Browse Articles

Rethink context engineering using an attention-based architecture

Abstract Accurate prediction of user actions is essential for optimizing digital platform workflows, enabling proactive recommendations, resource prefetching, and intelligent user assistance. Traditional Markov chain-based methods, though widely used for modeling sequential behavior, are fundamentally limited in capturing the complexity, long-range dependencies, and multi-objective nature of real-world user interactions. This paper introduces a multi-task attention-based transformer architecture for sequential API recommendation that addresses these gaps in robustness and generalizability. The core insight is that user behavior on enterprise platforms is driven by latent intent: users with different goals—such as executing a machine learning pipeline, conducting data analysis, managing user accounts, or generating quick visualizations—exhibit systematically different sequential patterns across functional API categories. Our framework exploits this structure through a shared transformer encoder backbone that produces a unified representation of the user’s action history, which is then decoded by three task-specific prediction heads operating simultaneously. The primary head predicts the next API action from a probability distribution over all available endpoints; an auxiliary goal classification head infers the user’s underlying session objective from the observed action sequence alone; and a session boundary detection head estimates the probability that the user is about to conclude their session. During inference, only the sequence of prior API calls is required as input—the model jointly infers what the user will do next, what they are trying to accomplish, and whether they are about to leave, all from the observed behavioral trace. Leveraging a large-scale simulated behavioral dataset encompassing 2, 000 user sessions and 20, 000 API calls across 100 APIs organized into 10 functional categories, with 4 distinct session goal types governing workflow-specific transition patterns, our model demonstrates strong performance across all tasks. The primary API prediction task achieves $$79.83\%$$ top-1 accuracy and $$99.97\%$$ top-5 hit rate, representing a $$+432\%$$ improvement over a first-order Markov chain baseline. Auxiliary tasks further validate the framework’s effectiveness, with goal prediction reaching $$81.6\%$$ accuracy and session-end detection achieving $$99.3\%$$ accuracy. To ensure full reproducibility, we release an open-source Python package, , available on PyPI, that enables researchers and practitioners to regenerate the experimental dataset, reproduce all reported results, and—critically—apply the same multi-task transformer pipeline to their own user log data by mapping proprietary action sequences and session labels into the framework’s integer-encoded input format. Our approach not only advances prediction accuracy over conventional sequential methods but also establishes a new, reproducible benchmark for modeling multi-objective sequential user behavior on digital platforms, with direct applicability to any enterprise environment where user actions can be represented as ordered sequences of discrete events.

Excited-state vibronic coherences and intramolecular charge transfer dynamics of the photoinactive cyanobacteriochrome NpF2164g5

We have characterized the excited-state structural dynamics that follow optical excitation of the photoinactive cyanobacteriochrome (CBCR) NpF2164g5 to determine the first events along the photoisomerization reaction coordinate of the phycocyanobilin (PCB) chromophore in red-to-green photoactive CBCRs. Within 25 fs of photoexcitation to the first excited singlet state, S1, a cross peak begins to develop below the diagonal of the broadband two-dimensional electronic spectrum (2DES) owing to the formation of a twisted conformation of the PCB chromophore with an enhanced intramolecular charge-transfer (ICT) character. Excited-state coherent wavepacket motions, including a torsion of the methine bridge between rings C and D and the carbon–hydrogen out-of-plane (HOOP) wagging vibration, are rapidly damped as the cross peak forms. This finding supports an assignment of the torsional and HOOP modes to the reaction coordinate of a coherent nonadiabatic mechanism. The ICT process is accompanied by an ultrafast Stokes shift and rapidly damped oscillations at the torsional mode’s frequency, which are sensed by measurements of the energy gap of the cross peak using its first moment with respect to the detection frequency axis of the 2DES spectrum. These results have the further implication that polar side chains of nearby amino acid residues in the binding site of the PCB chromophore can manipulate the barrier height for photoisomerization in red-to-green photoactive CBCRs and in phytochromes by redistributing the π-electron density along the methine bridge between the C and D rings.

