Inspector for Face Forgery Detection: Defending Against Adversarial Attacks From Coarse to Fine.

The emergence of face forgery has raised global concerns on social security, thereby facilitating the research on automatic forgery detection. Although current forgery detectors have demonstrated promising performance in determining authenticity, their susceptibility to adversarial perturbations remains insufficiently addressed. Given the nuanced discrepancies between real and fake instances are essential in forgery detection, previous defensive paradigms based on input processing and adversarial training tend to disrupt these discrepancies. For the detectors, the learning difficulty is thus increased, and the natural accuracy is dramatically decreased. To achieve adversarial defense without changing the instances as well as the detectors, a novel defensive paradigm called Inspector is designed specifically for face forgery detectors. Specifically, Inspector defends against adversarial attacks in a coarse-to-fine manner. In the coarse defense stage, adversarial instances with evident perturbations are directly identified and filtered out. Subsequently, in the fine defense stage, the threats from adversarial instances with imperceptible perturbations are further detected and eliminated. Experimental results across different types of face forgery datasets and detectors demonstrate that our method achieves state-of-the-art performances against various types of adversarial perturbations while better preserving natural accuracy. Code is available on https://github.com/xarryon/Inspector.

Efficacy and Safety of Early Treatment with Glibenclamide in Patients with Aneurysmal Subarachnoid Hemorrhage: A Randomized Controlled Trial.

BACKGROUND: This study aims to investigate the efficacy and safety of glibenclamide treatment in patients with acute aneurysmal subarachnoid hemorrhage (aSAH). METHODS: The randomized controlled trial was conducted from October 2021 to May 2023 at two university-affiliated hospitals in Beijing, China. The study included patients with aSAH within 48 h of onset, of whom were divided into the intervention group and the control group according to the random number table method. Patients in the intervention group received glibenclamide tablet 3.75 mg/day for 7 days. The primary end points were the levels of serum neuron-specific enolase (NSE) and soluble protein 100B (S100B) between the two groups. Secondary end points included evaluating changes in the midline shift and the gray matter-white matter ratio, as well as assessing the modified Rankin Scale scores during follow-up. The trial was registered at ClinicalTrials.gov (identifier NCT05137678). RESULTS: A total of 111 study participants completed the study. The median age was 55 years, and 52% were women. The mean admission Glasgow Coma Scale was 10, and 58% of the Hunt-Hess grades were no less than grade III. The baseline characteristics of the two groups were similar. On days 3 and 7, there were no statistically significant differences observed in serum NSE and S100B levels between the two groups (P > 0.05). The computer tomography (CT) values of gray matter and white matter in the basal ganglia were low on admission, indicating early brain edema. However, there were no significant differences found in midline shift and gray matter-white matter ratio (P > 0.05) between the two groups. More than half of the patients had a beneficial outcome (modified Rankin Scale scores 0-2), and there were no statistically significant differences between the two groups. The incidence of hypoglycemia in the two groups were 4% and 9%, respectively (P = 0.439). CONCLUSIONS: Treating patients with early aSAH with oral glibenclamide did not decrease levels of serum NSE and S100B and did not improve the poor 90-day neurological outcome. In the intervention group, there was a visible decreasing trend in cases of delayed cerebral ischemia, but no statistically significant difference was observed. The incidence of hypoglycemia did not differ significantly between the two groups.

Protecting Prostate Cancer Classification From Rectal Artifacts via Targeted Adversarial Training.

