Search

Biogenic Volatile Organic Compounds (BVOCs) are continuously emitted from terrestrial vegetation into the atmosphere and react with various atmospheric oxidants, with ozone being an important one. The reaction between BVOCs and ozone can lead to low volatile organic compounds, other pollutants, and the formation of secondary organic aerosols. To understand the chemical and physical processes taking place in the atmosphere, a complete picture of the BVOCs emitted is necessary. However, the large pool of BVOCs present makes it difficult to detect every compound. The total ozone reactivity method can help understand the ozone reactive potential of all BVOCs emitted into the atmosphere and also help determine if current analytical techniques can measure the total BVOC budget. In this study, we measured the total ozone reactivity from the emissions of a Norway spruce tree in Hyytiälä in late summer using the Total Ozone Reactivity Monitor (TORM) built at the Finnish Meteorological Institute (FMI). We also conducted chemical characterisation and quantification of the BVOC emissions using a gas chromatograph coupled with a mass spectrometer (GC-MS). The measured total ozone reactivity reached up to 7.3 x 10-9 m3 s-2 g-1, which corresponds to 64 μg g-1 h-1 of α-pinene. Stress-related sesquiterpenes such as β-farnesene and α-farnesene, and an unidentified sesquiterpene contributed the most to the observed emissions. However, the observed emissions made up only 35 % of the measured total ozone reactivity, with sesquiterpenes being the most important sink for the ozone. High total ozone reactivity was especially seen during high temperature periods, with up to 95 % of the reactivity remaining unexplained. Emissions of unidentified stress-related compounds could be the reason for the high fraction of missing reactivity.

Here we introduce a new method, termed “Nano Ranking Analysis,” for characterizing new particle formation (NPF) from atmospheric observations. Using daily variations of the particle number concentration at sizes immediately above the continuous mode of molecular clusters, here in practice 2.5–5 nm - ΔN2.5–5, we can determine the occurrence and estimate the strength of atmospheric NPF events. After determining the value of ΔN2.5–5 for all the days during a period under consideration, the next step of the analysis is to rank the days based on this simple metric. The analysis is completed by grouping the days either into a number of percentile intervals based on their ranking or into a few modes in the distribution of log(ΔN2.5–5) values. Using five years (2018–2022) of data from the SMEAR II station in Hyytiälä, Finland, we found that the days with higher (lower) ranking values had, on average, both higher (lower) probability of NPF events and higher (lower) particle formation rates. The new method provides probabilistic information about the occurrence and intensity of NPF events and is expected to serve as a valuable tool to define the origin of newly formed particles at many types of environments that are affected by multiple sources of aerosol precursors.

The interaction between nitrogen monoxide (NO) and organic peroxy radicals (RO$_{2}$) greatly impacts the formation of highly oxygenated organic molecules (HOM), the key precursors of secondary organic aerosols. It has been thought that HOM production can be significantly suppressed by NO even at low concentrations. Here, we perform dedicated experiments focusing on HOM formation from monoterpenes at low NO concentrations (0 – 82 pptv). We demonstrate that such low NO can enhance HOM production by modulating the RO$_{2}$ loss and favoring the formation of alkoxy radicals that can continue to autoxidize through isomerization. These insights suggest that HOM yields from typical boreal forest emissions can vary between 2.5%-6.5%, and HOM formation will not be completely inhibited even at high NO concentrations. Our findings challenge the notion that NO monotonically reduces HOM yields by extending the knowledge of RO$_{2}$-NO interactions to the low-NO regime. This represents a major advance towards an accurate assessment of HOM budgets, especially in low-NO environments, which prevails in the pre-industrial atmosphere, pristine areas, and the upper boundary layer.

EvalRS aims to bring together practitioners from industry and academia to foster a debate on rounded evaluation of recommender systems, with a focus on real-world impact across a multitude of deployment scenarios. Recommender systems are often evaluated only through accuracy metrics, which fall short of fully characterizing their generalization capabilities and miss important aspects, such as fairness, bias, usefulness, informativeness. This workshop builds on the success of last year's workshop at CIKM, but with a broader scope and an interactive format., EvalRS 2023 will be a workshop hosted at KDD23

New particle formation (NPF) in the tropical free troposphere (FT) is a globally important source of cloud condensation nuclei, affecting cloud properties and climate. Oxidized organic molecules (OOMs) produced from biogenic volatile organic compounds are believed to contribute to aerosol formation in the tropical FT, but without direct chemical observations. We performed in-situ molecular-level OOMs measurements at the Bolivian station Chacaltaya at 5240 meters above sea level, on the western edge of Amazonia. For the first time, we demonstrate the presence of OOMs, mainly with 4-5 carbon atoms, simultaneously in both gas and particulate phases in tropical FT air from Amazonia. These observations, combined with air mass history analyses, indicate that the observed OOMs are linked to isoprene emitted from the rainforests hundreds of kilometers away. Based on particle-phase measurements, we find that these compounds can contribute to the growth of newly formed particles, and are potentially crucial for new particle formation in the tropical free troposphere on a continental scale. Our study will thus improve the understanding of aerosol formation process in the tropics.

Pre-trained language models (PLMs) have outperformed other NLP models on a wide range of tasks. Opting for a more thorough understanding of their capabilities and inner workings, researchers have established the extend to which they capture lower-level knowledge like grammaticality, and mid-level semantic knowledge like factual understanding. However, there is still little understanding of their knowledge of higher-level aspects of language. In particular, despite the importance of sociodemographic aspects in shaping our language, the questions of whether, where, and how PLMs encode these aspects, e.g., gender or age, is still unexplored. We address this research gap by probing the sociodemographic knowledge of different single-GPU PLMs on multiple English data sets via traditional classifier probing and information-theoretic minimum description length probing. Our results show that PLMs do encode these sociodemographics, and that this knowledge is sometimes spread across the layers of some of the tested PLMs. We further conduct a multilingual analysis and investigate the effect of supplementary training to further explore to what extent, where, and with what amount of pre-training data the knowledge is encoded. Our overall results indicate that sociodemographic knowledge is still a major challenge for NLP. PLMs require large amounts of pre-training data to acquire the knowledge and models that excel in general language understanding do not seem to own more knowledge about these aspects., Accepted for publication at EMNLP 2022

The most common ways to explore latent document dimensions are topic models and clustering methods. However, topic models have several drawbacks: e.g., they require us to choose the number of latent dimensions a priori, and the results are stochastic. Most clustering methods have the same issues and lack flexibility in various ways, such as not accounting for the influence of different topics on single documents, forcing word-descriptors to belong to a single topic (hard-clustering) or necessarily relying on word representations. We propose PROgressive SImilarity Thresholds - ProSiT, a deterministic and interpretable method, agnostic to the input format, that finds the optimal number of latent dimensions and only has two hyper-parameters, which can be set efficiently via grid search. We compare this method with a wide range of topic models and clustering methods on four benchmark data sets. In most setting, ProSiT matches or outperforms the other methods in terms six metrics of topic coherence and distinctiveness, producing replicable, deterministic results.

