Towards cost-sensitive adaptation: when is it worth updating your predictive model?

Our digital universe is rapidly expanding,more and more daily activities are digitally recorded, data arrives in streams, it needs to be analyzed in real time and may evolve over time. In the last decade many adaptive learning algorithms and prediction systems, which can automatically update themselves with the new incoming data, have been developed. The majority of those algorithms focus on improving the predictive performance and assume that model update is always desired as soon as possible and as frequently as possible. In this study we consider potential model update as an investment decision, which, as in the financial markets, should be taken only if a certain return on investment is expected. We introduce and motivate a new research problem for data streams ? cost-sensitive adaptation. We propose a reference framework for analyzing adaptation strategies in terms of costs and benefits. Our framework allows to characterize and decompose the costs of model updates, and to asses and interpret the gains in performance due to model adaptation for a given learning algorithm on a given prediction task. Our proof-of-concept experiment demonstrates how the framework can aid in analyzing and managing adaptation decisions in the chemical industry.

Massive Open Online Courses and sustainability

Massive Open Online Courses (MOOCs) have become very popular among learners millions of users from around the world registered with leading platforms. There are hundreds of universities (and other organizations) offering MOOCs. However, sustainability of MOOCs is a pressing concern as MOOCs incur up front creation costs, maintenance costs to keep content relevant and on-going support costs to provide facilitation while a course is being run. At present, charging a fee for certification (for example Coursera Signature Track and FutureLearn Statement of Completion) seems a popular business model. In this paper, the authors discuss other possible business models and their pros and cons. Some business models discussed here are: Freemium model – providing content freely but charging for premium services such as course support, tutoring and proctored exams. Sponsorships – courses can be created in collaboration with industry where industry sponsorships are used to cover the costs of course production and offering. For example Teaching Computing course was offered by the University of East Anglia on the FutureLearn platform with the sponsorship from British Telecom while the UK Government sponsored the course Introduction to Cyber Security offered by the Open University on FutureLearn. Initiatives and Grants – The government, EU commission or corporations could commission the creation of courses through grants and initiatives according to the skills gap identified for the economy. For example, the UK Government’s National Cyber Security Programme has supported a course on Cyber Security. Similar initiatives could also provide funding to support relevant course development and offering. Donations – Free software, Wikipedia and early OER initiatives such as the MIT OpenCourseware accept donations from the public and this could well be used as a business model where learners could contribute (if they wish) to the maintenance and facilitation of a course. Merchandise – selling merchandise could also bring revenue to MOOCs. As many participants do not seek formal recognition (European Commission, 2014) for their completion of a MOOC, merchandise that presents their achievement in a playful way could well be attractive for them. Sale of supplementary material –supplementary course material in the form of an online or physical book or similar could be sold with the revenue being reinvested in the course delivery. Selective advertising – courses could have advertisements relevant to learners Data sharing – though a controversial topic, sharing learner data with relevant employers or similar could be another revenue model for MOOCs. Follow on events – the courses could lead to follow on summer schools, courses or other real-life or online events that are paid-for in which case a percentage of the revenue could be passed on to the MOOC for its upkeep. Though these models are all possible ways of generating revenue for MOOCs, some are more controversial and sensitive than others. Nevertheless unless appropriate business models are identified the sustainability of MOOCs would be problematic.

Towards pain-free diagnosis of skin diseases through multiplexed microneedles: biomarker extraction and detection using a highly sensitive blotting method

Immunodiagnostic microneedles provide a novel way to extract protein biomarkers from the skin in a minimally invasive manner for analysis in vitro. The technology could overcome challenges in biomarker analysis specifically in solid tissue, which currently often involves invasive biopsies. This study describes the development of a multiplex immunodiagnostic device incorporating mechanisms to detect multiple antigens simultaneously, as well as internal assay controls for result validation. A novel detection method is also proposed. It enables signal detection specifically at microneedle tips and therefore may aid the construction of depth profiles of skin biomarkers. The detection method can be coupled with computerised densitometry for signal quantitation. The antigen specificity, sensitivity and functional stability of the device were assessed against a number of model biomarkers. Detection and analysis of endogenous antigens (interleukins 1α and 6) from the skin using the device was demonstrated. The results were verified using conventional enzyme-linked immunosorbent assays. The detection limit of the microneedle device, at ≤10 pg/mL, was at least comparable to conventional plate-based solid-phase enzyme immunoassays.

