Search

When urban rail transit (URT) does not provide 24-hour services, passengers who travel at late night may not be able to reach their destinations with only URT trains. As a result, passengers have to find alternative transport means, or combine URT trains with other transport services to fulfill their journeys. This paper investigates the integrated optimization of last train timetabling and bridging service design with consideration of passenger path choices. Two bridging services are considered: taxis and buses. Based on pre-constructed path sets, a bi-objective mixed-integer nonlinear programming (MINLP) model is developed, aiming at minimizing total passenger travel time and total passenger travel cost. To reduce the model scale and improve solution efficiency, three path dominance principles are proposed to remove redundant passenger paths without loss of optimality. An adaptive iterative algorithm is designed to obtain the Pareto frontier curve. The proposed model and solution methods are demonstrated on the Chengdu URT network. Results indicate that passenger travel costs and travel times can be significantly reduced by the integrated optimization. It also provides passengers with a safer night travel environment due to the reduction in passenger travel times in taxis.

In convolutional neural networks, the convolutions are conventionally performed using a square kernel with a fixed N $\times$ N receptive field (RF). However, what matters most to the network is the effective receptive field (ERF) that indicates the extent with which input pixels contribute to an output pixel. Inspired by the property that ERFs typically exhibit a Gaussian distribution, we propose a Gaussian Mask convolutional kernel (GMConv) in this work. Specifically, GMConv utilizes the Gaussian function to generate a concentric symmetry mask that is placed over the kernel to refine the RF. Our GMConv can directly replace the standard convolutions in existing CNNs and can be easily trained end-to-end by standard back-propagation. We evaluate our approach through extensive experiments on image classification and object detection tasks. Over several tasks and standard base models, our approach compares favorably against the standard convolution. For instance, using GMConv for AlexNet and ResNet-50, the top-1 accuracy on ImageNet classification is boosted by 0.98% and 0.85%, respectively., Comment: 10 pages, 8 figures

Unlike language tasks, where the output space is usually limited to a set of tokens, the output space of visual tasks is more complicated, making it difficult to build a unified visual model for various visual tasks. In this paper, we seek to unify the output space of visual tasks, so that we can also build a unified model for visual tasks. To this end, we demonstrate a single unified model that simultaneously handles two typical visual tasks of instance segmentation and depth estimation, which have discrete/fixed-length and continuous/varied-length outputs, respectively. We propose several new techniques that take into account the particularity of visual tasks: 1) Soft token. We employ soft token to represent the task output. Unlike hard tokens in the common VQ-VAE which are assigned one-hot to discrete codebooks/vocabularies, the soft token is assigned softly to the codebook embeddings. Soft token can improve the accuracy of both the next token inference and decoding of the task output; 2) Mask augmentation. Many visual tasks have corruption, undefined or invalid values in label annotations, i.e., occluded area of depth maps. We show that a mask augmentation technique can greatly benefit these tasks. With these new techniques and other designs, we show that the proposed general-purpose task-solver can perform both instance segmentation and depth estimation well. Particularly, we achieve 0.279 RMSE on the specific task of NYUv2 depth estimation, setting a new record on this benchmark. The general-purpose task-solver, dubbed AiT, is available at \url{https://github.com/SwinTransformer/AiT}.

Normalization techniques have been widely used in the field of deep learning due to their capability of enabling higher learning rates and are less careful in initialization. However, the effectiveness of popular normalization technologies is typically limited to specific areas. Unlike the standard Batch Normalization (BN) and Layer Normalization (LN), where BN computes the mean and variance along the (N,H,W) dimensions and LN computes the mean and variance along the (C,H,W) dimensions (N, C, H and W are the batch, channel, spatial height and width dimension, respectively), this paper presents a novel normalization technique called Batch Channel Normalization (BCN). To exploit both the channel and batch dependence and adaptively and combine the advantages of BN and LN based on specific datasets or tasks, BCN separately normalizes inputs along the (N, H, W) and (C, H, W) axes, then combines the normalized outputs based on adaptive parameters. As a basic block, BCN can be easily integrated into existing models for various applications in the field of computer vision. Empirical results show that the proposed technique can be seamlessly applied to various versions of CNN or Vision Transformer architecture. The code is publicly available at https://github.com/AfifaKhaled/BatchChannel-Normalization