We show preliminary evidence that sound symbolic associations in the stimuli set of two relevant psycholinguistic studies emerge in a computational model trained to embed sentence- image pairs in a joint latent space. We thus show for the first time that sound symbolic cross- modal associations between sub-lexical patterns and visual features can be learned via associative mechanisms, leveraging co-occurrences in the environment.In detail, we use the CLIP model (Contrastive Language Image Pretraining [1]) to embed the stimuli from [2] and [3] and then use the WEAT score (Word Embedding Association Test [4]) to compute the association between sets of round/jagged images and pseudowords [5].CLIP is a neural network trained on sentence-image pairs and represents sentences and images as points in a latent multimodal embedding space via contrastive learning. We used CLIP ViT-B/32, where images are fed to a visual encoder which relies on Vision Transformers [6], while sentences are fed to a language encoder, which relies on text transformers [7]. Importantly, the text encoder is also able to embed out-of-vocabulary words leveraging part- words, enabling it to deal with pseudowords, typically used in studies on sound symbolism.The WEAT score has been widely used to uncover associations between concepts, e.g., flowers and pleasantness or insects and unpleasantness, from word embeddings. We adopt the same approach but use multimodal rather than purely language-based embeddings to test sound-symbolic associations between visually presented pseudowords and roundish or jagged shapes respectively, using CLIP embeddings.We first embedded the two sets of pseudowords (8 sharp-sounding, Wjagged, 8 round- sounding, Wround) and image stimuli (21 jagged shapes, Ijagged, 20 roundish shapes, Iround) in [2]. The pseudowords were grouped into sharp- or round-sounding by the authors of the original study based on the phonemes of which they consist of and their known sound symbolic asso- ciations, e.g., b and round or k and sharp. We then computed the WEAT score between the embedded targets, Wjagged and Wround, and the embedded attributes, Ijagged and Iround, observing an association in the predicted direction (effect size of 0.479).We then considered the stimuli in [3]. First, we embedded the two sets of target images (with 34 items each), featuring aliens with round v. jagged contours, again obtaining Ijagged and Iround. Then, we embedded two sets of target words, one, Wsharp, containing 10 sharp attributes like angry, and one, Wsoft, containing 10 soft attributes like friendly. The expected association between aggressive attributes and jagged aliens, based on lexical meaning rather than sound- symbolism, was observed (effect size of 0.372). We then embedded two lists of 20 and 40 made-up names, equally divided in round- and jagged-sounding (Wround and Wjagged respec- tively), and computed the WEAT score with Ijagged and Iround, finding an association in the predicted direction for both lists (effect sizes of 0.316 and 0.131). A last test, considering 20 names equally divided into male- v. female- sounding names and their expected association with jagged or round aliens respectively, also yielded an association (effect size of -0.346) but in the opposite direction, showing that CLIP may not be equally capable of capturing all documented sound-symbolic associations.To conclude, we show for the first time that cross-modal sound-symbolic associations between written words and images from two influential studies on sound symbolism are learned by CLIP, a computational model which maps text and images onto a joint embedding space exploiting co-occurrences between (sub-)lexical patterns and visual features. Unlike previous models [8], which were trained to map word-form representations to images, sound- symbolic associations emerge in CLIP as a by-product of a different objective. Cross-modal co-occurrences may thus be sufficient to learn sound-symbolic associations from experience via statistical learning [9]. Moreover, by showing that expected cross-modal associations are also captured when feeding CLIP with real words, we provide further evidence that a shared meaning space can embed words, about which we gather co-occurrence statistics, as well as pseudowords, which can be mapped onto the same representation space leveraging statisticsinvolving sub-lexical sequences [10]. Finally, we show that multimodal embedding models can be fruitfully used to study sound symbolism in a cross-modal set-up. Text encoder: the 12-layer transformer operates on a lower-cased byte pair encoding (BPE) representation of the text and has a vocabulary size of 49,152. The activations of the highest layer (512 units) of the transformer at the End-of-Sequence token (appended to any input sentence) are considered as the feature representation of the input text which is linearly projected into the multi-modal embedding space after unit-normalization. Visual encoder: the 12-layer vision transformer (VIT-base) operates on 32-pixel patches of the images and generates a feature representation (512 units) of the image which is projected into the multi-modal embedding space after unit-normalization. CLIP-training: CLIP learns representations using a contrastive task: given a batch of image-sentence pairs , the model is trained to increase the similarity of while decreasing the similarity of with i ≠ j. CLIP is trained on this task over more than 400M image-sentence pairs collected from the internet. We use the pre-trained CLIP model availa- ble through HuggingFace and do not apply further training. See [1] for further details. WEAT: this method takes in four sets of word embeddings: two sets of embedded target concepts (X and Y), e.g., round-sounding and sharp-sounding pseudowords, and two sets of embedded target attributes (A and B), e.g., round and jagged images. The WEAT score sum- marizes whether X is more closely associated with A and is computed following Eq. (1) and Eq. (2), where A and B are the set of attributes (images, in our case), X and Y are the set of targets (words or pseudowords in our case), x, y, and w are (pseudo)words, and s is a measure of similarity: (1) (,,,)= sum for x in X((,,)) − sum for y in Y((,,)) (2) (, , ) = 1/|A| * ( sum for a in A ((, ))) − 1/|B| * (sum for b in B ((, ))) References: [1] Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., ... & Sutskever, I. (2021). Learning transferable visual models from natural language supervision. ICML, 8748-8763. [2] Imai, M., & Kita, S. (2014). The sound symbolism bootstrapping hypothesis forlanguage acquisition and language evolution. Philos T Roy Soc B, 369(1651), 20130298. [3] Sidhu, D. M., & Pexman, P. M. (2015). What's in a Name? Sound Symbolism andGender in First Names. PLoS One, 10(5), e0126809. [4] Caliskan, A., Bryson, J. J., & Narayanan, A. (2017). Semantics derived automaticallyfrom language corpora contain human-like biases. Science, 356(6334), 183-186. [5] Ross, C., Katz, B., & Barbu, A. (2020). Measuring social biases in grounded vision and language embeddings. NAACL-2022. [6] Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T.,... & Houlsby, N. (2020). An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929. [7] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., and Polosukhin, I. (2017) Attention is all you need. NeurIPS, 5998–6008. [8] de Varda, A. G., & Strapparava, C. (2021). Phonovisual Biases in Language: is the Lexicon Tied to the visual world?. IJCAI-21, 643-649. [9] Sidhu, D. M., & Pexman, P. M. (2018). Five mechanisms of sound symbolicassociation. Psychon B Rev, 25(5), 1619-1643. [10] Cassani, G., & Limacher, N. (2021). Not just form, not just meaning: Words withconsistent form-meaning mappings are learned earlier. Q J Exp Psychol.

