Gist (Czechia)

Data-free Universal Adversarial Perturbation (UAP) is an image-agnostic adversarial attack that deceives deep neural networks using a single perturbation generated solely from random noise without relying on data priors. However, traditional data-free UAP methods often suffer from limited transferability due to the absence of semantic content in random noise. To address this issue, we propose a novel data-free universal attack method that recursively extracts pseudo-semantic priors directly from the UAPs during training to enrich the semantic content within the data-free UAP framework. Our approach effectively leverages latent semantic information within UAPs via region sampling, enabling successful input transformations—typically ineffective in traditional data-free UAP methods due to the lack of semantic cues—and significantly enhancing black-box transferability. Furthermore, we introduce a sample reweighting technique to mitigate potential imbalances from random sampling and transformations, emphasizing hard examples less affected by the UAPs. Comprehensive experiments on ImageNet show that our method achieves state-of-the-art performance in average fooling rate by a substantial margin, notably improves attack transferability across various CNN architectures compared to existing data-free UAP methods, and even surpasses data-dependent UAP methods. Code is available at: https://github.com/ChnanChan/PSP-UAP.

Size- and shape-tailored copper (Cu) nanocrystals can offer vicinal planes for facile carbon dioxide (CO2) activation. Despite extensive reactivity benchmarks, a correlation between CO2 conversion and morphology structure has not yet been established at vicinal Cu interfaces. Herein, ambient pressure scanning tunneling microscopy reveals step-broken Cu nanocluster evolutions on the Cu(997) surface under 1 mbar CO2(g). The CO2 dissociation reaction produces carbon monoxide (CO) adsorbate and atomic oxygen (O) at Cu step-edges, inducing complicated restructuring of the Cu atoms to compensate for increased surface chemical potential energy at ambient pressure. The CO molecules bound at under-coordinated Cu atoms contribute to the reversible Cu clustering with the pressure gap effect, whereas the dissociated oxygen leads to irreversible Cu faceting geometries. Synchrotron-based ambient pressure X-ray photoelectron spectroscopy identifies the chemical binding energy changes in CO-Cu complexes, which proves the characterized real-space evidence for the step-broken Cu nanoclusters under CO(g) environments. Our in situ surface observations provide a more realistic insight into Cu nanocatalyst designs for efficient CO2 conversion to renewable energy sources during C1 chemical reactions.

The electrochemical synthesis of ammonia under ambient condition has attracted significant attention as an alternative route for green ammonia production. However, competing hydrogen evolution reaction (HER) hinders selective ammonia synthesis through electrochemical nitrogen reduction reaction (NRR). For the broad adoption of such technology in industries, it is essential to overcome poor selectivity of NRR by developing effective electrocatalysts and controlling the reactant at the interface between electrode and electrolyte.[1,2] In this study, we prepared cobalt-molybdenum carbide-carbon nanofiber via electrospinning method as a selective NRR catalyst. Furthermore, we reveal the relationship between water adsorption ability and NRR selectivity on biphasic catalyst using isotopic-labeling and in-situ analyses. Reference [1] Journal of Electroanalytical Chemistry 2021, 896, 115157.[2] Journal of Energy Chemistry 2022, 67, 474–482.

Catalyst double layer (CDL) structure on gas diffusion electrode (GDE) without microporous layer (MPL) is developed for the anode of high temperature polymer electrolyte membrane fuel cell (HT–PEMFC). Polytetrafluoroethylene and polyvinylidene fluoride is applied for the outer and inner layer as the hydrophobic binder for fabricating 3D electrode structure. The effect of Pt ratio in anode CDLs on the GDE performance was investigated by single cell test and electrochemical analysis using poly(benzimidazole) (PBI)–based membrane, with Pt loading under 0.3 mg/cm2 at ambient pressure air on 150℃. The results show that optimal CDL anodes present performance of 0.65V at 0.2A/cm2 and peak power density is 0.41W/cm2 at mass transfer region which is higher than conventional single layer anode with MPL. Furthermore, the 3D structure of the outer catalyst layer in anode read high catalyst utilization and preventing leakage of electrolyte with a short–term durability test shows good stability with PBI membrane.