Speleothems as sensitive recorders of volcanic eruptions - the Bronze Age Minoan eruption recorded in a stalagmite from Turkey

Mounting evidence exists that variations in sulphur content in stalagmites are closely linked to changes in volcanic or anthropogenic atmospheric sulphur. The strong dependency of sulphur on soil pH and ecosystem storage, however, can result in a delay of several years to decades in the registration of volcanic eruptions and anthropogenic emissions by stalagmites. Here we present synchrotron-radiation based trace element analysis performed on a precisely-dated section of a stalagmite from Sofular Cave in Northern Turkey. As this section covers the time interval of the intensively studied Minoan volcanic eruption between 1600 and 1650 BC, we can test whether this vigorous eruption can be traced in a stalagmite. Of all measured trace elements, only bromine shows a clear short-lived peak at 1621±251621±25 BC, whereas sulphur and molybdenum show peaks later at 1617±251617±25 and 1589±251589±25 respectively. We suggest that all trace element peaks are related to the Minoan eruption, whereas the observed phasing of bromine, molybdenum and sulphur is related to differences in their retention rates in the soil above Sofular Cave. For the first time, we can show that bromine appears to be an ideal volcanic tracer in stalagmites, as it is a prominent volatile component in volcanic eruptions, can be easily leached in soils and rapidly transferred from the atmosphere through the soil and bedrock into the cave and stalagmite respectively. Highly resolved oxygen and carbon isotope profiles indicate that the Minoan eruption had no detectable climatic and environmental impact in Northern Turkey.

Discourse-sensitive clitic-doubled dislocations in heritage Spanish

This experimental study tests the predictions of the Interface Hypothesis (Sorace, 2011, 2012) using two constructions whose appropriateness depends on monitoring discourse information: Clitic Left Dislocation and Fronted Focus. Clitic Left Dislocation relates a dislocated and clitic-doubled object to an antecedent activated in previous discourse, while Fronted Focus does not relate the fronted constituent to a discourse antecedent. The Interface Hypothesis argues that speakers in language contact situations experience difficulties when they have to integrate syntactic with discourse information. We tested four groups of native speakers on these constructions: Spanish monolinguals, bilinguals with more than 7 years residence in the US, intermediate and advanced proficiency heritage speakers. Our findings suggest that attrition has not set in the adult L2 bilingual speakers, and that the heritage speakers perform similarly to the monolingual and the adult sequential bilingual natives.

Interacting effects of climate change and habitat fragmentation on drought-sensitive butterflies

Climate change is expected to increase the frequency of some climatic extremes. These may have drastic impacts on biodiversity, particularly if meteorological thresholds are crossed, leading to population collapses. Should this occur repeatedly, populations may be unable to recover, resulting in local extinctions. Comprehensive time series data on butterflies in Great Britain provide a rare opportunity to quantify population responses to both past severe drought and the interaction with habitat area and fragmentation. Here, we combine this knowledge with future projections from multiple climate models, for different Representative Concentration Pathways (RCPs), and for simultaneous modelled responses to different landscape characteristics. Under RCP8.5, which is associated with ‘business as usual’ emissions, widespread drought-sensitive butterfly population extinctions could occur as early as 2050. However, by managing landscapes and particularly reducing habitat fragmentation, the probability of persistence until mid-century improves from around zero to between 6 and 42% (95% confidence interval). Achieving persistence with a greater than 50% chance and right through to 2100 is possible only under both low climate change (RCP2.6) and semi-natural habitat restoration. Our data show that, for these drought-sensitive butterflies, persistence is achieved more effectively by restoring semi-natural landscapes to reduce fragmentation, rather than simply focusing on increasing habitat area, but this will only be successful in combination with substantial emission reductions.