Magnetic resonance imaging (MRI)-based deep neural networks (DNN) have been widely developed to perform prostate cancer (PCa) classification. However, in real-world clinical situations, prostate MRIs can be easily impacted by rectal artifacts, which have been found to lead to incorrect PCa classification. Existing DNN-based methods typically do not consider the interference of rectal artifacts on PCa classification, and do not design specific strategy to address this problem. In this study, we proposed a novel Targeted adversarial training with Proprietary Adversarial Samples (TPAS) strategy to defend the PCa classification model against the influence of rectal artifacts. Specifically, based on clinical prior knowledge, we generated proprietary adversarial samples with rectal artifact-pattern adversarial noise, which can severely mislead PCa classification models optimized by the ordinary training strategy. We then jointly exploited the generated proprietary adversarial samples and original samples to train the models. To demonstrate the effectiveness of our strategy, we conducted analytical experiments on multiple PCa classification models. Compared with ordinary training strategy, TPAS can effectively improve the single- and multi-parametric PCa classification at patient, slice and lesion level, and bring substantial gains to recent advanced models. In conclusion, TPAS strategy can be identified as a valuable way to mitigate the influence of rectal artifacts on deep learning models for PCa classification.

Class-Incremental Learning: A Survey.

Deep models, e.g., CNNs and Vision Transformers, have achieved impressive achievements in many vision tasks in the closed world. However, novel classes emerge from time to time in our ever-changing world, requiring a learning system to acquire new knowledge continually. Class-Incremental Learning (CIL) enables the learner to incorporate the knowledge of new classes incrementally and build a universal classifier among all seen classes. Correspondingly, when directly training the model with new class instances, a fatal problem occurs - the model tends to catastrophically forget the characteristics of former ones, and its performance drastically degrades. There have been numerous efforts to tackle catastrophic forgetting in the machine learning community. In this paper, we survey comprehensively recent advances in class-incremental learning and summarize these methods from several aspects. We also provide a rigorous and unified evaluation of 17 methods in benchmark image classification tasks to find out the characteristics of different algorithms empirically. Furthermore, we notice that the current comparison protocol ignores the influence of memory budget in model storage, which may result in unfair comparison and biased results. Hence, we advocate fair comparison by aligning the memory budget in evaluation, as well as several memory-agnostic performance measures. The source code is available at https://github.com/zhoudw-zdw/CIL_Survey/.

Kidney transplantation from a systemic lupus erythematosus donor and 1-year follow-up: A case report.

Systemic lupus erythematosus (SLE) is usually regarded as a relative contraindication for deceased kidney donation. The pathological variations because of the changes in the immune environment after kidney transplantation (KT) are unclear, and the recovery of renal function is poorly understood. We present a case of KT from a deceased donor with SLE who was followed-up for one year. Although SLE-related hemangioma developed during the perioperative period, it was cured after interventional treatment. A pre-planned biopsy was performed one year after KT, and it was found that most of the pathological changes and immunofluorescent markers of lupus had resolved. Renal function was stable, and urinary protein and occult blood levels reduced one year after KT.

Predictive factors of postoperative complications related to free flap reconstruction in head and neck cancer patients admitted to intensive care unit: a retrospective cohort study.

BACKGROUND: The epidemiology and risk factors for postoperative complications related to free flap reconstruction in head and neck cancer patients admitted to the Intensive Care Unit (ICU) are unknown. METHODS: We performed a retrospective cohort study of patients with free flap reconstruction of head and neck cancer between September 2015 and April 2023 admitted to the ICU of Beijing Tongren Hospital. The univariate and multivariate analyses were used to explore the risk factors for postoperative complications related to free flap reconstruction admitted to ICU, including flap necrosis, bleeding, fistula, and infection. RESULTS: A total of 239 patients were included in this study, and 38 (15.9%) patients had postoperative complications related to free flap reconstruction. The median length of ICU stay was 1 day (interquartile range, 1-2 days). Multivariate analysis found that low BMI (P < 0.001), high postoperative CRP (P = 0.005), low hemoglobin (P = 0.012), and inadequate fluid intake (P < 0.05) were independent risk factors for complications. CONCLUSIONS: Postoperative complications related to free flap reconstruction were common in this ICU population. Careful fluid management and monitoring of CRP and hemoglobin levels may reduce complications.