Manganese oxides are ubiquitous marine minerals which are redox sensitive. As major components of manganese nodules found on the ocean floor, birnessite and buserite have been known to be two distinct water-containing minerals with manganese octahedral interlayer separations of ~7 Å and ~10 Å, respectively. We show here that buserite is a super-hydrated birnessite formed near 5 km depth conditions. As one of the most hydrous minerals containing ca. 34.5 wt. % water, super-hydrated birnessite, i.e., buserite, remains stable up to ca. 70 km depth conditions, where it transforms into manganite by releasing ca. 24.3 wt. % water. Subsequent transformations to hausmannite and pyrochroite occur near 100 km and 120 km depths, respectively, concomitant with a progressive reduction of Mn4+ to Mn2+. Our work forwards an abiotic geochemical cycle of manganese minerals in subduction and/or other aqueous terrestrial environments, with implications for water storage and cycling, and the redox capacity of the region.

DNA polymerase plays a critical role in passing the genetic information of any living organism to its offspring. DNA polymerase from enterobacteria phage RB69 (RB69pol) has both polymerization and exonuclease activities and has been extensively studied as a model system for B-family DNA polymerases. Many binary and ternary complex structures of RB69pol are known, and they all contain a single polymerase-primer/template (P/T) DNA complex. Here, we report a crystal structure of the exonuclease-deficient RB69pol with the P/T duplex in a dimeric form at a resolution of 2.2 Å. The structure includes one new closed ternary complex with a single divalent metal ion bound and one new open binary complex in the pre-insertion state with a vacant dNTP-binding pocket. These complexes suggest that initial binding of the correct dNTP in the open state is much weaker than expected and that initial binding of the second divalent metal ion in the closed state is also much weaker than measured. Additional conformational changes are required to convert these complexes to high-affinity states. Thus, the measured affinities for the correct incoming dNTP and divalent metal ions are average values from many conformationally distinctive states. Our structure provides new insights into the order of the complex assembly involving two divalent metal ions. The biological relevance of specific interactions observed between one RB69pol and the P/T duplex bound to the second RB69pol observed within this dimeric complex is discussed.

Continual learning is a realistic learning scenario for AI models. Prevalent scenario of continual learning, however, assumes disjoint sets of classes as tasks and is less realistic rather artificial. Instead, we focus on ‘blurry’ task boundary; where tasks shares classes and is more realistic and practical. To address such task, we argue the importance of diversity of samples in an episodic memory. To enhance the sample diversity in the memory, we propose a novel memory management strategy based on per-sample classification uncertainty and data augmentation, named Rainbow Memory (RM). With extensive empirical validations on MNIST, CIFAR10, CIFAR100, and ImageNet datasets, we show that the proposed method significantly improves the accuracy in blurry continual learning setups, outperforming state of the arts by large margins despite its simplicity. Code and data splits will be available in https://github.com/clovaai/rainbow-memory.

There is growing interest in alkaline water electrolysis as a sustainable approach for producing hydrogen, but developing efficient and inexpensive catalysts for the oxygen evolution reaction, which can limit the operational efficiency of water electrolysis due to its considerable overpotential, is regarded as the most overriding challenge. Therefore, significant progress has been made in developing catalysts with transition metal and carbon materials as alternative catalysts. Here, we prepared cobalt containing carbon nanofibers via a facile route of electrospinning and pyrolysis, and metal leached carbon nanofibers were also prepared by subsequently leaching the metal. Despite metal leaching, the latter ones still show comparable activity and stability with iridium black in alkaline water electrolysis. After detailed physicochemical and electrochemical characterizations, we revealed that graphitic edge plane rich carbon is mainly responsible for the activity of our material rather than embedded metal species. In addition, the metal plays a role in forming the specific carbon structure along with improving graphitization based on the catalytic graphitization. This result indicates the importance of the graphitic edge plane and might be helpful to understand carbon anodes for alkaline water electrolysis.

The present study was performed to compare the morpho-physiological characteristics of the tetraploid and diploid varieties of Platycodon grandiflorum and to obtain basic data for cultivating a tetraploid variety with high yield and content of functional substances. The plant height of the tetraploid variety (54.0 cm) was slightly higher than that of the diploid variety. The leaf length and width of the tetraploid variety were 10.2 cm and 7.3 cm, respectively. The results obtained from the present study revealed that the form of the leaf changed from lanceolate to ovate, and the chlorophyll content in the tetraploid variety (16.7) was slightly higher than that in the diploid variety. The photosynthetic rate significantly increased (24%) to <TEX>$13.4{\mu}mol\;CO_2{\cdot}m^{-2}{\cdot}s^{-1}$</TEX> in the tetraploid variety from that of the diploid variety. The pollen viability of the tetraploid variety was decreased by approximately 33% with respect to that of the diploid variety, but this did not have a significant adverse effect on seed production. The fresh weight of tetraploid P. grandiflorum was 49.4 g, which was approximately 44% higher than that of the diploid variety.