In a decision-making scenario, one group member may need, independently from the other members, to choose an item, e.g., a restaurant, that will be experienced by the group. In a prior research on restaurant recommendation, we have identified two primary tasks of such an organizer of a lunch/dinner event, who is in charge of selecting a proper restaurant for the group: anticipating the other group members’ preferences, to properly enter these preferences in the recommender system, and reconciling incompatible preferences, if they arise in the elicitation phase. To support the first task, we augmented a group recommender system with a machine learning model that predicts group members’ food preferences, about dishes that can be consumed in the restaurant, based on other available information about the members (demographics and preferred cuisine). However, in a user study, we found that supporting functionality to be not effective in improving the quality of the choices made by the organizer. In this paper, we investigate the causes of this poor performance. We analyze the possibility that the poor performance may relate to: the deployed ML models used for food preference prediction, the amount of data about food preferences used for training the ML model, and the complexity of the preference anticipation task. The results of our experiments suggest that neither the ML model nor the scarcity of the preference data is responsible for the observed poor value of the implemented preference recalling function. While a user study seems to confirm the impossibility of accurately predicting food preferences from the considered user’s characteristics (demographics and preferred cuisine).

While most recommender systems cater to individual users’ needs, there are numerous situations where these systems are needed to meet groups’ demands. These systems are broadly labelled as Group Recommender Systems (GRSys). Traits like interpersonal relationships, group mood, and emotional contagion are essential to fulfilling the group’s needs. However, the group’s characteristics are frequently ill-defined and dynamic and are typically absent from systems modeling. Moreover, GRSys must maneuver between the needs of the group and the individuals when opinions differ and can contradict each other. The third edition of GMAP proposes consolidating a community of scholars interested in group modeling, adaptation, and personalization. Through the workshop, researchers continue their examination of the difficulties and possibilities of creating efficient procedures and instruments to facilitate collective decision-making. GMAP 2024 offered this unique opportunity to gather scholars from different fields to enrich discussions over GRSys’ research. The workshop also allowed attendees to strengthen their networks and establish new connections conducive to cutting-edge collaborative research.

Adaptive and personalized systems have become pervasive technologies, gradually playing an increasingly important role in our daily lives. Indeed, we are now accustomed to interacting with algorithms that leverage the power of Language Models (LLMs) to assist us in various scenarios, from services suggesting music or movies to personal assistants proactively supporting us in complex decision-making tasks. As these technologies continue to shape our everyday experiences, ensuring that the internal mechanisms guiding these algorithms are transparent and comprehensible becomes imperative. The EU General Data Protection Regulation (GDPR) recognizes the users’ right to explanation when confronted with intelligent systems, highlighting the significance of this aspect. Regrettably, current research often prioritizes the maximization of personalization strategy effectiveness, such as recommendation accuracy, at the expense of model explainability. To address this concern, the workshop aims to provide a platform for in-depth discussions on challenges, problems, and innovative research approaches in the field. The workshop specifically focuses on investigating the role of transparency and explainability in recent methodologies for constructing user models and developing personalized and adaptive systems.

BACKGROUND AND OBJECTIVES A better understanding of the factors influencing D-dimer levels in code stroke patients is needed to guide further investigations of concomitant thrombotic conditions. This study aimed to investigate the impact of time from symptom onset and other factors on D-dimer levels in patients with acute ischemic stroke (AIS) or transient ischemic attack (TIA). METHODS Data on consecutive AIS and TIA patients treated at our tertiary-care stroke center between January 2015 and December 2020 were retrospectively assessed. Patients with available D-dimer levels were evaluated for eligibility. Multivariable non-linear regression analyses were performed. RESULTS In total, 2467 AIS patients and 708 TIA patients were included. The median D-dimer levels differed between the AIS and TIA groups (746 µg/L [interquartile range 381-1468] versus 442 µg/L [interquartile range 244-800], p<0.001). In AIS patients, an early increase in D-dimer levels was demonstrated within the first 6 h (standardized beta coefficient [β] 0.728; 95% confidence interval [CI] 0.324-1.121). This was followed by an immediate decrease (β -13.022; 95% CI -20.401 to -5.643) and then by a second, late increase after 35 h (β 11.750; 95% CI 4.71-18.791). No time-dependent fluctuation in D-dimer levels was observed in TIA patients. CONCLUSION The time from symptom onset may affect D-dimer levels in patients with AIS but not those with TIA. Further studies confirming these findings and validating time-specific variations are needed to enable D-dimer levels to be used efficiently as an acute stroke and thrombotic risk biomarker.

[This corrects the article DOI: 10.1021/acsomega.3c08932.].