KDM5A/B contribute to HIV-1 latent infection and survival of HIV-1 infected cells.

Combinational antiretroviral therapy (cART) suppresses human immunodeficiency virus type 1 (HIV-1) viral replication and pathogenesis in acquired immunodeficiency syndrome (AIDS) patients. However, HIV-1 remains in the latent stage of infection by suppressing viral transcription, which hinders an HIV-1 cure. One approach for an HIV-1 cure is the "shock and kill" strategy. The strategy focuses on reactivating latent HIV-1, inducing the viral cytopathic effect and facilitating the immune clearance for the elimination of latent HIV-1 reservoirs. Here, we reported that the H3K4 trimethylation (H3K4me3)-specific demethylase KDM5A/B play a role in suppressing HIV-1 Tat/LTR-mediated viral transcription in HIV-1 latent cells. Furthermore, we evaluated the potential of KDM5-specific inhibitor JQKD82 as an HIV-1 "shock and kill" agent. Our results showed that JQKD82 increases the H3K4me3 level at HIV-1 5' LTR promoter regions, HIV-1 reactivation, and the cytopathic effects in an HIV-1-latent T cell model. In addition, we identified that the combination of JQKD82 and AZD5582, a non-canonical NF-κB activator, generates a synergistic impact on inducing HIV-1 lytic reactivation and cell death in the T cell. The latency-reversing potency of the JQKD82 and AZD5582 pair was also confirmed in peripheral blood mononuclear cells (PBMCs) isolated from HIV-1 aviremic patients and in an HIV-1 latent monocyte. In latently infected microglia (HC69) of the brain, either deletion or inhibition of KDM5A/B results in a reversal of the HIV-1 latency. Overall, we concluded that KDM5A/B function as a host repressor of the HIV-1 lytic reactivation and thus promote the latency and the survival of HIV-1 infected reservoirs.

The bioproduction of astaxanthin: A comprehensive review on the microbial synthesis and downstream extraction.

Astaxanthin is a valuable orange-red carotenoid with wide applications in agriculture, food, cosmetics, pharmaceuticals and nutraceuticals areas. At present, the biological synthesis of astaxanthin mainly relies on Haematococcus pluvialis and Xanthophyllomyces dendrorhous. With the rapid development of synthetic biology, more recombinant microbial hosts have been genetically constructed for astaxanthin production including Escherichia coli, Saccharomyces cerevisiae and Yarrowia lipolytica. As multiple genes (15) were involved in the astaxanthin synthesis, it is particularly important to adopt different strategies to balance the metabolic flow towards the astaxanthin synthesis. Furthermore, astaxanthin is a fat-soluble compound stored intracellularly, hence efficient extraction methods are also essential for the economical production of astaxanthin. Several efficient and green extraction methods of astaxanthin have been reported in recent years, including the superfluid extraction, ionic liquid extraction and microwave-assisted extraction. Accordingly, this review will comprehensively introduce the advances on the astaxanthin production and extraction by using different microbial hosts and strategies to improve the astaxanthin synthesis and extraction efficiency.

Enhancing C2+ product selectivity in CO2 electroreduction by enriching intermediates over carbon-based nanoreactors.

Electrochemical CO2 reduction reaction (CO2RR) to multicarbon (C2+) products faces challenges of unsatisfactory selectivity and stability. Guided by finite element method (FEM) simulation, a nanoreactor with cavity structure can facilitate C-C coupling by enriching *CO intermediates, thus enhancing the selectivity of C2+ products. We designed a stable carbon-based nanoreactor with cavity structure and Cu active sites. The unique geometric structure endows the carbon-based nanoreactor with a remarkable C2+ product faradaic efficiency (80.5%) and C2+-to-C1 selectivity (8.1) during the CO2 electroreduction. Furthermore, it shows that the carbon shell could efficiently stabilize and highly disperse the Cu active sites for above 20 hours of testing. A remarkable C2+ partial current density of-323 mA cm-2 was also achieved in a flow cell device. In situ Raman spectra and density functional theory (DFT) calculation studies validated that the *COatop intermediates are concentrated in the nanoreactor, which reduces the free energy of C-C coupling. This work unveiled a simple catalyst design strategy that would be applied to improve C2+ product selectivity and stability by rationalizing the geometric structures and components of catalysts.