Much of the complexity of Recommender Systems (RSs) comes from the fact that they are used as part of more complex applications and affect user experience through a varied range of user interfaces. However, research focused almost exclusively on the ability of RSs to produce accurate item rankings while giving little attention to the evaluation of RS behavior in real-world scenarios. Such narrow focus has limited the capacity of RSs to have a lasting impact in the real world and makes them vulnerable to undesired behavior, such as reinforcing data biases. We propose EvalRS as a new type of challenge, in order to foster this discussion among practitioners and build in the open new methodologies for testing RSs "in the wild"., CIKM 2022 Data Challenge Paper

This article surveys a field of empirical social research linking economic exchange relations to the generation of interpersonal bonds of solidarity. This issue is particularly interesting for both social sciences and current societies because of the increasing global integration of markets and spatial mobility. Experimental research in Social Exchange Theory has provided mixed evidence on how various structural properties of exchange relations generate solidarity. Findings suggest that risk is key to generate trust between exchange partners, which is linked to feelings of solidarity. This casts doubts on the positive effect of economic exchange on solidarity, as actual market-based transactions mainly rely on assurance structures to reduce risk and prevent opportunism. However, further developments suggest that solidarity can be indirectly generated by economic exchange ties framed by loose agreements. This opens up opportunities for new empirical research to study economic exchange and solidarity in actual social systems. Furthermore, negative consequences of the link between economic exchange and solidarity are discussed, such as centralization and segregation., Sociologica, Vol 14, No 1 (2020)

Contemporary West European shantytowns have essentially been studied with qualitative methods. Questions related to their ethnic structure, homophily and interaction with local institutions have not been analysed through large samples and survey data. Based on the example of Romanian and Bulgarian Roma living in shantytowns in the Parisian metropolitan area, we analyse the ‘historical’ region of origin (autochthony), networks between individuals and households having lived together in a shantytown, as well as some of their attitudes, skills, and behaviour (i.e. expectations as to local welfare, French proficiency, children’s schooling). We used a database of slum-dwellers in Paris collected by social workers (N = 12,019). The paper looks at how autochthonies combine with socialisation in shantytowns and with territorial institutional effects of local policies. Evidence shows that while there are limited differences in the socio-professional backgrounds of slum-dwellers, there are several differences in attitudes, behaviour and skills. Moreover, social network analysis shows that these differences are only weakly related to the households’ region of origin (autochthony), while they are more correlated to emergent structural clusters of co-habitation connections, where individuals socialise in acting together to build and manage a shantytown. Moreover, local policies at the city level play a role in shaping shantytown dwellers expectations and skills for integration.

BackgroundA previous meta-analysis (Kuiper et al., 2016) has shown that multiple aspects of social relationships are associated with cognitive decline in older adults. Yet, results indicated possible bias in estimations of statistical effects due to the heterogeneity of study design and measurements. We have updated this meta-analysis adding all relevant publications from 2012 to 2020 and performed a cumulative meta-analysis to map the evolution of this growing field of research (+80% of studies from 2012-2020 compared to the period considered in the previous meta-analysis).MethodsScopus and Web of Science were searched for longitudinal cohort studies examining structural, functional and combined effects of social relationships. We combined Odds Ratios (OR) with 95% confidence intervals (CI) using random effects meta-analysis and assessed sources of heterogeneity and the likelihood of publication bias. The risk of bias was evaluated with the Quality of Prognosis Studies in Systematic Reviews (QUIPS) tool.ResultsThe review was prospectively registered on PROSPERO (ID: CRD42019130667). We identified 34 new articles published in 2012-2020. Poor social relationships were associated with cognitive decline with increasing precision of estimates compared to previously reviewed studies [(for structural, 17 articles, OR: 1.11; 95% CI: 1.08; 1.14) (for functional, 16 articles, OR: 1.12; 95% CI: 1.05; 1.20) (for combined, 5 articles, OR: 1.15; 95% CI: 1.06; 1.24)]. Meta-regression, risk and subgroup analyses showed that the precision of estimations improved in recent studies mostly due to increased sample sizes.ConclusionsOur cumulative meta-analysis would confirm that multiple aspects of social relationships are associated with cognitive decline. Yet, there is still evidence of publication bias and relevant information on study design is often missing, which could lead to an over-estimation of their statistical effects. published

Organic aerosol (OA) is a major component of fine particulate matter (PM), affecting air quality, human health, and the climate. The absorptive and reflective behavior of OA components contributes to determining particle optical properties and thus their effects on the radiative budget of the troposphere. There is limited knowledge on the influence of the molecular composition of OA on particle optical properties in the polluted urban environment. In this study, we characterized the molecular composition of oxygenated OA collected on filter samples in the autumn of 2018 in Beijing, China, with a filter inlet for gases and aerosols coupled to a high-resolution time-of-flight chemical ionization mass spectrometer (FIGAERO–CIMS). Three haze episodes occurred during our sampling period with daily maximum concentrations of OA of 50, 30, and 55 µg m−3. We found that the signal intensities of dicarboxylic acids and sulfur-containing compounds increased during the two more intense haze episodes, while the relative contributions of wood-burning markers and other aromatic compounds were enhanced during the cleaner periods. We further assessed the optical properties of oxygenated OA components by combining detailed chemical composition measurements with collocated particle light absorption measurements. We show that light absorption enhancement (Eabs) of black carbon (BC) was mostly related to more oxygenated OA (e.g., dicarboxylic acids), likely formed in aqueous-phase reactions during the intense haze periods with higher relative humidity, and speculate that they might contribute to lensing effects. Aromatics and nitro-aromatics (e.g., nitrocatechol and its derivatives) were mostly related to a high light absorption coefficient (babs) consistent with light-absorbing (brown) carbon (BrC). Our results provide information on oxygenated OA components at the molecular level associated with BrC and BC particle light absorption and can serve as a basis for further studies on the effects of anthropogenic OA on radiative forcing in the urban environment.

Aunque el término prevención cuaternaria no es conocido o dominado por una gran mayoría de los profesionales sanitarios, en los últimos años, ha existido un creciente e importante auge de este concepto y se ha expandido en todo el mundo. Cada médico debe tener la suficiente determinación para ofrecer o desaconsejar la realización de un determinado test o estudio en base a la evidencia disponible, aunque en la práctica habitual, muchas veces esto resulta complejo. Se le debe explicar a cada paciente los riesgos y beneficios de realizar un estudio y ante un determinado resultado, actuar en consecuencia. Lamentablemente, en Ginecología Oncológica sólo unas pocas patologías cuentan con la posibilidad de realizar screening poblacional de detección temprana, por lo que se espera con ansias avances en diagnóstico y tecnología que permitan ofrecer una luz ante estas enfermedades. La prevención cuaternaria tiene la finalidad de evitar los posibles daños ocasionados en la práctica habitual de la medicina, basándose en la evidencia científica disponible. Facultad de Ciencias Médicas

