, we trained a Random Forest Classifier to assess the level of creativity within the restaurant's recipes. The application of this model to elBulli's recipes allowed us to answer the following questions: how can creativity in a culinary product be characterized? Is there a single trait that is the key to culinary creativity? Is the creative culinary product a truly original outcome? Our key findings are (1) although ingredients and preparation techniques prove to be suitable attributes to characterize creativity in a recipe, none of them nor an elite subset clearly identifies creative cuisine, (2) new products are not radically novel with respect to previous creations, (3) creativity is more related to new uses and combinations of existing ingredients and techniques rather than to new ingredients or techniques. We believe that our method can find practical applications in supporting and guiding the creation of culinary products, identifying their strengths and weaknesses, and helping restaurants maintain a sustainable competitive advantage. [ABSTRACT FROM AUTHOR]

Nowadays, mobile telephony interruptions in our daily life activities are common because of the inappropriate ringing notifications of incoming phone calls in different contexts. Such interruptions may impact on the work attention not only for the mobile phone owners, but also for the surrounding people. Decision tree is the most popular machine-learning classification technique that is used in existing context-aware mobile intelligent interruption management (MIIM) model to overcome such issues. However, a single-decision tree-based context-aware model may cause over-fitting problem and thus decrease the prediction accuracy of the inferred model. Therefore, in this paper, we propose an ensemble machine-learning-based context-aware mobile telephony model for the purpose of intelligent interruption management by taking into account multi-dimensional contexts and name it "E-MIIM". The experimental results on individuals' real-life mobile telephony data sets show that our E-MIIM model is more effective and outperforms existing MIIM model for predicting and managing individual's mobile telephony interruptions based on their relevant contextual information. [ABSTRACT FROM AUTHOR]

Most of the existing classification techniques concentrate on learning the datasets as a single similar unit, in spite of so many differentiating attributes and complexities involved. However, traditional classification techniques are required to analyze the datasets prior to learning, and if not doing so, they loss their performance in terms of accuracy and AUC. To this end, many of the machine learning problems can be very easily solved just by carefully observing human learning and training nature and then mimicking the same in the machine learning. In response to these issues, we present a comprehensive suite of experiments carefully designed to provide conclusive, reliable, and significant results to the problem of efficient learning. This paper proposes a novel, simple, and effective machine learning paradigm that explicitly exploits this important similar-to-different (S2D) human learning strategy and implements it based on two algorithms (C4.5 and CART) efficiently. The framework not only analyzes the data sets prior to implementation, but also carefully allows classifier to have a systematic study so as to mimic the human training technique designed for efficient learning. Experimental results show that the method outperforms the state-of-the-art methods in terms of learning capability and breaks through the gap between human and machine learning. In fact, the proposed similar-to-different (S2D) strategy may also be useful in efficient learning of real-world complex and high-dimensional data sets, especially which are very typical to learn with traditional classifiers. [ABSTRACT FROM AUTHOR]