Single-cell analysis of CD4+ tissue residency memory cells (TRMs) in adult atopic dermatitis: A new potential mechanism.

The pathophysiology of atopic dermatitis (AD) is complex. CD4+ T cells play an essential role in the development of lesions in AD. However, the underlying mechanism remains unclear. In the present study, we investigated the differentially expressed genes (DEGs) between adult AD lesioned and non-lesioned skin using two datasets from the Gene Expression Omnibus (GEO) database. 62 DEGs were shown to be related to cytokine response. Compared to non-lesioned skin, lesioned skin showed immune infiltration with increased numbers of activated natural killer (NK) cells and CD4+ T memory cells (p < 0.01). We then identified 13 hub genes with a strong association with CD4+ T cells using weighted correlation network analysis. Single-cell analysis of AD detected a novel CD4+ T subcluster, CD4+ tissue residency memory cells (TRMs), which were verified through immunohistochemistry (IHC) to be increased in the dermal area of AD. The significant relationship between CD4+ TRM and AD was assessed through further analyses. FOXO1 and SBNO2, two of the 13 hub genes, were characteristically expressed in the CD4+ TRM, but down-regulated in IFN-γ/TNF-α-induced HaCaT cells, as shown using quantitative polymerase chain reaction (qPCR). Moreover, SBNO2 expression was associated with increased Th1 infiltration in AD (p < 0.05). In addition, genes filtered using Mendelian randomization were positively correlated with CD4+ TRM and were highly expressed in IFN-γ/TNF-α-induced HaCaT cells, as determined using qPCR and western blotting. Collectively, our results revealed that the newly identified CD4+ TRM may be involved in the pathogenesis of adult AD.

The early change in pH values after out-of-hospital cardiac arrest is not associated with neurological outcome at hospital discharge.

Background: The association between pH values and outcome for patients after out-of-hospital cardiac arrest (OHCA) was not fully elucidated; besides, the relationship of change in pH values and neurological outcome was unknown. The aim was to explore the association of pH values as well as change in pH values and neurological outcome for OHCA cardiac patients. Methods: The adult patients with non-traumatic out-of-hospital cardiac arrest, shock-refractory ventricular fibrillation or pulseless ventricular tachycardia, and at least two arterial blood gases analysis recorded after admission were included. The change in pH values is calculated as the difference between the second and first pH value, and divided by time interval got the rate of change in pH values. The primary outcome was modified Rankin Score (mRS), dichotomized to good (mRS 0-3) and poor (mRS 4-6) outcomes at hospital discharge. The independent relationship of the first pH value, second pH value, and changes in pH values with neurological outcome was investigated with multivariable logistic regression models, respectively. Results: A total of 1388 adult patients were included for analysis, of which 514 (37%) had good neurological outcome. The median first pH value and second pH value after admission were 7.21 (interquartile range [IQR] 7.09-7.29) and 7.28 (IQR 7.20-7.36), respectively. The median absolute, relative change, and rate of changes in pH values were 0.08 (IQR 0.01-0.16), 1.10% (IQR 0.11-2.22%), and 0.02 (IQR 0-0.06) per hour, respectively. After adjusting for confounders, the higher first pH value (odds ratio [OR] 3.81, confidence interval [CI] 1.60-9.24, P = 0.003) and higher second pH value (OR 9.54, CI 3.45-26.87, P < 0.001) after admission were associated with good neurological outcome, respectively. The absolute (OR 1.58, CI 0.58-4.30, P = 0.368) and relative (OR 1.03, CI 0.96-1.11, P = 0.399) change as well as the rate of change (OR 0.98, CI 0.33-2.71, P = 974) in pH values were not associated with neurological outcome. Conclusions: For OHCA patients, abnormality in pH values was very common, with a more acidic pH value indicating poor neurological outcome. However, the change in pH values was not associated with outcomes.