El cáncer de mama es el tipo de cáncer más frecuente y la causa más común de muerte por cáncer en mujeres a nivel mundial. El cáncer de mama causa más años de vida ajustados por discapacidad perdidos en mujeres que cualquier otro cáncer. Esta afirmación de la OMS permite enfatizar la importancia del diagnóstico temprano de una patología cuyo órgano diana permite accesibilidad desde la clínica y la metodología diagnóstica. En Argentina, ocupa el segundo lugar en frecuencia, seguido por el cáncer de colon, pulmón, cérvix y tiroides. Sin duda que los programas de detección para el cáncer ginecológico y mamaria deben estimularse y realizarse. Las políticas de salud pública no pueden postergar ni omitir estas intervenciones y, de esa forma, aspirar a una mejor calidad de vida de las mujeres afectadas., Facultad de Ciencias Médicas

Dimethyl sulfide (DMS) influences climate via cloud condensation nuclei (CCN) formation resulting from its oxidation products (mainly methanesulfonic acid, MSA, and sulfuric acid, H$_{2}$SO$_{4}$). Despite their importance, accurate prediction of MSA and H$_{2}$SO$_{4}$ from DMS oxidation remains challenging. With comprehensive experiments carried out in the Cosmics Leaving Outdoor Droplets (CLOUD) chamber at CERN, we show that decreasing the temperature from +25 to −10 °C enhances the gas-phase MSA production by an order of magnitude from OH-initiated DMS oxidation, while H$_{2}$SO$_{4}$ production is modestly affected. This leads to a gas-phase H$_{2}$SO$_{4}$-to-MSA ratio (H$_{2}$SO$_{4}$/MSA) smaller than one at low temperatures, consistent with field observations in polar regions. With an updated DMS oxidation mechanism, we find that methanesulfinic acid, CH$_{3}$S(O)OH, MSIA, forms large amounts of MSA. Overall, our results reveal that MSA yields are a factor of 2–10 higher than those predicted by the widely used Master Chemical Mechanism (MCMv3.3.1), and the NO$_{x}$ effect is less significant than that of temperature. Our updated mechanism explains the high MSA production rates observed in field observations, especially at low temperatures, thus, substantiating the greater importance of MSA in the natural sulfur cycle and natural CCN formation. Our mechanism will improve the interpretation of present-day and historical gas-phase H$_{2}$SO$_{4}$/MSA measurements.

Hate speech is a global phenomenon, but most hate speech datasets so far focus on English-language content. This hinders the development of more effective hate speech detection models in hundreds of languages spoken by billions across the world. More data is needed, but annotating hateful content is expensive, time-consuming and potentially harmful to annotators. To mitigate these issues, we explore data-efficient strategies for expanding hate speech detection into under-resourced languages. In a series of experiments with mono- and multilingual models across five non-English languages, we find that 1) a small amount of target-language fine-tuning data is needed to achieve strong performance, 2) the benefits of using more such data decrease exponentially, and 3) initial fine-tuning on readily-available English data can partially substitute target-language data and improve model generalisability. Based on these findings, we formulate actionable recommendations for hate speech detection in low-resource language settings., Accepted at EMNLP 2022 (Main Conference)

Aunque el término prevención cuaternaria no es conocido o dominado por una gran mayoría de los profesionales sanitarios, en los últimos años, ha existido un creciente e importante auge de este concepto y se ha expandido en todo el mundo. Cada médico debe tener la suficiente determinación para ofrecer o desaconsejar la realización de un determinado test o estudio en base a la evidencia disponible, aunque en la práctica habitual, muchas veces esto resulta complejo. Se le debe explicar a cada paciente los riesgos y beneficios de realizar un estudio y ante un determinado resultado, actuar en consecuencia. Lamentablemente, en Ginecología Oncológica sólo unas pocas patologías cuentan con la posibilidad de realizar screening poblacional de detección temprana, por lo que se espera con ansias avances en diagnóstico y tecnología que permitan ofrecer una luz ante estas enfermedades. La prevención cuaternaria tiene la finalidad de evitar los posibles daños ocasionados en la práctica habitual de la medicina, basándose en la evidencia científica disponible., Facultad de Ciencias Médicas

El cáncer de mama es el tipo de cáncer más frecuente y la causa más común de muerte por cáncer en mujeres a nivel mundial. El cáncer de mama causa más años de vida ajustados por discapacidad perdidos en mujeres que cualquier otro cáncer. Esta afirmación de la OMS permite enfatizar la importancia del diagnóstico temprano de una patología cuyo órgano diana permite accesibilidad desde la clínica y la metodología diagnóstica. En Argentina, ocupa el segundo lugar en frecuencia, seguido por el cáncer de colon, pulmón, cérvix y tiroides. Sin duda que los programas de detección para el cáncer ginecológico y mamaria deben estimularse y realizarse. Las políticas de salud pública no pueden postergar ni omitir estas intervenciones y, de esa forma, aspirar a una mejor calidad de vida de las mujeres afectadas. Facultad de Ciencias Médicas

Twitter data have become essential to Natural Language Processing (NLP) and social science research, driving various scientific discoveries in recent years. However, the textual data alone are often not enough to conduct studies: especially social scientists need more variables to perform their analysis and control for various factors. How we augment this information, such as users' location, age, or tweet sentiment, has ramifications for anonymity and reproducibility, and requires dedicated effort. This paper describes Twitter-Demographer, a simple, flow-based tool to enrich Twitter data with additional information about tweets and users. Twitter-Demographer is aimed at NLP practitioners and (computational) social scientists who want to enrich their datasets with aggregated information, facilitating reproducibility, and providing algorithmic privacy-by-design measures for pseudo-anonymity. We discuss our design choices, inspired by the flow-based programming paradigm, to use black-box components that can easily be chained together and extended. We also analyze the ethical issues related to the use of this tool, and the built-in measures to facilitate pseudo-anonymity.

Product discovery is a crucial component for online shopping. However, item-to-item recommendations today do not allow users to explore changes along selected dimensions: given a query item, can a model suggest something similar but in a different color? We consider item recommendations of the comparative nature (e.g. "something darker") and show how CLIP-based models can support this use case in a zero-shot manner. Leveraging a large model built for fashion, we introduce GradREC and its industry potential, and offer a first rounded assessment of its strength and weaknesses., Comment: Accepted at ACL 2022 (ECNLP)

Despite the success of large vision and language models (VLMs) in many downstream applications, it is unclear how well they encode compositional information. Here, we create the Attribution, Relation, and Order (ARO) benchmark to systematically evaluate the ability of VLMs to understand different types of relationships, attributes, and order. ARO consists of Visual Genome Attribution, to test the understanding of objects' properties; Visual Genome Relation, to test for relational understanding; and COCO & Flickr30k-Order, to test for order sensitivity. ARO is orders of magnitude larger than previous benchmarks of compositionality, with more than 50,000 test cases. We show where state-of-the-art VLMs have poor relational understanding, can blunder when linking objects to their attributes, and demonstrate a severe lack of order sensitivity. VLMs are predominantly trained and evaluated on large datasets with rich compositional structure in the images and captions. Yet, training on these datasets has not been enough to address the lack of compositional understanding, and evaluating on these datasets has failed to surface this deficiency. To understand why these limitations emerge and are not represented in the standard tests, we zoom into the evaluation and training procedures. We demonstrate that it is possible to perform well on retrieval over existing datasets without using the composition and order information. Given that contrastive pretraining optimizes for retrieval on datasets with similar shortcuts, we hypothesize that this can explain why the models do not need to learn to represent compositional information. This finding suggests a natural solution: composition-aware hard negative mining. We show that a simple-to-implement modification of contrastive learning significantly improves the performance on tasks requiring understanding of order and compositionality., ICLR 2023 Oral (notable-top-5%)