Activation of the mitogen-activated protein kinase cascade by pertussis toxin-sensitive and -insensitive pathways in cultured ventricular cardiomyocytes.

The involvement of pertussis toxin (PTX)-sensitive and -insensitive pathways in the activation of the mitogen-activated protein kinase (MAPK) cascade was examined in ventricular cardiomyocytes cultured from neonatal rats. A number of agonists that activate heterotrimeric G-protein-coupled receptors stimulated MAPK activity after exposure for 5 min. These included foetal calf serum (FCS), endothelin-1 (these two being the most effective of the agonists examined), phenylephrine, endothelin-3, lysophosphatidic acid, carbachol, isoprenaline and angiotensin II. Activation of MAPK and MAPK kinase (MEK) by carbachol returned to control levels within 30-60 min, whereas activation by FCS was more sustained. FPLC on Mono Q showed that carbachol and FCS activated two peaks of MEK and two peaks of MAPK (p42MAPK and p44MAPK). Pretreatment of cells with PTX for 24 h inhibited the activation of MAPK by carbachol, FCS and lysophosphatidic acid, but not that by endothelin-1, phenylephrine or isoprenaline. Involvement of G-proteins in the activation of the cardiac MAPK cascade was demonstrated by the sustained (PTX-insensitive) activation of MAPK (and MEK) after exposure of cells to AlF4-. AlF4- activated PtdIns hydrolysis, as did endothelin-1, endothelin-3, phenylephrine and FCS. In contrast, the effect of lysophosphatidic acid on PtdIns hydrolysis was small and carbachol was without significant effect even after prolonged exposure. We conclude that PTX-sensitive (i.e. Gi/G(o)-linked) and PTX-insensitive (i.e. Gq/Gs-linked) pathways of MAPK activation exist in neonatal ventricular myocytes. FCS may stimulate the MAPK cascade through both pathways.

Ionic liquids and other liquid matrices for sensitive MALDI MS analysis

Although liquid matrix-assisted laser desorption/ionization (MALDI) has been used in mass spectrometry (MS) since the early introduction of MALDI, its substantial lack of sensitivity compared to solid (crystalline) MALDI was for a long time a major hurdle to its analytical competitiveness. In the last decade, this situation has changed with the development of new sensitive liquid matrices, which are often based on a binary matrix acid/base system. Some of these matrices were inspired by the recent progress in ionic liquid research, while others were developed from revisiting previous liquid MALDI work as well as from a combination of these two approaches. As a result, two high-performing liquid matrix classes have been developed, the ionic liquid matrices (ILMs) and the liquid support matrices (LSMs), now allowing MS measurements at a sensitivity level that is very close to the level of solid MALDI and in some cases even surpasses it. This chapter provides some basic information on a selection of highly successful representatives of these new liquid matrices and describes in detail how they are made and applied in MALDI MS analysis.

Doing well by talking good? A topic modelling-assisted discourse study of corporate social responsibility

Using the novel technique of topic modelling, this paper examines thematic patterns and their changes over time in a large corpus of corporate social responsibility (CSR) reports produced in the oil sector. Whereas previous research on corporate communications has been small-scale or interested in selected lexical aspects and thematic categories identified ex ante, our approach allows for thematic patterns to emerge from the data. The analysis reveals a number of major trends and topic shifts pointing to changing practices of CSR. Nowadays ‘people’, ‘communities’ and ‘rights’ seem to be given more prominence, whereas ‘environmental protection’ appears to be less relevant. Using more established corpus-based methods, we subsequently explore two top phrases - ‘human rights’ and ‘climate change’ that were identified as representative of the shifting thematic patterns. Our approach strikes a balance between the purely quantitative and qualitative methodologies and offers applied linguists new ways of exploring discourse in large collections of texts.