Large Batches Research Articles

Adaptation of an rVSV Ebola vaccine purification process for rapid development of a viral vaccine candidate for SARS-CoV-2.

During the COVID-19 pandemic, long development timelines typically associated with vaccines were challenged. The urgent need for a vaccine provided a strong driver to reevaluate existing vaccine development approaches. Innovative approaches to regulatory approval were realized, including the use of platform-based technology. In collaboration with the International AIDS Vaccine Initiative, Inc. (IAVI), Merck & Co., Inc., Rahway, NJ, USA rapidly advanced an investigational SARS-CoV-2 vaccine based on the recombinant vesicular stomatitis virus (rVSV) platform used for the Ebola vaccine ERVEBO (rVSV∆G-ZEBOV-GP). An rVSV∆G-SARS-CoV-2 vaccine candidate was generated using the SARS-CoV-2 spike protein to replace the VSV G protein. The purification process development for this vaccine candidate was detailed in this paper. Areas were highlighted where the ERVEBO platform process was successfully adopted and where additional measures were needed for the SARS-CoV-2 vaccine candidate. These included: (i) endonuclease addition directly into the bioreactor prior to harvest, (ii) inclusion of a core-shell chromatography step for improved purification, and (iii) incorporation of a terminal, sterile filtration step to eliminate the need for aseptic, closed processing. High infectious virus titers were achieved in Phase 3 clinical drug substance (>108 PFUmL-1 ), and process consistency was demonstrated across four large scale batches that were completed in 6 months from clone selection.

Open acid dissolution-Ammonia solution extraction-ICP OES rapid determination of 7 trace metal elements in soil.

To overcome the corrosion of hydrofluoric acid on the ICP OES injection system in the acid dissolution system, this paper makes some improvements based on the traditional open digestion. The improved method does not require the complete removal of hydrofluoric acid. After appropriate digestion of the sample with a mixed acid, the solution can be transferred to a colorimetric tube containing ammonium hydroxide solution to give the final volume for analysis. In this paper, two-point standard curves are plotted using soil standards and process blanks, which is not only convenient but also overcomes the interference of the matrix effect. Through continuous experiments, the preferred ratio of mixed acid is 3 mL nitric acid + 5 mL hydrofluoric acid, and the concentration of ammonia solution is 0.5%. The spectral lines of the measured elements V (292.4), Cr (283.5), Co (228.6), Ni (231.6), Cu (324.7), Zn (213.8) and Pb (220.3) were determined. The method quantification limits of the seven measured elements V, Cr, Co, Ni, Cu, Zn and Pb were 0.909, 4.32, 0.269, 0.261, 0.968, 3.69 and 2.64 μg g-1, respectively, and the precision was 3.5%, 5.2%, 4.8%, 2.4%, 6.1% and 4.5%, respectively. After processing six national standard materials according to the experimental method, the measured values of each measured element were basically in agreement with the certified values, indicating that this method is fully feasible for the measurement of V, Cr, Co, Ni, Cu, Zn and Pb in soil. This method greatly improves the efficiency of pretreatment and is particularly suitable for analysing large batches of samples.

Dispersive liquid-liquid microextraction coupled with gas chromatography-mass spectrometry (GC-MS) for the determination of soy sauce aroma compounds

In the present work, a dispersive liquid-liquid microextraction (DLLME) method combined with gas chromatography-mass spectrometry (GC-MS) was successfully developed for accurate determination of aroma-active components in soy sauce. Several key experimental parameters were optimized, including the type of dispersant, vortex time and the volume ratio of extractant and dispersant. The optimal conditions were 1200 μL acetone as dispersant, 800 μL dichloromethane as extractant and 3 min of vortex mixing. The proposed method was validated, achieving good linearities with R2 ≥ 0.9941, recoveries between 82 and 129% with 4-hydroxy-2,5-dimethyl-3(2H)-furanone (HDMF) and 4-hydroxy-2 (or 5)-ethly-5 (or 2)-methyl-3(2H)-furanone (HEMF) within 84–103%, relative standard deviations ranging from 0.02 to 8.60% with HDMF and HEMF between 0.26 and 1.44%, and limits of detection and limits of quantification in the range of 0.08–35.95 μg/L and 0.25–119.83 μg/L, respectively. The entire sample processing with DLLME was completed in roughly 15 min with a total consumption of 2 mL of organic solvent, and 93 volatiles identified in the obtained extracts, with 87.1% (81 volatiles) showing relative standard deviations lower than 10%. Compared to conventional liquid-liquid extraction (LLE), DLLME exhibited obvious advantages of simplicity, rapidity, stability and low solvent consumption, favouring the extraction of acids and ketones. This is the first application of the DLLME technique for the analysis of aroma compounds in soy sauce and is suitable for rapid determination of key aroma compounds like HDMF and HEMF, especially for routine detection and large batch analysis of soy sauce, which can provide reliable information for soy sauce flavour analysis.