Language is constantly changing and evolving, leaving language models to become quickly outdated. Consequently, we should continuously update our models with new data to expose them to new events and facts. However, that requires additional computing, which means new carbon emissions. Do any measurable benefits justify this cost? This paper looks for empirical evidence to support continuous training. We reproduce existing benchmarks and extend them to include additional time periods, models, and tasks. Our results show that the downstream task performance of temporally adapted English models for social media data do not improve over time. Pretrained models without temporal adaptation are actually significantly more effective and efficient. However, we also note a lack of suitable temporal benchmarks. Our findings invite a critical reflection on when and how to temporally adapt language models, accounting for sustainability., Comment: 8 pages

Well-annotated data is a prerequisite for good Natural Language Processing models. Too often, though, annotation decisions are governed by optimizing time or annotator agreement. We make a case for nuanced efforts in an interdisciplinary setting for annotating offensive online speech. Detecting offensive content is rapidly becoming one of the most important real-world NLP tasks. However, most datasets use a single binary label, e.g., for hate or incivility, even though each concept is multi-faceted. This modeling choice severely limits nuanced insights, but also performance. We show that a more fine-grained multi-label approach to predicting incivility and hateful or intolerant content addresses both conceptual and performance issues. We release a novel dataset of over 40,000 tweets about immigration from the US and UK, annotated with six labels for different aspects of incivility and intolerance. Our dataset not only allows for a more nuanced understanding of harmful speech online, models trained on it also outperform or match performance on benchmark datasets., Comment: EMNLP 2022

Additional file 1 supplementary-material.pdf includes: ��� search strings for the selection of papers from Scopus and Web of Science databases ��� a table summing up characteristics of articles included in the systematic review ��� funnel plots of structural, functional and combination of aspects of social relationships ��� tables of subgroup analysis ��� code for the analysis

The aim of the present study is to evaluate the difference in terms of feasibility and detection rate of two magnetic resonance imaging (MRI) guided biopsy approaches (MRI fusion versus "in-bore" MRI) in a single tertiary centre.

We present recent results related to extensions and applications of Partially Observable Monte Carlo Planning to real-world problems involving robotic and cyber-physical systems. Experiments were performed in the context of two projects: "Dipartimenti di Eccellenza 2018-22", Italian Ministry of Education, Universities and Research, and "Safe Place: Sistemi IoT per ambienti di vita salubri e sicuri", POR FESR 2014-20. Three main research lines are developed. The first concerns the explanation and monitoring of POMCP policies by integrating symbolic and non-symbolic approaches. The second regards the applicability of POMCP to real robotic scenarios. The third concerns POMCP policy improvement and adaptation.

Welcome to EVALITA 2020! EVALITA is the evaluation campaign of Natural Language Processing and Speech Tools for Italian. EVALITA is an initiative of the Italian Association for Computational Linguistics (AILC, http://www.ai-lc.it) and it is endorsed by the Italian Association for Artificial Intelligence (AIxIA, http://www.aixia.it) and the Italian Association for Speech Sciences (AISV, http://www.aisv.it).

The 2021 SIGIR workshop on eCommerce is hosting the Coveo Data Challenge for "In-session prediction for purchase intent and recommendations". The challenge addresses the growing need for reliable predictions within the boundaries of a shopping session, as customer intentions can be different depending on the occasion. The need for efficient procedures for personalization is even clearer if we consider the e-commerce landscape more broadly: outside of giant digital retailers, the constraints of the problem are stricter, due to smaller user bases and the realization that most users are not frequently returning customers. We release a new session-based dataset including more than 30M fine-grained browsing events (product detail, add, purchase), enriched by linguistic behavior (queries made by shoppers, with items clicked and items not clicked after the query) and catalog meta-data (images, text, pricing information). On this dataset, we ask participants to showcase innovative solutions for two open problems: a recommendation task (where a model is shown some events at the start of a session, and it is asked to predict future product interactions); an intent prediction task, where a model is shown a session containing an add-to-cart event, and it is asked to predict whether the item will be bought before the end of the session., Comment: SIGIR eCOM 2021 Data Challenge

Gaseous sulfuric acid (SA) has received a lot of attention for its crucial role in atmospheric new particle formation (NPF). And for this reason, studies until now have mainly focused on daytime SA when most NPF events occur. While daytime SA production is driven by SO2 oxidation of OH radicals of photochemical origin, the formation of SA during nighttime and its potential influence on particle formation remains poorly understood. Here we present evidence for significant nighttime SA production in urban Beijing during winter, yielding concentrations between 1.0 and 3.0 × 106 cm−3. We found a high frequency (∼ 30 %) of nighttime SA events, which are defined by the appearance of a distinct SA peak observed between 20:00 and 04:00 local time, with the maximum concentration exceeding 1.0 × 106 cm−3. These events mostly occurred during unpolluted nights with a low vapor condensation sink. Furthermore, we found that under very clean conditions (visibility > 16.0 km) with abundant ozone (concentration > 2.0 × 1011 cm−3, ∼ 7 ppb), the overall sink of SA was strongly correlated with the products of O3, alkenes and SO2 concentrations, suggesting that the ozonolysis of alkenes played a major role in nighttime SA formation under such conditions. This is in light of previous studies showing that the ozonolysis of alkenes can form OH radicals and stabilized Criegee intermediates (SCIs), both of which are able to oxidize SO2 and thus lead to SA formation. However, we also need to point out that there exist additional sources of SA under more polluted conditions, which are not investigated in this study. Moreover, there was a strong correlation between SA concentration and the number concentration of sub-3 nm particles on both clean and polluted nights. Different from forest environments, where oxidized biogenic vapors are the main driver of nighttime clustering, our study demonstrates that the formation of nighttime cluster mode particles in urban environments is mainly driven by nighttime SA production.

Atmospheric organic vapors play essential roles in the formation of secondary organic aerosol. Source identification of these vapors is thus fundamental to understanding their emission sources and chemical evolution in the atmosphere and their further impact on air quality and climate change. In this study, a Vocus proton-transfer-reaction time-of-flight mass spectrometer (PTR-TOF) was deployed in two forested environments, the Landes forest in southern France and the boreal forest in southern Finland, to measure atmospheric organic vapors, including both volatile organic compounds (VOCs) and their oxidation products. For the first time, we performed binned positive matrix factorization (binPMF) analysis on the complex mass spectra acquired with the Vocus PTR-TOF and identified various emission sources as well as oxidation processes in the atmosphere. Based on separate analysis of low- and high-mass ranges, 15 PMF factors in the Landes forest and nine PMF factors in the Finnish boreal forest were resolved, showing a high similarity between the two sites. Particularly, terpenes and various terpene reaction products were separated into individual PMF factors with varying oxidation degrees, such as lightly oxidized compounds from both monoterpene and sesquiterpene oxidation, monoterpene-derived organic nitrates, and monoterpene more oxidized compounds. Factors representing monoterpenes dominated the biogenic VOCs in both forests, with lower contributions from the isoprene factors and sesquiterpene factors. Factors of the lightly oxidized products, more oxidized products, and organic nitrates of monoterpenes/sesquiterpenes accounted for 8 %-12% of the measured gas-phase organic vapors in the two forests. Based on the interpretation of the results relating to oxidation processes, further insights were gained regarding monoterpene and sesquiterpene reactions. For example, a strong relative humidity (RH) dependence was found for the behavior of sesquiterpene lightly oxidized compounds. High concentrations of these compounds only occur at high RH; yet similar behavior was not observed for monoterpene oxidation products.