In this paper, we explore FP8 low-bit data formats for efficient training of large language models (LLMs). Our key insight is that most variables, such as gradients and optimizer states, in LLM training can employ low-precision data formats without compromising model accuracy and requiring no changes to hyper-parameters. Specifically, we propose a new FP8 automatic mixed-precision framework for training LLMs. This framework offers three levels of FP8 utilization to streamline mixed-precision and distributed parallel training for LLMs. It gradually incorporates 8-bit gradients, optimizer states, and distributed learning in an incremental manner. Experiment results show that, during the training of GPT-175B model on H100 GPU platform, our FP8 mixed-precision training framework not only achieved a remarkable 39% reduction in real memory usage but also ran 75% faster than the widely adopted BF16 framework (i.e., Megatron-LM), surpassing the speed of Nvidia Transformer Engine by 37%. This largely reduces the training costs for large foundation models. Furthermore, our FP8 mixed-precision training methodology is generic. It can be seamlessly applied to other tasks such as LLM instruction tuning and reinforcement learning with human feedback, offering savings in fine-tuning expenses. Our FP8 low-precision training framework is open-sourced at {https://github.com/Azure/MS-AMP}{aka.ms/MS.AMP}.

Differentiable architecture search (DARTS) has attracted much attention due to its simplicity and significant improvement in efficiency. However, the excessive accumulation of the skip connection makes it suffer from long-term weak stability and low robustness. Many works attempt to restrict the accumulation of skip connections by indicators or manual design, however, these methods are susceptible to thresholds and human priors. In this work, we suggest a more subtle and direct approach that removes skip connections from the operation space. Then, by introducing an adaptive channel allocation strategy, we redesign the DARTS framework to automatically refill the skip connections in the evaluation stage, resolving the performance degradation caused by the absence of skip connections. Our method, dubbed Adaptive-Channel-Allocation-DARTS (ACA-DRATS), could eliminate the inconsistency in operation strength and significantly expand the architecture diversity. We continue to explore smaller search space under our framework, and offer a direct search on the entire ImageNet dataset. Experiments show that ACA-DRATS improves the search stability and significantly speeds up DARTS by more than ten times while yielding higher accuracy.

Jia-Qi Ning,1,2 Jian-Sheng Luo,1,2 Ling-Ling Ding,2 Yu-Hong Guo,3 Zhuo-Ya Chen,2 Qi Wang,2 Rui-Ling Zhou2 1Department of Anesthesiology, Capital Medical University, Beijing, 100069, People’s Republic of China; 2Department of Anesthesiology, Beijing Hospital of Traditional Chinese Medicine, Capital Medical University, Beijing, 100010, People’s Republic of China; 3Department of Emergency, Beijing Hospital of Traditional Chinese Medicine, Capital Medical University, Beijing, 100010, People’s Republic of ChinaCorrespondence: Ling-Ling Ding, Department of Anesthesiology, Beijing Hospital of Traditional Chinese Medicine, Capital Medical University, Beijing, 100010, People’s Republic of China, Tel/Fax +86 1087906647, Email dignlilbc@outlook.com; dinglingling301@126.comObjective: This study aimed to evaluate the effect of electroacupuncture preconditioning on regional cerebral oxygen saturation (rSO2) levels in elderly patients with diabetes.Methods: Forty patients undergoing elective diabetic foot surgery were enrolled in this study. All patients were aged 65 years and above and weighed 45– 75 kg. All were characterized as class II or III according to the American Society of Anesthesiologists’ physical status classification system. Patients were divided randomly into an electroacupuncture group (group E) and a control group (group C); both groups comprised 20 patients. In group E, the DU20 (Baihui), DU24 (Shenting), and EX-HN1 (Sishencong) acupoints were selected for electroacupuncture 30 min prior to administering anesthesia, while in group C, patients underwent routine anesthesia without electroacupuncture. The patients in both groups were anesthetized using a sciatic nerve block. The number of cases with increased or decreased regional oxygen saturation (rSO2) compared with the baseline as well as rSO2 variability in the two groups were recorded and compared.Results: There was no significant difference in the preoperative rSO2 values between the two groups (