Ultrasmall Fe3O4 nanoparticles self-assembly induced dual-mode T1/T2-weighted magnetic resonance imaging and enhanced tumor synergetic theranostics.

Individual theranostic agents with dual-mode MRI responses and therapeutic efficacy have attracted extensive interest due to the real-time monitor and high effective treatment, which endow the providential treatment and avoid the repeated medication with side effects. However, it is difficult to achieve the integrated strategy of MRI and therapeutic drug due to complicated synthesis route, low efficiency and potential biosafety issues. In this study, novel self-assembled ultrasmall Fe3O4 nanoclusters were developed for tumor-targeted dual-mode T1/T2-weighted magnetic resonance imaging (MRI) guided synergetic chemodynamic therapy (CDT) and chemotherapy. The self-assembled ultrasmall Fe3O4 nanoclusters synthesized by facilely modifying ultrasmall Fe3O4 nanoparticles with 2,3-dimercaptosuccinic acid (DMSA) molecule possess long-term stability and mass production ability. The proposed ultrasmall Fe3O4 nanoclusters shows excellent dual-mode T1 and T2 MRI capacities as well as favorable CDT ability due to the appropriate size effect and the abundant Fe ion on the surface of ultrasmall Fe3O4 nanoclusters. After conjugation with the tumor targeting ligand Arg-Gly-Asp (RGD) and chemotherapy drug doxorubicin (Dox), the functionalized Fe3O4 nanoclusters achieve enhanced tumor accumulation and retention effects and synergetic CDT and chemotherapy function, which serve as a powerful integrated theranostic platform for cancer treatment.

Growth of millimeter-sized 2D metal iodide crystals induced by ion-specific preference at water-air interfaces.

Conventional liquid-phase methods lack precise control in synthesizing and processing materials with macroscopic sizes and atomic thicknesses. Water interfaces are ubiquitous and unique in catalyzing many chemical reactions. However, investigations on two-dimensional (2D) materials related to water interfaces remain limited. Here we report the growth of millimeter-sized 2D PbI2 single crystals at the water-air interface. The growth mechanism is based on an inherent ion-specific preference, i.e. iodine and lead ions tend to remain at the water-air interface and in bulk water, respectively. The spontaneous accumulation and in-plane arrangement within the 2D crystal of iodide ions at the water-air interface leads to the unique crystallization of PbI2 as well as other metal iodides. In particular, PbI2 crystals can be customized to specific thicknesses and further transformed into millimeter-sized mono- to few-layer perovskites. Additionally, we have developed water-based techniques, including water-soaking, spin-coating, water-etching, and water-flow-assisted transfer to recycle, thin, pattern, and position PbI2, and subsequently, perovskites. Our water-interface mediated synthesis and processing methods represents a significant advancement in achieving simple, cost-effective, and energy-efficient production of functional materials and their integrated devices.

Development and Validation of a Deep Learning Model to Reduce the Interference of Rectal Artifacts in MRI-based Prostate Cancer Diagnosis.