New particle formation in the Arctic atmosphere is an important source of aerosol particles. Understanding the processes of Arctic secondary aerosol formation is crucial due to their significant impact on cloud properties and therefore Arctic amplification. We observed the molecular formation of new particles from low-volatility vapors at two Arctic sites with differing surroundings. In Svalbard, sulfuric acid (SA) and methane sulfonic acid (MSA) contribute to the formation of secondary aerosol and to some extent to cloud condensation nuclei (CCN). This occurs via ion-induced nucleation of SA and NH3 and subsequent growth by mainly SA and MSA condensation during springtime and highly oxygenated organic molecules during summertime. By contrast, in an ice-covered region around Villum, we observed new particle formation driven by iodic acid but its concentration was insufficient to grow nucleated particles to CCN sizes. Our results provide new insight about sources and precursors of Arctic secondary aerosol particles.

Iodic acid (HIO3) is known to form aerosol particles in coastal marine regions, but predicted nucleation and growth rates are lacking. Using the CERN CLOUD (Cosmics Leaving Outdoor Droplets) chamber, we find that the nucleation rates of HIO3 particles are rapid, even exceeding sulfuric acid-ammonia rates under similar conditions. We also find that ion-induced nucleation involves IO3- and the sequential addition of HIO3 and that it proceeds at the kinetic limit below +10 degrees C. In contrast, neutral nucleation involves the repeated sequential addition of iodous acid (HIO2) followed by HIO3, showing that HIO2 plays a key stabilizing role. Freshly formed particles are composed almost entirely of HIO3, which drives rapid particle growth at the kinetic limit. Our measurements indicate that iodine oxoacid particle formation can compete with sulfuric acid in pristine regions of the atmosphere.

8 pags., 5 figs., Iodic acid (HIO) is known to form aerosol particles in coastal marine regions, but predicted nucleation and growth rates are lacking. Using the CERN CLOUD (Cosmics Leaving Outdoor Droplets) chamber, we find that the nucleation rates of HIOparticles are rapid, even exceeding sulfuric acid-ammonia rates under similar conditions. We also find that ion-induced nucleation involves IOand the sequential addition of HIOand that it proceeds at the kinetic limit below +10°C. In contrast, neutral nucleation involves the repeated sequential addition of iodous acid (HIO) followed by HIO, showing that HIOplays a key stabilizing role. Freshly formed particles are composed almost entirely of HIO, which drives rapid particle growth at the kinetic limit. Our measurements indicate that iodine oxoacid particle formation can compete with sulfuric acid in pristine regions of the atmosphere., We thank the European Organization for Nuclear Research (CERN) for supporting CLOUD with important technical and financial resources and for providing a particle beam from the CERN Proton Synchrotron. This research has received support from the Academy of Finland (projects 316114, 307331, 310682, 266388, 3282290, 306853, 296628, 229574, 333397, 326948, and 1325656); the European Research Council (projects 692891, 616075, 764991, 316662, 742206, and 714621); CSC – Finnish IT center; the EC Seventh Framework Programme and the EU H2020 programme Marie Skłodowska Curie ITN “CLOUD-TRAIN” (316662) and “CLOUD-MOTION” (764991); Austrian Science Fund (FWF) (J3951-N36 and P27295-N20); the Swiss National Science Foundation (20FI20_159851, 200021_169090, 200020_172602, and 20FI20_172622); the U.S. National Science Foundation (grants AGS1447056, AGS1439551, AGS1801574, AGS1620530, AGS1801897, AGS153128, AGS1649147, AGS1801280, AGS1602086, and AGS1801329); MSCA H2020 COFUND-FP-CERN2014 fellowship (665779); German Federal Ministry of Education and Research: CLOUD-16 (01LK1601A); Portuguese Foundation for Science and Technology (CERN/FIS-COM/0014/2017); Academy of Finland Centre of Excellence in Atmospheric Sciences (grant 272041); European Regional Development Fund (project MOBTT42); Jiangsu Collaborative Innovation Center for Climate Change; Yangtze River Delta Atmosphere and Earth System Science National Observation and Research Station; Estonian Research Council (project PRG714); Hungarian National Research, Development and Innovation Office (K116788 and K132254); NASA Graduate Fellowship (NASA-NNX16AP36H); and ACTRIS 2TNA H2020 OCTAVE (654109).

CLIP (Contrastive Language-Image Pre-training) is a very recent multi-modal model that jointly learns representations of images and texts. The model is trained on a massive amount of English data and shows impressive performance on zero-shot classification tasks. Training the same model on a different language is not trivial, since data in other languages might be not enough and the model needs high-quality translations of the texts to guarantee a good performance. In this paper, we present the first CLIP model for the Italian Language (CLIP-Italian), trained on more than 1.4 million image-text pairs. Results show that CLIP-Italian outperforms the multilingual CLIP model on the tasks of image retrieval and zero-shot classification.

Understanding differences of viewpoints across corpora is a fundamental task for computational social sciences. In this paper, we propose the Sliced Word Embedding Association Test (SWEAT), a novel statistical measure to compute the relative polarization of a topical wordset across two distributional representations. To this end, SWEAT uses two additional wordsets, deemed to have opposite valence, to represent two different poles. We validate our approach and illustrate a case study to show the usefulness of the introduced measure., Published as a conference paper at EMNLP2021

This article examines the emergence of solidarity from interactions between professionals competing for collaboration. Research on multiplex collaboration networks has shown that economic exchange can elicit solidarity when mediated by trust but did not consider the effect of competition. To fill this gap, we built an agent-based model that simulates the evolution of a multiplex network of collaboration, trust, and support expectations. Simulations show that while resource heterogeneity is key for collaboration, competition for attractive collaboration partners penalizes low-resource professionals, who are less connected and highly segregated. Heterogeneous resource distribution can trigger segregation because of preferential selection of resourceful peers and reciprocity. Interestingly, we also found that low-resource professionals can reduce their marginalization by building in-group mutual support expectations.