Admixed pellets for fast and efficient delivery of plasma enhancement gases: Investigations at AUG exploring the option for EU-DEMO

Gas and pellet injection are envisaged for particle fuelling in EU-DEMO. The gas system will provide edge and divertor fuelling and any further gas species required for operation. Pellets, mm-sized bodies formed from solid hydrogen fuel, are designed for efficient and fast core fuelling. However, they can also be employed for a more efficient delivery of plasma enhancement gases, by admixing them with the fuelling pellets. To check this option for EU-DEMO, explorative investigations have been performed at ASDEX Upgrade (AUG).The AUG system produces ice in a batch process sufficient for about 100 pellets, initially designed for operation with pure H2 or D2. On a trial basis, pellet formation was tested using an H2/D2 mixture and admixtures containing small amounts (up to 2 mol%) of N2, Ar, Kr or Xe in the D2 host. A homogeneous and reproducible ice composition was found for the H2/D2 = 1:1 case. For all the admixed gases, a depletion of the admixture in the ice with increasing atomic number is observed. Nevertheless, the fast and efficient delivery of admixed pellets was clearly demonstrated in dedicated plasma experiments at AUG. Detailed investigations showed that the Ar supplied via admixed pellets has a higher radiation efficiency and a faster radiation rise than an Ar/D2 gas puff. Furthermore, Ar density measurements in a discharge with admixed pellet injection show reasonable agreement with findings of a fading admixed species’ concentration along the ice rod and assumptions on the pellet ablation location in the plasma. Investigations performed at the Oak Ridge National Laboratory with a large batch extruder using up to 2 mol% Ne in D2 confirmed that production of much larger ice quantities can be achieved.These initial explorative investigations clearly reveal the great potential of admixed pellets, although they also demonstrate that further technology efforts are required before their benefits can be utilized.

Development of Non-Spherical Platinum Nanoparticles on Carbon Supports for Oxygen Reduction Reaction

Proton exchange membrane fuel cells are anticipated to play an important role in decarbonizing the global energy system, but the performance of platinum (Pt) catalysts must be improved to make this technology more economical. Studies have identified non-spherical Pt nanoparticles on carbon supports as promising approaches to address this challenge. However, to realize the full benefits of these strategies, the catalyst synthesis procedures must be successfully simplified and scaled up, and the catalyst must perform well in half and full-cell tests. In this study, a surfactant-free one-pot method is developed to synthesize non-spherical Pt nanoparticles on Ketjen Black carbon, which is either non-treated (Pt/KB), acid-treated (Pt/KB-O), or nitrogen-doped (Pt/KB-N). The catalysts are synthesized in both small and large batches to determine the effect of scaling up the synthesis procedure. The nitrogen-doped carbon support shows a nearly identical morphological structure with uniform distribution of non-spherical Pt nanoparticles for both small and large batches’ synthesis compared with non-treated and acid-treated carbon samples. The comparative oxygen reduction reaction (ORR) activity shows that the Pt/KB-N prepared in small and large batches has better ORR activity, which is likely caused by uniformly distributed non-spherical Pt nanoparticles on the nitrogen-doped carbon support. All three catalysts show similar ORR durability, testing from 0.5–1.0 V, while Pt/KB-O displays slightly better durability from 1.0–1.5 V for carbon corrosion. These results will help inform the implementation of shape-controlled Pt catalysts on modified carbon supports in large scale.