Survival prediction for bladder cancer using multimodal data with quantum neural networks and transformer architectures

Predicting copolymer critical parameters with a theory-integrated neural network

The phase behavior of polymer solutions is essential for designing materials with targeted properties, but classical theories such as Flory–Huggins are limited by mean-field assumptions and often fail to capture sequence-specific effects. Here, we develop a machine learning framework to predict the critical temperature and critical volume fraction of copolymer sequences, combining neural networks (NNs) with physically motivated scaling relations in a theory-integrated neural network (TI-NN). Using grand canonical Monte Carlo simulations of 3351 model copolymer sequences in solvents of varying selectivity across diverse architectures, we show that a standard NN achieves reasonable accuracy, while a TI-NN significantly reduces prediction errors and enables robust extrapolation beyond the training set. Feature analysis reveals that solvent selectivity and sequence blockiness are the dominant determinants of copolymer critical parameters. Overall, our work demonstrates that embedding theoretical insights within machine learning models enhances both accuracy and interpretability for predictions of copolymer phase behavior.

Pharmacogenetics of RAS-affecting AGT and ACE variants and the efficacy of Valsartan/HCTZ therapy

Threshold density for electron self-localization in gaseous H2

There is a revamped interest in the transport properties of quasi-free electrons in dense and cold hydrogen gas because it has recently been shown that multiple scattering effects modify their drift mobility in the same way as they do in helium and neon gases. Owing to the similarities in the electron-atom/molecule scattering cross sections, there is the possibility that also in hydrogen, electrons self-localize in bubbles as they do in dense gaseous helium and neon and in their liquids. A very scant number of experiments suggest that this may happen. In this paper, we investigate this possibility by numerically carrying out the prediction of the optimum fluctuation model. We show that electron self-localization is a very likely process and that the model accurately predicts the density at which quasi-free electrons and electron bubbles coexist in equal proportions, as inferred from the experiments.

Skin metabolomic response to medicinal shrub Myrothamnus flabellifolia and effect on skin phenotypes

Cation dominated but negatively charged Na2SO4,aq–graphene interfaces

The distribution of ions and their impact on the structure of electrolyte interfaces plays an important role in many applications. Interestingly, recent experimental studies have suggested the preferential accumulation of SO42− ions at the Na2SO4,aq–graphene interface in disagreement with the generally known tendency of cations to accumulate at graphene–electrolyte interfaces. Herein, we resolve the atomistic structure of the Na2SO4,aq–graphene interface in the 0.1–2.0M concentration range using machine learning interatomic potential-based simulations and simulated sum frequency generation (SFG) spectra to reveal the molecular origins of the conundrum. Our results show that Na+ ions accumulate between the outermost and second interfacial water layers whereas SO42− ions accumulate within the second interfacial water layer indicating cation dominated interfaces. We find that the interfacial region (within ∼10 Å of the graphene sheet) is negatively charged due to the sub-stoichiometric Na+/SO42− ratio at the interface. Our simulated SFG spectra show enhancement and a redshift of the spectra in the hydrogen bonded region as a function of Na2SO4 concentration similar to measurements due to SO42−-induced changes in the orientational order of water molecules in the second interfacial layer. Our study demonstrates that ion stratification and ion-induced water reorganization are key elements of understanding the electrolyte–graphene interface.

The effects of music and virtual reality on pain and anxiety during central venous port implantation: a randomised clinical trial

Abstract The value of music (MUS) and virtual reality (VR) in reducing pain or anxiety during central venous port implantation (CVPI) is controversial. We conducted a randomised multicenter controlled trial in 127 patients who received either MUS (38) or VR (38) during CVPI compared to standard (STAND) group (51). The primary outcome was a co-criterion related to pain or anxiety experienced during CVPI assessed using a Numerical Rating Scale. Pain and anxiety were considered independently. The secondary outcomes were the tolerance of the MUS and VR devices, patient satisfaction, and the correlation between mean pain scores and Analgesia Nociceptive Index scores (ANI). There were no differences in pain or anxiety between MUS and STAND. Mean pain was 3.3 ± 2.2 (SD) vs. 3.3 ± 2.6; ( P  > 0.9) and mean anxiety was 4.4 ± 2.8 vs. 4.2 ± 3.1; ( P  = 0.6). There were no differences in pain or anxiety between VR and STAND. Pain was 3.6 ± 2.3 vs. 3.3 ± 2.6; ( P  = 0.5) and anxiety was 3.2 ± 2.3 vs. 4.2 ± 3.1; ( P  = 0.11). The MUS and VR devices were well tolerated. Patients were very satisfied. No correlation was observed between pain scores and ANI scores in the three groups. The use of MUS or VR during CVPI had no beneficial effect on reducing pain or anxiety. Trial resgistration ClinicalTrials.gov: NCT04804735; Registred on 16/03/2021.