We study the relation between Age of Acquisition (AoA) and semantic change, quantifying it using word embeddings extracted from the Corpus of Historic American (CoHA) English (Davis, 2010), a diachronic corpus of American English spanning the period 1800-2009. Previous research (Monagahn, 2014; Monaghan & Roberts, 2019) looked at the relation between language evolution and AoA, but quantified language evolution through cognate proliferation and rate of lexical substitution for a limited set of words. We extend this line of work by exploring semantic change, using a highly scalable corpus-based method to quantify it for thousands of words. Our results show that the relation between AoA and language change also holds for semantic shift on a smaller time scale than previously established. We first bin the CoHA in six slices containing a similar number of tokens and then align embedding spaces across time slices using Temporal Word Embeddings with a Compass (TWEC, Di Carlo et al. (2019)). TWEC extends the Continuous Bag of Words model (CBoW, Mikolov et al. (2013)) and aligns different vector spaces to a shared coordinate system in such a way that the different embedded slices can be compared. We consider two different measures of semantic change, tracking the coherence of the embeddings of a word over time (vector coherence, VC) and the local neighborhood coherence (LNC) of a word over time, in line with the work by Hamilton et al. (2016). Words with high VC have similar embeddings over time while words with high LNC have a consistent relation with their semantic neighbors across time slices. For each measure, we also compute a quasi-random baseline to ensure that any observed effects are robust. We fit regression models using AoA as the dependent variable and focus on the 8296 words for which we have AoA norms (Kuperman et al., 2012), concreteness ratings (Brysbaert et al., 2014), frequency estimates from SUBTLEX US (Brysbaert and New, 2009) and which occurred in all temporal slices with a frequency count >= 25 in order to estimate reliable embeddings. We add our target measures separately to a baseline model including log frequency, OLD20 (Yarkoni et al., 2008), length in letters, concreteness, dominant Part of Speech, and frequency in the CoHA. We then compare the degree to which they improve the fit over the baseline model, observing that both explain additional variance over the base model and that words with lower rates of semantic change are acquired earlier. In spite of the fact that VC has a stronger correlation with AoA, LNC explains a higher proportion of extra variance over the base model. Random baseline measures, on the contrary, have far weaker correlations with AoA. Qualitative analyses show that words with high VC tend to be basic level, concrete concepts (e.g. dress, water, fish, door). On the contrary, among concepts with high LNC we primarily see words referring to body parts, parts of the day, time expressions, family relations, trees and fruit. Phonosymbolic words and words including phonaestemes also tend to have high LNC (e.g. howl, shriek, glimmer, fling, snore, ...), suggesting that words with high LNC may capture statistical regularities in form-meaning relations. Future studies should investigate this possibility further. Summing up, we report evidence of a robust relation between semantic change (quantified with a scalable, corpus-based method for thousands of words) and AoA, with early acquired words showing higher semantic coherence over time, beyond the effects of standard covariates. We also show that different measures of semantic change highlight different aspects of this phenomenon, characterizing it as a multidimensional construct. On the basis of preliminary qualitative analyses, we hypothesize that this relation signals that words with consistent meanings over time refer to aspects of the world which are more salient to children and serve to structure later vocabulary learning, as they denote stable concepts, whose meaning is less culturally determined. Giovanni Cassani (Tilburg University), Federico Bianchi (Bocconi University) & Marco Marelli (University of Milan – Bicocca) Technical details of the computational model: The TWEC (https://github.com/valedica/twec) model is implemented using the gensim (https://radimrehurek.com/gensim/) library and is based on the CBoW model, which is instantiated in a neural network with one hidden layer. The CBoW model uses two matrices to learn word embeddings, a target matrix and a context matrix. TWEC exploits this aspect by first training a general embedding using the whole corpus, ignoring the time dimension. This embedding space is the compass, i.e. a general representation to which the embedding spaces derived from each time slice can be aligned. The context matrix of the compass is extracted and used to initialize (and freeze) the second matrix of a slice-specific CBoW e bedding space. This approach ensures that all slice-specific embedding spaces share the same context matrix, making the slice-specific embeddings aligned. The reported analysis is performed on word embeddings with 100 dimensions, learned using a window of 5 words to the left and to the right of the target word. However, we performed a small grid search exploring windows of 3 and 20 words, as well as embeddings with 40 dimensions. Results do not depend on these hyper-parameter choices. Parameter optimization is carried out with a learning rate of 0.025 and 10 negative samples. We use 5 iterations to train the compass embeddings and 5 iterations to train the slice specific embeddings. We initialize all the other hyper-parameters using the default settings provided by the library. References: Brysbaert, M., & New, B. (2009). Moving beyond Kučera and Francis: A critical evaluation of current word frequency norms and the introduction of a new and improved word frequency measure for American English. Behavior research methods, 41(4), 977-990. Brysbaert, M., Warriner, A. B., & Kuperman, V. (2014). Concreteness ratings for 40 thousand generally known English word lemmas. Behavior research methods, 46(3), 904-911. Di Carlo, V., Bianchi, F., & Palmonari, M. (2019, July). Training temporal word embeddings with a compass. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 33, pp. 6326-6334). Hamilton, W. L., Leskovec, J., & Jurafsky, D. (2016). Cultural Shift or Linguistic Drift? Comparing Two Computational Measures of Semantic Change. In Proceedings of the Conference on Empirical Methods in Natural Language Processing. Conference on Empirical Methods in Natural Language Processing (Vol. 2016, p. 2116). Kuperman, V., Stadthagen-Gonzalez, H., & Brysbaert, M. (2012). Age-of-acquisition ratings for 30,000 English words. Behavior research methods, 44(4), 978-990. Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S., & Dean, J. (2013). Distributed representations of words and phrases and their compositionality. In Advances in neural information processing systems (pp. 3111-3119). Monaghan, P. (2014). Age of acquisition predicts rate of lexical evolution. Cognition, 133(3), 530-534. Monaghan, P. and Roberts, S. G. (2019). Cognitive influences in language evolution: Psycholinguistic predictors of loan words borrowing. Cognition, 186, 147-158. Yarkoni, T., Balota, D., & Yap, M. (2008). Moving beyond Coltheart’s N: A new measure of orthographic similarity. Psychonomic bulletin & review, 15(5), 971-979.