A novel three-step deep learning approach for the classification of breast cancer histopathological images

Breast cancer is a widespread and significant health concern among women globally. Accurately categorizing breast cancer is essential for effective treatment, ultimately improving survival rates. Moreover, deep learning (DL) has emerged as a widely adopted approach for precise medical image classification in recent years, showing promise in this domain. However, despite the availability of DL models proposed in the literature for automated classification of breast cancer histopathology images, achieving high accuracy remains challenging. A minor modification to pre-trained models and simple training strategies can further enhance model accuracy. Based on the approach, this paper proposed an anti-aliased filter in a pre-trained ResNet-34 and a novel three-step training process to improve BC histopathology image classification accuracy. The training involves systematically unfreezing layers and imposing additional constraints on the rate of change of learnable parameters. In addition, four-fold on-the-fly data augmentation enhances model generalization. The Ada-Hessian optimizer adjusts learning rates based on first and second-order gradients to improve convergence speed. The training process utilizes a large batch size to minimize the training loss associated with batch normalization layers. Even with the limited GPU size, the gradient accumulation technique achieves a large batch size. Collectively, these strategies minimize training time while maintaining or improving the accuracy of BC histopathology image classification models. In the experimental implementation, the proposed architecture achieves superior results compared to recent existing models, with an accuracy of 98.64%, recall (98.98%), precision (99.35%), F1-Score (99.17%), and MCC (97.36%) for binary classification. Similarly, the model achieves an accuracy of 95.01%, recall (95.01%), precision (94.95%), F1-Score (94.94%), and MCC (93.42%) for the eight-class category of BC images.

A Gram Scale Soft-Template Synthesis of Heteroatom Doped Nanoporous Hollow Carbon Spheres for Oxygen Reduction Reaction.

Heteroatom-doped nanoporous carbon materials with unique hierarchical structures have been shown to be promising supports and catalysts for energy conversion; however, hard-template methods are limited by their inflexibility and time-consuming process. Soft-template methods have been suggested as an alternative, but they are limited by their picky requirements for stable reactions and the few known precursors for small-batch synthesis. In this study, a gram-scale soft-template-based silica-assisted method was investigated for producing nitrogen-doped hollow nanoporous carbon spheres (N-HNCS). Nitrogen doping is accomplished during preparation with enhanced electrocatalytic activity without complicating the methodology. To investigate the effect of the unique structural characteristics of N-HNCS (specific surface area: 1250 m2 g-1; pore volume: 1.2 cm3 g-1), cobalt was introduced as an active center for the oxygen reduction reaction. Finely tuned reaction conditions resulted in well-dispersed cobalt particles with minimal agglomeration. This sheds light on the advancement of new experimental procedures for developing more active and promising non-noble catalysts in large and stable batches.

A quantitative risk assessment for the incursion of lumpy skin disease virus into Australia via long-distance windborne dispersal of arthropod vectors.

Lumpy skin disease (LSD) is an infectious disease of cattle and water buffalo caused by lumpy skin disease virus (LSDV). It is primarily transmitted mechanically by biting insects. LSDV has spread from Africa to the Middle-East, the Balkans, Caucasus, Russia, Kazakhstan, China, Asia and India, suggesting that a wide variety of arthropod vectors are capable of mechanical transmission. In 2022, LSD was detected in Indonesia, heightening awareness for Australia's livestock industries. To better understand the risk of LSDV incursion to Australia we undertook a quantitative risk assessment (QRA) looking at windborne dispersal of arthropod vectors, assuming a hypothetical situation where LSD is endemic in south-east Asia and Papua New Guinea. We estimated the risk of LSDV incursion to be low, with a median incursion rate of one incursion every 403 years, based on a model where several infectious insects (i.e. a 'small batch' of 3-5) must bite a single bovine to transmit infection. The incursion risk increases substantially to one incursion every 7-8 years if a bite from a single insect is sufficient for transmission. The risk becomes negligible (one incursion every 20,706 years) if bites from many insects (i.e. a 'large batch' of 30-50 insects) are necessary. Critically, several of our parameter estimates were highly uncertain during sensitivity analyses. Thus, a key outcome of this QRA was to better prioritise surveillance activities and to understand the key research gaps associated with LSDV in the Australasian context. The current literature shows that multiple vectors are required for successful bovine-to-vector transmission of LSDV, suggesting that our estimate of one outbreak every 403 years more accurately represents the risk to Australia; however, the role of single insects in transmission has not yet been evaluated. Similarly, attempts to transmit LSDV between bovines by Culicoides have not been successful, although midges were the highest risk vector category in our model due to the high vector-to-host ratio for midges compared to other vector categories. Our findings provide further insight into the risk of LSD to Australian cattle industries and identify the Tiwi Islands and areas east of Darwin as priority regions for LSDV surveillance, especially between December and March.