Large-scale NLP models have been shown to significantly improve the performance on language tasks with no signs of saturation. They also demonstrate amazing few-shot capabilities like that of human beings. This paper aims to explore large-scale models in computer vision. We tackle three major issues in training and application of large vision models, including training instability, resolution gaps between pre-training and fine-tuning, and hunger on labelled data. Three main techniques are proposed: 1) a residual-post-norm method combined with cosine attention to improve training stability; 2) A log-spaced continuous position bias method to effectively transfer models pre-trained using low-resolution images to downstream tasks with high-resolution inputs; 3) A self-supervised pre-training method, SimMIM, to reduce the needs of vast labeled images. Through these techniques, this paper successfully trained a 3 billion-parameter Swin Transformer V2 model, which is the largest dense vision model to date, and makes it capable of training with images of up to 1,536$\times$1,536 resolution. It set new performance records on 4 representative vision tasks, including ImageNet-V2 image classification, COCO object detection, ADE20K semantic segmentation, and Kinetics-400 video action classification. Also note our training is much more efficient than that in Google's billion-level visual models, which consumes 40 times less labelled data and 40 times less training time. Code is available at \url{https://github.com/microsoft/Swin-Transformer}.

The vision community is witnessing a modeling shift from CNNs to Transformers, where pure Transformer architectures have attained top accuracy on the major video recognition benchmarks. These video models are all built on Transformer layers that globally connect patches across the spatial and temporal dimensions. In this paper, we instead advocate an inductive bias of locality in video Transformers, which leads to a better speed-accuracy trade-off compared to previous approaches which compute self-attention globally even with spatial-temporal factorization. The locality of the proposed video architecture is realized by adapting the Swin Transformer designed for the image domain, while continuing to leverage the power of pre-trained image models. Our approach achieves state-of-the-art accuracy on a broad range of video recognition benchmarks, including on action recognition (84.9 top-1 accuracy on Kinetics-400 and 86.1 top-1 accuracy on Kinetics-600 with ~20x less pre-training data and ~3x smaller model size) and temporal modeling (69.6 top-1 accuracy on Something-Something v2). The code and models will be made publicly available at https://github.com/SwinTransformer/Video-Swin-Transformer.

Toxicokinetics influences the toxicity of chemicals. This also holds for 1,2-unsaturated pyrrolizidine alkaloids (PAs), which need bioactivation to become toxic. Given that only for a limited number of 1,2-unsaturated PAs in vivo toxicity data are available, alternative testing strategies including read-across and quantitative in vitro to in vivo extrapolation (QIVIVE) are important. This paper presents how physiologically-based kinetic (PBK) models for the PAs lasiocarpine and riddelliine were developed for rat and human, and used for conversion of in vitro data for toxicity in primary hepatocytes to quantitatively predict in vivo acute liver toxicity for both rat and human. Marked differences in toxicokinetics were observed between the two model PAs influencing the predicted in vivo toxicity. In a next step, in vitro toxicokinetic data that predicted relative bioactivation of the PAs, were shown to provide a possible basis for read-across from the BMDL10 for tumor formation by riddelliine of 237 μg/kg bw per day to other PAs for which tumor data are lacking. It is concluded that when comparing toxicity of different PAs, or when extrapolating in vitro toxicity data for PAs to the in vivo situation, differences in toxicokinetics should be taken into account, while future challenges are also discussed.