Surface architectural changes upon sonication revealed with sum frequency generation spectroscopy of perfluoroalkyl carboxylic acids

Understanding the interfacial behavior of molecular films is essential for elucidating key processes in both environmental and industrial systems. In this study, we investigate the interfacial behavior of perfluorotetradecanoic acid using vibrational sum frequency generation (SFG) spectroscopy to directly compare its behavior in different monolayer architectures, i.e., a traditional Langmuir monolayer and a monolayer formed from a sonicated bulk solution. Our results reveal significant differences in the vibrational SFG spectra of the carboxylic acid head group between the two monolayer types. These findings demonstrate that monolayer type can fundamentally alter the interfacial structure of perfluoroalkyl carboxylic acids and highlight the importance of monolayer architecture in interpreting interfacial spectroscopic data.

Intelligent risk assessment and early warning for human–machine–environment coupling in coal preparation plants

Publisher’s Note: “‘Ensemblization’ of density functional theory” [J. Chem. Phys. 164, 040901 (2026)]

Real-time eructation event prediction in livestock using head vibrations and machine-learning in an IoT wearable device

Topology and rigidity controlled coarsening in miktoarm star polymer melts

We study the phase-separation kinetics of miktoarm star polymer (MSP) melts using three-dimensional dissipative particle dynamics simulations. Three MSP architectures with distinct arm sequencing and connectivity are examined while systematically varying arm length and rigidity at fixed composition to quantify how architectural constraints regulate domain growth and dynamical scaling. For flexible architectures with long arms, the coarsening kinetics exhibit an early diffusive regime with L(t) ∼ t1/3, followed by saturation controlled by topological constraints. The saturated domain size grows as a power law with arm length, Ls∼lpμ with μ ranging from sublinear values (μ ≃ 0.77) in homopolymer-arm systems to nearly linear scaling (μ ≃ 1) for diblock-arm architectures. MSP melts with short arms deviates strongly from dynamical scaling and rapidly enters kinetically arrested, weakly structured states. Increasing arm rigidity suppresses mobility, leading to slower coarsening, and for fully rigid architectures, arrested morphologies comprising small, stable clusters. Architectural asymmetry further frustrates segregation and reduces the growth rate relative to symmetric designs. The results demonstrate how topology, chain length, and stiffness govern microphase-separation pathways, providing guidelines for tuning the domain size and kinetic arrest in architecturally complex polymer melts.

Deep learning for high-resolution material texture enhancement in 3D environments

Full-dimensional investigation of the photoionization spectrum of benzonitrile

The present investigation employs the state-of-the-art ab initio electronic structure and quantum dynamical method to examine the photoionization spectrum of benzonitrile (BN), also called cyanobenzene. A multi-state and multi-mode vibronic coupling model is developed and employed for the purpose. Utilizing both multi-configuration time-dependent Hartree (MCTDH) and its multi-layer (ML-MCTDH) variant, the dynamics and spectral feature of the BN radical cation (BN·+) are examined. The findings from the present theoretical work in conjunction with experimental photoionization spectroscopy aids to the understanding of complex interplay of electronic structural features and nuclear dynamics of this molecular system. This approach not only elucidates crucial features of individual photoionization bands of BN but also sheds light on the discrepancies of previous studies, as revealed by the reported spectral features. The results contribute further to the understanding of BN’s potential applications in various dimensions, highlighting its unique electronic properties influenced by the cyano group. A strong correlation was found between the full dimensional ML-MCTDH findings, the experiment, and reduced dimensional MCTDH calculations. This indicates that the ML-MCTDH method is very efficient, which enabled understanding the finer details of spectral and dynamical features.