Communication-efficient federated learning with stagewise training strategy

The efficiency of communication across workers is a significant factor that affects the performance of federated learning. Though periodic communication strategy is applied to reduce communication rounds in training, the communication cost is still high when the training data distributions are not independently and identically distributed (non-IID) which is common in federated learning. Recently, some works introduce variance reduction to eliminate the effect caused by non-IID data among workers. Nevertheless the provable optimal communication complexity O(log(ST)) and convergence rate O(1/(ST)) cannot be achieved simultaneously, where S denotes the number of sampled workers in each round and T is the number of iterations. To deal with this dilemma, we propose an optimization algorithm SQUARFA that adopts stagewise training framework coupling with variance reduction and uses a quick-start phase in each loop. Theoretical results show that SQUARFA achieves both optimal convergence rate and communication complexity for both strongly convex objectives and non-convex objectives under PL condition, thus fills the gap mentioned above. Then, a variant of SQUARFA yields the optimal theoretical results for general non-convex objectives. We further extend the technique in SQUARFA to the large batch setting and achieve optimal communication complexity. Experimental results demonstrate the superiority of the proposed algorithms.

A Versatile and Efficient Method to Isolate DNA-Polymer Conjugates.

We present a facile and adaptable method to purify and isolate DNA-polymer conjugates from different uncharged homo, random, or block copolymer families. Anion exchange chromatography is used to separate the reaction solution and retrieve the excess unreacted polymer and oligonucleotide. The stationary phase has a high efficiency (25 nmol of DNA per run), facilitating the purification of large batches without compromising the peak shape and resolution. To demonstrate the versatility of this method, different types of polymers, including acrylates, methacrylates, and acrylamides containing hydrophilic and hydrophobic blocks, were purified with high yields. Additionally, DNA-polymer conjugates with various DNA block lengths were also successfully purified, further highlighting the broad applicability of this method.

Study of terbium production from enriched Gd targets via the reaction 155Gd(d,2n)155Tb

The terbium (Tb) family has attracted much attention in recent years thanks to the diagnostic and therapeutic applications of the quadruplet 149Tb, 152Tb, 155Tb and 161Tb. However, the scarce availability of Tb radioisotopes is one of the main reasons hindering its clinical applications. To increase its availability, this work proposes to use enriched gadolinium (Gd) targets to produce some Tb radioisotopes (149Tb, 152Tb, and 155Tb) via deuteron-induced reactions in cyclotrons. The production of the Auger and gamma emitter 155Tb was chosen as a case study because the 155Gd enrichment (92.8%) is the highest available from all Gd stable isotopes. The involved reaction is 155Gd(d,2n)155Tb. Using enriched thin Gd-containing targets, cross-sections of the reactions 155Gd(d,x)153,154,155,156Tb have been measured at the GIP ARRONAX cyclotron facility with a beam energy ranging from 8 MeV to 30 MeV. This measurement allows for estimating the production yield and the purity of 155Tb, and for determining the irradiation parameters for large production batches. A thick enriched 155Gd2O3 target has been then irradiated with an incident energy of 15.1 MeV and a beam current of 368 nA for 1 h. The production yield of 155Tb is 10.2 MBq/μA/h at End Of Bombardment (EOB) and the purity is 89% after 14 days of decay. These experimental values are consistent with estimation based on measured cross-sections. A comparison of the deuteron-induced and proton-induced production routes is also presented in this paper.

Novel cross wedge rolling method for producing railcar axles

Railcar axles are large-size parts manufactured in large batches using forging methods. More efficient methods of manufacturing railcar axles are currently being sought. A promising line of research in this respect is connected with the use of cross wedge rolling (CWR). Nonetheless, there exist no rolling mills that would make it possible to manufacture such large-size parts as railcar axles. A solution to this problem was inspired by the symmetric shape of a railcar axle, which makes it possible to roll this part in two passes using the same tools. Taking advantage of the potential offered by numerical modelling, simulations were made of two rolling processes for producing a railcar axle type BA302, i.e. one process was conducted with the use of two rolls and the other with two flat tools moving in the opposite directions. Numerical results demonstrated that railcar axles produced by the proposed rolling process had neither external nor internal (cracks) defects. The effectiveness of the developed CWR method was ultimately confirmed by the high quality of railcar axles that were produced (in a scale of 1:5) under industrial conditions at the Lublin University of Technology.