The aim of the present thesis was to perform the risk and safety assessment of two groups of natural food-borne toxins, namely alkenylbenzenes and pyrrolizidine alkaloids (PAs) for the Chinese population as compared to the overall Caucasian population using methods including the Margin of Exposure (MOE) approach, physiologically based kinetic (PBK) modelling-based reverse dosimetry and the combination of PBK modelling and Monte Carlo simulation. Chapter 1 introduced the model compounds studied in the thesis, namely alkenylbenzenes and 1,2-unsaturated PAs, the aim of the thesis and the strategies applied to perform a mode of action based safety and risk assessment. The regulations and consumer behaviour towards botanical preparations in China and in several EU countries were also presented. In chapter 2, a risk assessment of 10 plant food supplements (PFS), 23 traditional Chinese medicines (TCM) and 38 herbal teas containing alkenylbenzenes obtained at the Chinese market was performed using the Margin of Exposure (MOE) approach. Use of about half of the samples would result in estimated intakes upon daily life-time exposure that give rise to MOE values lower than 10 000, suggesting a potential priority for risk management. For short-term exposure such as two weeks consumption, applying Haber’s rule, only one TCM (TCM 6) still had a MOE value below 10 000. It is concluded that consumption of Chinese botanical preparations raise a concern because of exposure to alkenylbenzenes, especially when exposure is for longer periods of time. Chapter 3 studied the differences in bioactivation and detoxification of the food-borne genotoxic carcinogen estragole between Chinese and Caucasians using a mode of action based PBK modelling approach. The outcomes of the model predictions showed that the detoxification of the proximate carcinogenic metabolite 1’-hydroxyestragole, mainly by conversion to 1’-oxoestragole, was similar in both ethnic groups. For the bioactivation pathway, at

Lasiocarpine and riddelliine are pyrrolizidine alkaloids (PAs) known to cause liver toxicity. The aim of this study was to predict the inter-species and inter-ethnic human differences in acute liver toxicity of lasiocarpine and riddelliine using physiologically based kinetic (PBK) modelling based reverse dosimetry of in vitro toxicity data. The concentration–response curves of in vitro cytotoxicity of lasiocarpine and riddelliine defined in pooled human hepatocytes were translated to in vivo dose–response curves by PBK models developed using kinetic data obtained from incubations with pooled tissue fractions from Chinese and Caucasian individuals, providing PBK models for the average Chinese and average Caucasian, respectively. From the predicted in vivo dose–response curves, the benchmark dose lower and upper confidence limits for 5% effect (BMDL5 and BMDU5) were derived and subsequently compared to those previously obtained in rat to evaluate inter-species differences. The inter-species differences amounted to 2.0-fold for lasiocarpine and 8.2-fold for riddelliine with humans being more sensitive than rats. The inter-ethnic human differences varied 2.0-fold for lasiocarpine and 5.0-fold for riddelliine with the average Caucasian being more sensitive than the average Chinese. In conclusion, the present study provides the proof-of-principle to predict inter-species and inter-ethnic differences in in vivo liver toxicity for PAs by an alternative testing strategy integrating in vitro cytotoxicity data with PBK modelling-based reverse dosimetry.

The aim of the present study was to predict the effect of inter-individual and inter-ethnic human kinetic variation on the sensitivity towards acute liver toxicity of lasiocarpine in the Chinese and the Caucasian population, and to derive chemical specific adjustment factors (CSAFs) by integrating variation in the in vitro kinetic constants Vmax and Km, physiologically based kinetic (PBK) modelling and Monte Carlo simulation. CSAFs were derived covering the 90th and 99th percentile of the population distribution of pyrrole glutathione adduct (7-GS-DHP) formation, reflecting bioactivation. The results revealed that in the Chinese population, as compared to the Caucasian population, the predicted 7-GS-DHP formation at the geometric mean, the 90th and the 99th percentile were 2.1-, 3.3- and 4.3-fold lower respectively. The CSAFs obtained using the 99th percentile values were 8.3, 17.0 and 19.5 in the Chinese, the Caucasian population and the two populations combined, respectively, while the CSAFs were generally 3.0-fold lower at the 90th percentile. These results indicate that when considering the formation of 7-GS-DHP the Caucasian population may be more sensitive towards acute liver toxicity of lasiocarpine, and further point out that the default safety factor of 3.16 for inter-individual human kinetic differences may not be sufficiently protective. Altogether, the results obtained demonstrate that integrating PBK modelling with Monte Carlo simulations using human in vitro data is a powerful strategy to quantify inter-individual variations in kinetics, and can be used to refine the human risk assessment of pyrrolizidine alkaloids.