Purpose To develop an MRI-based model for clinically significant prostate cancer (csPCa) diagnosis that can resist rectal artifact interference. Materials and Methods This retrospective study included 2203 male patients with prostate lesions who underwent biparametric MRI and biopsy between January 2019 and June 2023. Targeted adversarial training with proprietary adversarial samples (TPAS) strategy was proposed to enhance model resistance against rectal artifacts. The automated csPCa diagnostic models trained with and without TPAS were compared using multicenter validation datasets. The impact of rectal artifacts on the diagnostic performance of each model at the patient and lesion levels was compared using the area under the receiver operating characteristic curve (AUC) and the area under the precision-recall curve (AUPRC). The AUC between models was compared using the DeLong test, and the AUPRC was compared using the bootstrap method. Results The TPAS model exhibited diagnostic performance improvements of 6% at the patient level (AUC: 0.87 vs 0.81, P < .001) and 7% at the lesion level (AUPRC: 0.84 vs 0.77, P = .007) compared with the control model. The TPAS model demonstrated less performance decline in the presence of rectal artifact-pattern adversarial noise than the control model (ΔAUC: -17% vs -19%, ΔAUPRC: -18% vs -21%). The TPAS model performed better than the control model in patients with moderate (AUC: 0.79 vs 0.73, AUPRC: 0.68 vs 0.61) and severe (AUC: 0.75 vs 0.57, AUPRC: 0.69 vs 0.59) artifacts. Conclusion This study demonstrates that the TPAS model can reduce rectal artifact interference in MRI-based csPCa diagnosis, thereby improving its performance in clinical applications. Keywords: MR-Diffusion-weighted Imaging, Urinary, Prostate, Comparative Studies, Diagnosis, Transfer Learning Clinical trial registration no. ChiCTR23000069832 Supplemental material is available for this article. Published under a CC BY 4.0 license.

Vapor Phase Growth of Air-Stable Hybrid Perovskite FAPbBr3 Single-Crystalline Nanosheets.

Organic-inorganic hybrid perovskites have attracted tremendous attention owing to their fascinating optoelectronic properties. However, their poor air stability seriously hinders practical applications, which becomes more serious with thickness down to the nanoscale. Here we report a one-step vapor phase growth of HC(NH2)2PbBr3 (FAPbBr3) single-crystalline nanosheets of tunable size up to 50 µm and thickness down to 20 nm. The FAPbBr3 nanosheets demonstrate high stability for over months of exposure to air with no degradation in surface roughness and photoluminescence efficiency. Besides, the FAPbBr3 photodetectors exhibit superior overall performance as compared to previous devices based on nonlayered perovskite nanosheets, such as an ultralow dark current of 24 pA, an ultrahigh responsivity of 1033 A/W, an external quantum efficiency over 3000%, a rapid response time around 25 ms, and a high on/off ratio of 104. This work provides a strategy to tackle the challenges of hybrid perovskites toward integrated optoelectronics with requirements of nanoscale thickness, high stability, and excellent performance.

Cooperation of Different Active Sites to Promote CO2 Electroreduction to Multi-carbon Products at Ampere-Level.

Electroreduction of CO2 to C2+ products provides a promising strategy for reaching the goal of carbon neutrality. However, achieving high selectivity of C2+ products at high current density remains a challenge. In this work, we designed and prepared a multi-sites catalyst, in which Pd was atomically dispersed in Cu (Pd-Cu). It was found that the Pd-Cu catalyst had excellent performance for producing C2+ products from CO2 electroreduction. The Faradaic efficiency (FE) of C2+ products could be maintained at approximately 80.8 %, even at a high current density of 0.8 A cm-2 for at least 20 hours. In addition, the FE of C2+ products was above 70 % at 1.4 A cm-2. Experiments and density functional theory (DFT) calculations revealed that the catalyst had three distinct catalytic sites. These three active sites allowed for efficient conversion of CO2, water dissociation, and CO conversion, ultimately leading to high yields of C2+ products.

A systematic evaluation of computation methods for cell segmentation.

Cell segmentation is a fundamental task in analyzing biomedical images. Many computational methods have been developed for cell segmentation, but their performances are not well understood in various scenarios. We systematically evaluated the performance of 18 segmentation methods to perform cell nuclei and whole cell segmentation using light microscopy and fluorescence staining images. We found that general-purpose methods incorporating the attention mechanism exhibit the best overall performance. We identified various factors influencing segmentation performances, including training data and cell morphology, and evaluated the generalizability of methods across image modalities. We also provide guidelines for choosing the optimal segmentation methods in various real application scenarios. We developed Seggal, an online resource for downloading segmentation models already pre-trained with various tissue and cell types, which substantially reduces the time and effort for training cell segmentation models.