Cumulative and Rolling Horizon Prediction of Overall Equipment Effectiveness (OEE) with Machine Learning

Nowadays, one of the important and indispensable conditions for the effectiveness and competitiveness of industrial companies is the high efficiency of manufacturing and assembly. These enterprises based on different methods and tools systematically monitor their efficiency metrics with Key Performance Indicators (KPIs). One of these most frequently used metrics is Overall Equipment Effectiveness (OEE), the product of availability, performance and quality. In addition to monitoring, it is also necessary to predict efficiency, which can be implemented with the support of machine learning techniques. This paper presents and compares several supervised machine learning techniques amongst other polynomial regression, lasso regression, ridge regression and gradient boost regression. The aim of this article is to determine the best estimation method for semiautomatic assembly line and large batch size. The case study presented with a real industrial example gives the answer as to which of the cumulative or rolling horizon prediction methods is more accurate.

Target product profile: Trypanosoma brucei gambiense test to verify elimination.

Human African trypanosomiasis is a life-threatening parasitic infection transmitted by the tsetse fly in sub-Saharan Africa. The most common form is caused by Trypanosoma brucei gambiense, with humans as the main reservoir. Diagnosis in the field requires microscopic examination performed by specifically trained personnel. After over two decades of sustained efforts, the incidence of the disease is strongly declining, and some historically endemic countries are no longer detecting cases. The World Health Organization (WHO) has targeted the elimination of transmission of gambiense human African trypanosomiasis by 2030, defined as zero autochthonous cases for at least five consecutive years. Endemic countries reaching this goal must maintain dedicated surveillance to detect re-emergence or re-introduction. With this new agenda, new tools are needed for verification of the absence of transmission. WHO has therefore developed a target product profile calling for development of a method for population-level cross-cutting surveillance of T. b. gambiense transmission. The method needs to be performed in national or sub-national reference laboratories, and to test in parallel numerous samples shipped from remote rural areas. Among other characteristics the product profile specifies: (i)a simple specimen collection procedure; (ii)no cold-chain requirement to transfer specimens to reference laboratories; (iii)high sensitivity and specificity; (iv)high-throughput, substantially automatized; (v)low cost per specimen, when analysed in large batches; and (vi)applicable also in animals.

Comparison of forming processes for micro textured cups

Friction losses are present in all kinds of technical systems. This is always associated with additional energy consumption, which can be partly avoided as it arises. Energy savings can be achieved by a wide range of approaches. The application of friction reducing coatings on bodies in relative motion is widespread. The coating of parts is however only suitable to a limited extent for the production in large batches. Therefore, research approaches consider alternative methods to achieve friction reductions. A promising approach is the integration of miniaturized pockets on the surface of the component. In lubricated contacts, the lubricant can then accumulate in these indentations. The relative motion between two bodies thus builds up a hydrodynamic pressure and consequently reduces friction. Forming technology is suitable for the efficient production of metallic components in large quantities. The aim of this study is therefore to identify challenges in the integration of micro coining in established forming processes. Two processes from sheet metal forming and bulk metal forming are investigated. In a combined experimental and numerical approach, process-specific challenges are derived regarding the texturing process and the numerical process modelling for combined micro coining. The analysis is carried out on a cup geometry, as used in the automotive sector as tappets in the valve train. The case hardening steel 16MnCr5 (1.7131) is used as workpiece material.

P4SGD: Programmable Switch Enhanced Model-Parallel Training on Generalized Linear Models on Distributed FPGAs

Generalized linear models (GLMs) are a widely utilized family of machine learning models in real-world applications. As data size increases, it is essential to perform efficient distributed training for these models. However, existing systems for distributed training have a high cost for communication and often use large batch sizes to balance computation and communication, which negatively affects convergence. Therefore, we argue for an efficient distributed GLM training system that strives to achieve linear scalability, while keeping batch size reasonably low. As a start, we propose P4SGD, a distributed heterogeneous training system that efficiently trains GLMs through model parallelism between distributed FPGAs and through forward-communication-backward pipeline parallelism within an FPGA. Moreover, we propose a light-weight, latency-centric in-switch aggregation protocol to minimize the latency of the AllReduce operation between distributed FPGAs, powered by a programmable switch. As such, to our knowledge, P4SGD is the first solution that achieves almost linear scalability between distributed accelerators through model parallelism. We implement P4SGD on eight Xilinx U280 FPGAs and a Tofino P4 switch. Our experiments show P4SGD converges up to 6.5X faster than the state-of-the-art GPU counterpart.