Fokus rada je na motivaciji kao jednom od glavnih čimbenika u poučavanju i učenju stranih jezika, kao i na načinu na koji se učenici motiviraju za njihovo učenje. Kao potpora empirijskom istraživanju testirani su aspekti motivacije (intrinzične i ekstrinzične) kako bi se ispitao njihov doprinos tijekom online nastave izazvane Covidom-19 i nakon povratka u škole. U prilog dobivenim rezultatima dane su preporuke koje bi mogle poslužiti budućoj organizaciji online nastave, ali i poslužiti pedagoško-psihološkim i didaktičko-metodičkim kompetencijama nastavnika. Istraživanje je potvrdilo povećanje i intrinzične i ekstrinzične motivacije nakon povratka u školu, za razliku od razdoblja provedenoga u online nastavi s izuzetkom aspekta učenja njemačkoga jezika. Prisutna intrinzična motivacija za učenje njemačkoga jezika u predikciji učenja njemačkoga jezika daje bolje rezultate iz razdoblja 2019. u odnosu na razdoblje 2021. Zanimljivo je da su učenici 2019. više uživali u samostalnom učenju tijekom online nastave nego danas u učionicama. Utjecaj analiziranih aspekata ekstrinzične motivacije proveden je uz pomoć korištenih metoda i nastavnih oblika rada. Utvrđene vrijednosti pokazuju veću snagu u 2021. godini u odnosu na 2019. godinu. Implikacije nedostatka digitalnih kompetencija ozbiljno zahtijevaju restrukturiranje obrazovnoga rada učitelja.

Data curation is an essential component of large-scale pretraining. In this work, we demonstrate that jointly selecting batches of data is more effective for learning than selecting examples independently. Multimodal contrastive objectives expose the dependencies between data and thus naturally yield criteria for measuring the joint learnability of a batch. We derive a simple and tractable algorithm for selecting such batches, which significantly accelerate training beyond individually-prioritized data points. As performance improves by selecting from larger super-batches, we also leverage recent advances in model approximation to reduce the associated computational overhead. As a result, our approach--multimodal contrastive learning with joint example selection (JEST)--surpasses state-of-the-art models with up to 13$\times$ fewer iterations and 10$\times$ less computation. Essential to the performance of JEST is the ability to steer the data selection process towards the distribution of smaller, well-curated datasets via pretrained reference models, exposing the level of data curation as a new dimension for neural scaling laws.

A two-year experiment was conducted with a local maize hybrid under full (F) and deficit (D) drip irrigation and rainfed conditions (R) to estimate maize evapotranspiration in Bosnia and Herzegovina (BiH). Three approaches, namely, A&P, SIMDualKc (SD), and vegetation index (VI), to estimate the actual crop coefficient (Kc act), the actual basal crop coefficient (Kcb act), and the actual crop evapotranspiration (ETc act), were applied with the dual crop coefficient method and remote sensing (RS) data for the first time. While Kcb act from all approaches matched FAO56 tabulated values, SD showed differences in comparison to A&P of up to 0.24 in D and R conditions, especially in the initial and mid-season stages. VI demonstrated very good performance in all treatments. In F, the obtained Kc act for all approaches during the initial and end stages were higher than the tabulated values, ranging from 0.71 to 0.87 for the Kc ini act and from 0.80 to 1.06 for the Kc end act, while the mid-season period showed very good agreement with the literature. The maize crop evapotranspiration range is 769–813 mm, 480–752 mm, and 332–618 mm for F, D, and R, respectively. The results confirmed the suitability of both approaches (SD and VI) to estimate maize crop evapotranspiration under F, with the VI approach demonstrating an advantage in calculating Kcb act, Kc act, and ETc act values under water stress conditions. The higher observed yields (67.6%) under irrigation conditions emphasize the need to transition from rainfed to irrigation-dependent agriculture in BiH, even for drought-resistant crops like maize.

Electromagnetic levitation represents a contemporary line of technology with a wide spectrum of applications in several areas of engineering. Regarding the increased demands for this technology in-depth research into its dynamics and the influence of different material characteristics used in these systems is needed. This paper presents simulation results of electrodynamic levitation system regarding different types of materials used for levitating disc. The main intention is to provide a comparison of electrodynamic levitation systems with different materials used for levitating disc in order to improve the system itself. Therefore, the impact of different materials used for the levitating disc of the electrodynamic levitation system would be investigated through some parameters of interest such as an analysis of the electromagnetic force, disk displacement, and the time required to achieve a stable disk position.