Prognostic value of the fibrosis-4 index for predicting in-hospital mortality in sepsis patients: evidence from MIMIC-IV and eICU databases

Imaging vibrationally mediated photodissociation dynamics of MgCl

Magnesium monochloride (MgCl), a reactive intermediate in high-temperature combustion and plasma chemistry, has been theoretically recognized as a promising candidate for laser cooling. However, its key molecular parameters—such as bond dissociation energy, vibrational frequency, and excited-state topography—remain unclear due to inconsistencies between experimental and theoretical results. Here, we directly observe the photodissociation of MgCl molecules in the coupled electronically excited 42Σ+ and 52Σ+ states via developing laser-ablation molecular beam and time sliced ion velocity imaging techniques. By analyzing the vibrationally state-resolved velocity of recoiling magnesium atoms from the photolysis of MgCl at 193 nm, we determined a bond dissociation energy of 29 464(110) cm−1 and a vibrational frequency of 435(61) cm−1 for MgCl(X2Σ+). The measured dissociation energy serves as a lower bound for D0 and is notably higher than prior reported data, indicating that vibrational excitation effectively reduces the energy required for bond dissociation—a plausible reason for the underestimations observed in earlier measurements. Insights into the dissociation energy and dissociation dynamics of MgCl provide a key experimental basis for advancing theoretical models of reactive intermediates and prospective applications in laser cooling.

Integrated effects of vermicompost and town refuse on growth and nutritional status of onion cultivated in calcareous soil

Abstract This study investigates the influence of organic amendments vermicompost and town refuse on the growth parameters, nutrient uptake, and biochemical quality of onion ( Allium cepa L) grown in calcareous soils. A pot experiment was carried out at Burg El-Arab City, Alex, Egypt. By using three levels (10, 15, and 20 ton fed -1 ) of both amendments compared to a non-treated control. Results indicated that the application of vermicompost, particularly at 20 ton fed -1 significantly enhanced chlorophyll a, b and carotene levels (0.086, 0.036 and 0.131mg g -1 FW) as well as growth indicators such as plant height, bulb diameter, and total leaf area (44.58 cm, 4.268 cm and 2.894 cm 2 . Both fresh and dry weights of bulbs and leaves increased notably with organic treatments. Macronutrient (N, P, K) and micronutrient (Fe, Zn, Cu) concentrations and uptake were significantly elevated by increasing amendment levels, with vermicompost showing superior results over town refuse. Additionally, the oil content (9.83%), total soluble sugars (TSS) (15.25%), and protein percentage (28.95%) in onion ( Allium cepa L ) bulbs improved remarkably under organic treatments. Strong positive correlations were found among most growth and nutrient parameters, indicating that enhanced nutrient availability translates directly to improve physiological and yield responses. The findings suggest that vermicompost is an effective and sustainable amendment for improving onion ( Allium cepa L ) productivity and nutritional quality under nutrient-poor calcareous soil conditions.

Fast calculation of plasmonic metals nanofilms permittivity in THz–ultraviolet range

A theoretical study of the dielectric permittivity tensor for the plasmonic metals (Au, Ag, and Cu) nanofilms of different thicknesses in the THz–ultraviolet frequency range was conducted. Using DFT calculation data, the models for calculating longitudinal and transverse to nanofilm surface components ɛ‖(ω, h) and ɛ⊥(ω, h) for the metal nanofilms of any thickness h are proposed. The model for the ɛ‖(ω, h) calculation uses a sum of the interband contribution of the bulk metal obtained from density functional perturbation theory, the Drude contribution of intraband excitations, and the Lorentz model to account for the film surface influence. The model for the ɛ⊥(ω, h) calculation uses, in addition to the interband term, the contribution of the electron’s motion perpendicular to the film surface, described as its motion in a 1D infinitely deep potential well. This contribution is calculated using transient perturbation theory and Fermi’s golden rule. Based on both of these models, a Python program has been written that allows one to calculate in seconds longitudinal and transverse permittivity components in the THz–ultraviolet range for nanofilms of these metals with surface indices (001), (110), or (111) and any film thickness h. Using Bruggeman’s effective medium theory, the effective permittivities ɛeff(ω, h) of several thin gold films were calculated and shown to be in good agreement with the available experimental data.