Self-assembly of MOF-801 into robust hierarchically porous monoliths for scale-up atmospheric water harvesting

MOFs are becoming one of the most promising sorbents for atmospheric water harvesting (AWH), but their powder form limits the scale-up application. Conventional shaping methods tend to deteriorate the sorption performance or weaken the mechanical strength of the sorbents. Here, monolithic MOF-801 was constructed by coupling sol–gel phase separation and acid-heat post-treatment method. The porosity can be optimized up to 59.7% by regulating the phase separation progress and gelation time, and the specific surface area and micropore volume can be optimized up to 977.8 m2 g−1 and 0.234 cm3 g−1 by the acid-heat post-treatment. The special properties ensured the monolithic MOF-801 with an excellent moisture sorption capacity of 1.2 times higher than that of the conventional MOF-801 powder in small batches and 1.4 times higher than that of the conventional compressed tablet in large batches at 30% RH. The monolithic MOF-801 had centimeter-scale morphology and good mechanical strength with compressive Young's modulus up to 61.3 MPa. This study provides a feasible strategy for achieving the monolithic shaping of MOFs for industrial-scale AWH.

Foreign RNA spike-ins enable accurate allele-specific expression analysis at scale.

Analysis of allele-specific expression is strongly affected by the technical noise present in RNA-seq experiments. Previously, we showed that technical replicates can be used for precise estimates of this noise, and we provided a tool for correction of technical noise in allele-specific expression analysis. This approach is very accurate but costly due to the need for two or more replicates of each library. Here, we develop a spike-in approach which is highly accurate at only a small fraction of the cost. We show that a distinct RNA added as a spike-in before library preparation reflects technical noise of the whole library and can be used in large batches of samples. We experimentally demonstrate the effectiveness of this approach using combinations of RNA from species distinguishable by alignment, namely, mouse, human, and Caenorhabditis elegans. Our new approach, controlFreq, enables highly accurate and computationally efficient analysis of allele-specific expression in (and between) arbitrarily large studies at an overall cost increase of ∼5%. Analysis pipeline for this approach is available at GitHub as R package controlFreq (github.com/gimelbrantlab/controlFreq).

CowClip: Reducing CTR Prediction Model Training Time from 12 Hours to 10 Minutes on 1 GPU

The click-through rate (CTR) prediction task is to predict whether a user will click on the recommended item. As mind-boggling amounts of data are produced online daily, accelerating CTR prediction model training is critical to ensuring an up-to-date model and reducing the training cost. One approach to increase the training speed is to apply large batch training. However, as shown in computer vision and natural language processing tasks, training with a large batch easily suffers from the loss of accuracy. Our experiments show that previous scaling rules fail in the training of CTR prediction neural networks. To tackle this problem, we first theoretically show that different frequencies of ids make it challenging to scale hyperparameters when scaling the batch size. To stabilize the training process in a large batch size setting, we develop the adaptive Column-wise Clipping (CowClip). It enables an easy and effective scaling rule for the embeddings, which keeps the learning rate unchanged and scales the L2 loss. We conduct extensive experiments with four CTR prediction networks on two real-world datasets and successfully scaled 128 times the original batch size without accuracy loss. In particular, for CTR prediction model DeepFM training on the Criteo dataset, our optimization framework enlarges the batch size from 1K to 128K with over 0.1% AUC improvement and reduces training time from 12 hours to 10 minutes on a single V100 GPU. Our code locates at github.com/bytedance/LargeBatchCTR.

Faster Adaptive Federated Learning

Federated learning has attracted increasing attention with the emergence of distributed data. While extensive federated learning algorithms have been proposed for the non-convex distributed problem, the federated learning in practice still faces numerous challenges, such as the large training iterations to converge since the sizes of models and datasets keep increasing, and the lack of adaptivity by SGD-based model updates. Meanwhile, the study of adaptive methods in federated learning is scarce and existing works either lack a complete theoretical convergence guarantee or have slow sample complexity. In this paper, we propose an efficient adaptive algorithm (i.e., FAFED) based on the momentum-based variance reduced technique in cross-silo FL. We first explore how to design the adaptive algorithm in the FL setting. By providing a counter-example, we prove that a simple combination of FL and adaptive methods could lead to divergence. More importantly, we provide a convergence analysis for our method and prove that our algorithm is the first adaptive FL algorithm to reach the best-known samples O(epsilon(-3)) and O(epsilon(-2)) communication rounds to find an epsilon-stationary point without large batches. The experimental results on the language modeling task and image classification task with heterogeneous data demonstrate the efficiency of our algorithms.