This research aimed to analyze the impact of two different non-Saccharomyces yeast species on the aromatic profile of red wines made from the cv. Babić (Vitis vinifera L.) red grape variety. The grapes were obtained from two positions in the Middle and South of Dalmatia. This study compared a control treatment with the Saccharomyces cerevisiae (Sc) strain as a type of sequential inoculation treatment with Lachancea thermotolerans (Lt x Sc) and Torulaspora delbrueckii (Td x Sc). The focus was on the basic wine parameters and volatile aromatic compound concentrations determined using the SPME-Arrow-GC/MS method. The results revealed significant differences in cis-linalool oxide, geraniol, neric acid, and nerol, which contribute to the sensory profile with floral and rose-like aromas; some ethyl esters, such as ethyl furoate, ethyl hexanoate, ethyl lactate, ethyl 2-hydroxy-3-methylbutanoate, ethyl 3-hydroxy butanoate, diethyl glutarate, and diethyl succinate, contribute to the aromatic profile with fruity, buttery, overripe, or aging aromas. A sensory evaluation of wines confirmed that Td x Sc treatments exhibited particularly positive aromatic properties together with a more intense fullness, harmony, aftertaste, and overall impression.

This paper presents a novel event-triggered control (ETC) design framework based on measured $\mathcal{L}_{p}$ norms. We consider a class of systems with finite $\mathcal{L}_{p}$ gain from the network-induced error to a chosen output. The $\mathcal{L}_{p}$ norms of the network-induced error and the chosen output since the last sampling time are used to formulate a class of triggering rules. Based on a small-gain condition, we derive an explicit expression for the $\mathcal{L}_{p}$ gain of the resulting closed-loop systems and present a time-regularization, which can be used to guarantee a lower bound on the inter-sampling times. The proposed framework is based on a different stability- and triggering concept compared to ETC approaches from the literature, and thus may yield new types of dynamical properties for the closed-loop system. However, for specific output choices it can lead to similar triggering rules as “standard” static and dynamic ETC approaches based on input-to-state stability and yields therefore a novel interpretation for some of the existing triggering rules. We illustrate the proposed framework with a numerical example from the literature.

Mammography is the leading methodology used to diagnose breast cancer. Effective, cheap and reliable, the mammography can be used to screen large populations, if the imagery produced can be analysed efficiently. State-of-the-art generative artificial intelligence approaches can be used to create tools able to aid in this task. Here we present a study focused on the emerging research topic of the application of generative diffusion models to the task of anomaly detection and we apply if for detecting anomalies on mammograms. Diffusion models exhibit promising results in making pixel-level predictions with image level annotations, but no such application has been published so far regarding mammography. We have, therefore, developed a novel approach utilizing U-net backbone that is able to generate mammograms with Fréchet Inception Distance (FID) of 14.62. We showed its ability to perform anomaly detection with Intersection over Union (IoU) of 0.195 which demonstrates the viability of our approach for early-stage research.

Calocucullia celsiae (Herrich-Schäffer, [1850]) is an easily recognizable noctuid species, differing from all other similar species in its subfamily. Within this survey, it was recorded at two localities in Bosnia and Herzegovina. Two specimens were collected near Hutovo village in the southern Herzegovina region in April 2023, and a single specimen was collected near Zoranovići village in the central part of the country in May 2023. These are the first records of this species for Bosnia and Herzegovina and the westernmost known data on the presence of this species on the Balkan Peninsula.

New accurate, precise, and sensitive spectrophotometric method were developed for the assay of L-ascorbic acid in pharmaceutical preparations. The determination of L-ascorbic acid was based on its oxidation by potassium peroxydisulfate in the presence of Ag(I) as a catalyst. The molar absorptivity of the proposed method was found to be 8.61 · 103 L mol-1 cm-1 at 248 nm. Beer's law was obeyed in the concentration range of 0.46–20.0 μg mL–1. Other compounds commonly found in vitamin C and multivitamin products did not interfere with the determination of L-ascorbic acid. The proposed method was successfully applied for the determination of L-ascorbic acid in pharmaceutical formulations. The results obtained with the proposed method showed good agreement with those given by the titrimetric method using iodine.

The availability of powerful open-source large language models (LLMs) opens exciting use-cases, such as automated personal assistants that adapt to the user's unique data and demands. Two key desiderata for such assistants are personalization-in the sense that the assistant should reflect the user's own style-and privacy-in the sense that users may prefer to always store their personal data locally, on their own computing device. We present a new design for such an automated assistant, for the specific use case of personal assistant for email generation, which we call Panza. Specifically, Panza can be both trained and inferenced locally on commodity hardware, and is personalized to the user's writing style. Panza's personalization features are based on a new technique called data playback, which allows us to fine-tune an LLM to better reflect a user's writing style using limited data. We show that, by combining efficient fine-tuning and inference methods, Panza can be executed entirely locally using limited resources-specifically, it can be executed within the same resources as a free Google Colab instance. Finally, our key methodological contribution is a careful study of evaluation metrics, and of how different choices of system components (e.g. the use of Retrieval-Augmented Generation or different fine-tuning approaches) impact the system's performance.