Search

Modern natural language processing (NLP) state-of-the-art (SoTA) deep learning (DL) models have hundreds of millions of parameters, making them extremely complex. Large datasets are required for training these models, and while pretraining has reduced this requirement, human-labelled datasets are still necessary for fine-tuning. Few-shot learning (FSL) techniques, such as meta-learning, try to train models from smaller datasets to mitigate this cost. However, the tasks used to evaluate these meta-learners frequently diverge from the problems in the real world that they are meant to resolve. This work aims to apply meta-learning to a problem that is more pertinent to the real world: class incremental learning (IL). In this scenario, after completing its training, the model learns to classify newly introduced classes. One unique quality of meta-learners is that they can generalise from a small sample size to classes that have never been seen before, which makes them especially useful for class incremental learning (IL). The method describes how to emulate class IL using proxy new classes. This method allows a meta-learner to complete the task without the need for retraining. To generate predictions, the transformer-based aggregation function in a meta-learner that modifies data from examples across all classes has been proposed. The principal contributions of the model include concurrently considering the entire support and query sets, and prioritising attention to crucial samples, such as the question, to increase the significance of its impact during inference. The outcomes demonstrate that the model surpasses prevailing benchmarks in the industry. Notably, most meta-learners demonstrate significant generalisation in the context of class IL even without specific training for this task. This paper establishes a high-performing baseline for subsequent transformer-based aggregation techniques, thereby emphasising the practical significance of meta-learners in class IL., (© 2024. The Author(s).)

Identifying the microscopic nature of non-equilibrium energy transfer mechanisms among electronic, spin, and lattice degrees of freedom is central to understanding ultrafast phenomena such as manipulating magnetism on the femtosecond timescale. Here, we use time- and angle-resolved photoemission spectroscopy to go beyond the often-used ensemble-averaged view of non-equilibrium dynamics in terms of quasiparticle temperature evolutions. We show for ferromagnetic Ni that the non-equilibrium electron and spin dynamics display pronounced variations with electron momentum, whereas the magnetic exchange interaction remains isotropic. This highlights the influence of lattice-mediated scattering processes and opens a pathway toward unraveling the still elusive microscopic mechanism of spin-lattice angular momentum transfer.

Wide-band-gap insulators such as NiO offer the exciting prospect of coherently manipulating electronic correlations with strong optical fields. Contrary to metals where rapid dephasing of optical excitation via electronic processes occurs, the sub-gap excitation in charge-transfer insulators has been shown to couple to low-energy bosonic excitations. However, it is currently unknown if the bosonic dressing field is composed of phonons or magnons. Here we use the prototypical charge-transfer insulator NiO to demonstrate that 1.5 eV sub-gap optical excitation leads to a renormalised NiO band-gap in combination with a significant reduction of the antiferromagnetic order. We employ element-specific X-ray reflectivity at the FLASH free-electron laser to demonstrate the reduction of the upper band-edge at the O 1s-2p core-valence resonance (K-edge) whereas the antiferromagnetic order is probed via X-ray magnetic linear dichroism (XMLD) at the Ni 2p-3d resonance (L 2 -edge). Comparing the transient XMLD spectral line shape to ground-state measurements allows us to extract a spin temperature rise of 65 ± 5 K for time delays longer than 400 fs while at earlier times a non-equilibrium spin state is formed. We identify transient mid-gap states being formed during the first 200 fs accompanied by a band-gap reduction lasting at least up to the maximum measured time delay of 2.4 ps. Electronic structure calculations indicate that magnon excitations significantly contribute to the reduction of the NiO band gap.

The performance of time-resolved photoemission experiments at fs-pulsed photon sources is ultimately limited by the e-e Coulomb interaction, downgrading energy and momentum resolution. Here, we present an approach to effectively suppress space-charge artifacts in momentum microscopes and photoemission microscopes. A retarding electrostatic field generated by a special objective lens repels slow electrons, retaining the k-image of the fast photoelectrons. The suppression of space-charge effects scales with the ratio of the photoelectron velocities of fast and slow electrons. Fields in the range from -20 to -1100 V/mm for E kin = 100 eV to 4 keV direct secondaries and pump-induced slow electrons back to the sample surface. Ray tracing simulations reveal that this happens within the first 40 to 3 μm above the sample surface for E kin = 100 eV to 4 keV. An optimized front-lens design allows switching between the conventional accelerating and the new retarding mode. Time-resolved experiments at E kin = 107 eV using fs extreme ultraviolet probe pulses from the free-electron laser FLASH reveal that the width of the Fermi edge increases by just 30 meV at an incident pump fluence of 22 mJ/cm 2 (retarding field -21 V/mm). For an accelerating field of +2 kV/mm and a pump fluence of only 5 mJ/cm 2 , it increases by 0.5 eV (pump wavelength 1030 nm). At the given conditions, the suppression mode permits increasing the slow-electron yield by three to four orders of magnitude. The feasibility of the method at high energies is demonstrated without a pump beam at E kin = 3830 eV using hard x rays from the storage ring PETRA III. The approach opens up a previously inaccessible regime of pump fluences for photoemission experiments.

A joint theoretical and experimental investigation is performed to understand the underlying physics of laser-induced demagnetization in Ni and Co films with varying thicknesses excited by 10 fs optical pulses. Experimentally, the dynamics of spins is studied by determining the time-dependent amplitude of the Voigt vector, retrieved from a full set of magnetic and nonmagnetic quantities performed on both sides of films, with absolute time reference. Theoretically, ab initio calculations are performed using time-dependent density functional theory. Overall, we demonstrate that spin-orbit induced spin flips are the most significant contributors with superdiffusive spin transport, which assumes only that the transport of majority spins without spin flips induced by scattering does not apply in Ni. In Co it plays a significant role during the first ∼20  fs only. Our study highlights the material dependent nature of the demagnetization during the process of thermalization of nonequilibrium spins.

Background and Aims: HCV GT-3 has a more pronounced effect on hepatic steatosis and host lipids than other HCV genotypes and is proving less responsive to all oral interferon-free treatment with direct acting antiviral agents. As both HCV GT3 infection and NASH can result in steatosis and cirrhosis, we asked whether hepatic transcriptional profiles reflective of the host response to inflammation differed based on the etiology of injury., Methods: Hepatic gene expression was determined for 48 pre-selected genes known to be associated with hepatic interferon signaling and lipid metabolic pathways in treatment-naïve HCV GT-3 (n = 9) and NASH (n = 14) patients., Results: Genes with significantly higher expression in HCV included chemokines CXCL10, CXCL11 interferon IFNA2, interferon receptors IFNAR1, IL10RB negative regulators of interferon signaling SOCS3, USP18, JAK/STAT and IRF family members STAT1, STAT2, and IRF, and TGFB family members TGFB1, TGFBR1, and TGFBR2 and other ISGs like OAS2, IF127, IF144 and ISG15. HCV infection was also associated with higher expression of genes associated with lipid metabolism APOE, APOL3, SREBF1 and HMBS. Furthermore, our results suggest that, in HCV GT3-infected patients, IL28B (CC) genotype is associated with lower baseline ISG expression such as IRF9, ISG15, MX1, STAT1, CXCL10, CXCL11, and IFI27 compared to CT/TT genotype., Conclusions: HCV GT-3 and NASH both induce hepatic steatosis and inflammation, while HCV GT-3 infection is uniquely associated with elevated transcription of hepatic ISGs and genes associated with lipid metabolism. These changes likely reflect the unique host response to HCV replication distinct from the inflammatory response induced by NASH.