Highly Reliable Van Der Waals Memory Boosted by a Single 2D Charge Trap Medium.

Charge trap materials that can store carriers efficiently and controllably are desired for memory applications. 2D materials are promising for highly compacted and reliable memory mainly due to their ease of constructing atomically uniform interfaces, however, remain unexplored as being charge trap media. Here it is discovered that 2D semiconducting PbI2 is an excellent charge trap material for nonvolatile memory and artificial synapses. It is simple to construct PbI2 -based charge trap devices since no complicated synthesis or additional defect manufacturing are required. As a demonstration, MoS2 /PbI2 device exhibits a large memory window of 120 V, fast write speed of 5 µs, high on-off ratio around 106 , multilevel memory of over 8 distinct states, high reliability with endurance up to 104 cycles and retention over 1.2 × 104 s. It is envisioned that PbI2 with ionic activity caused by the natively formed iodine vacancies is unique to combine with unlimited 2D materials for versatile van der Waals devices with high-integration and multifunctionality.

Optimizing copper nanoparticles with a carbon shell for enhanced electrochemical CO2 reduction to ethanol.

The electrochemical reduction of carbon dioxide (CO2RR) holds great promise for sustainable energy utilization and combating global warming. However, progress has been impeded by challenges in developing stable electrocatalysts that can steer the reaction toward specific products. This study proposes a carbon shell coating protection strategy by an efficient and straightforward approach to prevent electrocatalyst reconstruction during the CO2RR. Utilizing a copper-based metal-organic framework as the precursor for the carbon shell, we synthesized carbon shell-coated electrocatalysts, denoted as Cu-x-y, through calcination in an N2 atmosphere (where x and y represent different calcination temperatures and atmospheres: N2, H2, and NH3). It was found that the faradaic efficiency of ethanol over the catalysts with a carbon shell could reach ∼67.8%. In addition, the catalyst could be stably used for more than 16 h, surpassing the performance of Cu-600-H2 and Cu-600-NH3. Control experiments and theoretical calculations revealed that the carbon shell and Cu-C bonds played a pivotal role in stabilizing the catalyst, tuning the electron environment around Cu atoms, and promoting the formation and coupling process of CO*, ultimately favoring the reaction pathway leading to ethanol formation. This carbon shell coating strategy is valuable for developing highly efficient and selective electrocatalysts for the CO2RR.

Few-Shot Class-Incremental Learning by Sampling Multi-Phase Tasks.

New classes arise frequently in our ever-changing world, e.g., emerging topics in social media and new types of products in e-commerce. A model should recognize new classes and meanwhile maintain discriminability over old classes. Under severe circumstances, only limited novel instances are available to incrementally update the model. The task of recognizing few-shot new classes without forgetting old classes is called few-shot class-incremental learning (FSCIL). In this work, we propose a new paradigm for FSCIL based on meta-learning by LearnIng Multi-phase Incremental Tasks (Limit), which synthesizes fake FSCIL tasks from the base dataset. The data format of fake tasks is consistent with the 'real' incremental tasks, and we can build a generalizable feature space for the unseen tasks through meta-learning. Besides, Limit also constructs a calibration module based on transformer, which calibrates the old class classifiers and new class prototypes into the same scale and fills in the semantic gap. The calibration module also adaptively contextualizes the instance-specific embedding with a set-to-set function. Limit efficiently adapts to new classes and meanwhile resists forgetting over old classes. Experiments on three benchmark datasets (CIFAR100, miniImageNet, and CUB200) and large-scale dataset, i.e., ImageNet ILSVRC2012 validate that Limit achieves state-of-the-art performance.