Highly oxygenated organic molecules (HOMs) contribute substantially to the formation and growth of atmospheric aerosol particles, which affect air quality, human health and Earth s climate. HOMs are formed by rapid, gasphase autoxidation of volatile organic compounds (VOCs) such as -pinene, the most abundant monoterpene in the atmosphere. Due to their abundance and low volatility, HOMs can play an important role in new-particle formation (NPF) and the early growth of atmospheric aerosols, even without any further assistance of other low-volatility compounds such as sulfuric acid. Both the autoxidation reaction forming HOMs and their NPF rates are expected to be strongly dependent on temperature. However, experimental data on both effects are limited. Dedicated experiments were performed at the CLOUD (Cosmics Leaving OUtdoor Droplets) chamber at CERN to address this question. In this study, we show that a decrease in temperature (from C25 to 50 C) results in a reduced HOM yield and reduced oxidation state of the products, whereas the NPF rates (J1:7 nm) increase substantially. Measurements with two different chemical ionization mass spectrometers (using nitrate and protonated water as reagent ion, respectively) provide the molecular composition of the gaseous oxidation products, and a two-dimensional volatility basis set (2D VBS) model provides their volatility distribution. The HOM yield decreases with temperature from 6.2% at 25 C to 0.7% at 50 C. However, there is a strong reduction of the saturation vapor pressure of each oxidation state as the temperature is reduced. Overall, the reduction in volatility with temperature leads to an increase in the nucleation rates by up to 3 orders of magnitude at 50 C compared with 25 C. In addition, the enhancement of the nucleation rates by ions decreases with decreasing temperature, since the neutral molecular clusters have increased stability against evaporation. The resulting data quantify how the interplay between the temperature-dependent oxidation pathways and the associated vapor pressures affect biogenic NPF at the molecular level. Our measurements, therefore, improve our understanding of pure biogenic NPF for a wide range of tropospheric temperatures and precursor concentrations. publishedVersion

Highly oxygenated organic molecules (HOMs) contribute substantially to the formation and growth of atmospheric aerosol particles, which affect air quality, human health and Earth's climate. HOMs are formed by rapid, gas-phase autoxidation of volatile organic compounds (VOCs) such as alpha-pinene, the most abundant monoterpene in the atmosphere. Due to their abundance and low volatility, HOMs can play an important role in new-particle formation (NPF) and the early growth of atmospheric aerosols, even without any further assistance of other low-volatility compounds such as sulfuric acid. Both the autoxidation reaction forming HOMs and their NPF rates are expected to be strongly dependent on temperature. However, experimental data on both effects are limited. Dedicated experiments were performed at the CLOUD (Cosmics Leaving OUtdoor Droplets) chamber at CERN to address this question. In this study, we show that a decrease in temperature (from + 25 to -50 degrees C) results in a reduced HOM yield and reduced oxidation state of the products, whereas the NPF rates (J(1.7) (nm)) increase substantially. Measurements with two different chemical ionization mass spectrometers (using nitrate and protonated water as reagent ion, respectively) provide the molecular composition of the gaseous oxidation products, and a two-dimensional volatility basis set (2D VBS) model provides their volatility distribution. The HOM yield decreases with temperature from 6.2 % at 25 degrees C to 0.7 % at -50 degrees C. However, there is a strong reduction of the saturation vapor pressure of each oxidation state as the temperature is reduced. Overall, the reduction in volatility with temperature leads to an increase in the nucleation rates by up to 3 orders of magnitude at -50 degrees C compared with 25 degrees C. In addition, the enhancement of the nucleation rates by ions decreases with decreasing temperature, since the neutral molecular clusters have increased stability against evaporation. The resulting data quantify how the interplay between the temperature-dependent oxidation pathways and the associated vapor pressures affect biogenic NPF at the molecular level. Our measurements, therefore, improve our understanding of pure biogenic NPF for a wide range of tropospheric temperatures and precursor concentrations.

This paper addresses the challenge of leveraging multiple embedding spaces for multi-shop personalization, proving that zero-shot inference is possible by transferring shopping intent from one website to another without manual intervention. We detail a machine learning pipeline to train and optimize embeddings within shops first, and support the quantitative findings with additional qualitative insights. We then turn to the harder task of using learned embeddings across shops: if products from different shops live in the same vector space, user intent - as represented by regions in this space - can then be transferred in a zero-shot fashion across websites. We propose and benchmark unsupervised and supervised methods to "travel" between embedding spaces, each with its own assumptions on data quantity and quality. We show that zero-shot personalization is indeed possible at scale by testing the shared embedding space with two downstream tasks, event prediction and type-ahead suggestions. Finally, we curate a cross-shop anonymized embeddings dataset to foster an inclusive discussion of this important business scenario., accepted at 2020 SIGIR Workshop On eCommerce

L'intelligenza artificiale cerca di spiegare come gli agenti intelligenti si comportano. Il linguaggio è uno dei media di comunicazioni più importanti e studiare delle teorie che permettano di definire il significato di espressioni naturali è molto importante. I linguisti hanno usato con successo linguaggi artificiali basati su logiche, ma una theory che ha avuto un impatto significativo in intelligenza artificiale è la semantica distribuzionale. La semantica distribuzionale afferma che il significato di espressioni in linguaggio naturale può essere derivato dal contesto in cui tali espressioni compaiono. Questa teoria è stata implementata da algoritmi che permettono di generare rappresentazioni vettoriali delle espressioni del linguaggio natural in modo che espressioni simili vengano rappresentate con vettori simili. Negli ultimi anni, gli scienziati cognitivi hanno sottolineato che queste rappresentazioni sono correlate con l'associative learning e che sono anche in grado di catturare bias e stereotype del testo. Diventa quindi importante trovare metodologie per comparare rappresentazioni che arrivano da sorgenti diverse. Ad esempio, usare questi algoritmi su testi di periodi differenti genera rappresentazioni differenti: visto che il linguaggio muta nel tempo, trovare delle metododoloie per comparare come le parole si sono mosse è un task imporante per l'intelligenza artificiale (e.g., la parola "amazon" ha cambiato il suo significato principale negli ultimi anni) In questa tesi, introduciamo un modello comparative basato su testi che permette di comparare rappresentazioni di sorgenti diverse generate con la semantica distribuzionale. Proponiamo un modello che è efficiente ed efficace e mostriamo che possiamo anche gestire nomi di entità e non solo paorle, superando problemi legati all'ambiguità del linguaggio. Alla fine, mostriamo che è possibile combinare questi metodi con approcci logici e fare comparazioni utilizzando costrutti logici. One of the main goals of artificial intelligence is understanding how intelligent agent acts. Language is one of the most important media of communication, and studying theories that can account for the meaning of natural language expressions is an important task. Language is one of the most important media of communication, and studying theories that can account for the meaning of natural language expressions is a crucial task in artificial intelligence. Distributional semantics states that the meaning of natural language expressions can be derived from the context in which the expressions appear. This theory has been implemented by algorithms that generate vector representations of natural language expressions that represent similar natural language expressions with similar vectors. In the last years, several cognitive scientists have shown that these representations are correlated with associative learning and they capture cognitive biases and stereotypes as they are encoded in text corpora. If language is encoding important aspects of cognition and our associative knowledge, and language usage change across the contexts, the comparison of language usage in different contexts may reveal important associative knowledge patterns. Thus, if we want to reveal these patterns, we need ways to compare distributional representations that are generated from different text corpora. For example, using these algorithms on textual documents from different periods will generate different representations: since language evolves during time, finding a way to compare words that have shifted over time is a valuable task for artificial intelligence (e.g., the word "Amazon" has changed its prevalent meaning during the last years). In this thesis, we introduce a corpus-based comparative model that allows us to compare representations of different sources generated under the distributional semantic theory. We propose a model that is both effective and efficient, and we show that it can also deal with entity names and not just words, overcoming some problems that follow from the ambiguity of natural language. Eventually, we combine these methods with logical approaches. We show that we can do logical reasoning on these representations and make comparisons based on logical constructs.