In the present study, a risk assessment of plant food supplements (PFS), traditional Chinese medicines (TCM) and herbal teas containing alkenylbenzenes was performed using the Margin of Exposure (MOE) approach. The levels of alkenylbenzenes in botanical preparations collected on the Chinese market were quantified and the combined estimated daily intake (EDI) was determined using dose additivity. The combined EDI values obtained assuming equal potency of all alkenylbenzenes detected in the PFS, TCM and herbal teas were 0.3 to 14.3, 0.05 to 539.4 and 0.04 to 42.5 μg/kg bw/day, respectively. Calculating combined EDI values taking into account the toxic equivalency (TEQ) approach, the values for PFS, TCM and herbal teas were 0.3 to 7.7, 0.05 to 278.0 and 0.02 to 16.5 μg estragole equivalents/kg bw/day, respectively. The MOE values resulting from consumption of these PFS, TCM and one cup of herbal tea per day during life-time were generally lower than 10 000, suggesting a potential priority for risk management. For short-term exposure such as two weeks consumption, applying Haber's rule, only one TCM 6 (四神丸) still had an MOE value below 10 000. It is concluded that selected consumption of Chinese botanical preparations raise a concern because of exposure to alkenylbenzenes, especially when exposure is for longer periods of time.

Yunjia Song,1,2 Aobo Ma,1 Jia Ning,1 Xue Zhong,1 Qian Zhang,1 Xu Zhang,1 Guang Hong,3,4 Ying Li,1 Keiichi Sasaki,2 Changyi Li1 1School of Dentistry, Stomatological Hospital, Tianjin Medical University, Tianjin, China; 2Division of Advanced Prosthetic Dentistry, Tohoku University Graduate School of Dentistry, Sendai, Japan; 3Liaison Center for Innovative Dentistry, Tohoku University Graduate School of Dentistry, Sendai, Japan; 4Faculty of Dental Medicine, Airlangga University, Surabaya, Indonesia Purpose: Icariin (ICA) is one of the main active constituents of Herba Epimedii for improving osteogenesis. It is necessary to create a simple and efficient method to load ICA onto the surface of titanium (Ti) implant. The purpose of this study was to establish a local ICA delivery system via a layer-by-layer (LbL) self-assembly system on phase-transited lysozyme (PTL)-primed Ti surface.Materials and methods: A PTL nanofilm was first firmly coated on the pristine Ti. Then, the ICA-loaded hyaluronic acid/chitosan (HA/CS) multilayer was applied via the LbL system to form the HA/CS-ICA surface. This established HA/CS-ICA surface was characterized by scanning electron microscopy (SEM), X-ray photoelectron spectroscopy (XPS) and contact angle measurement. The ICA release pattern of the HA/CS-ICA surface was also examined. MC3T3-E1 osteoblast culture test and a rat model were used to evaluate the effects of the HA/CS-ICA surface in vitro and in vivo.Results: SEM, XPS and contact angle measurement demonstrated successful fabrication of the HA/CS-ICA surface. The HA/CS-ICA surfaces with different ICA concentrations revealed a controlled release profile of ICA during a 2-week monitoring span. Osteoblasts grown on the coated substrates displayed higher adhesion, viability, proliferation and ALP activity than those on the polished Ti surface. Furthermore, in vivo histological evaluation revealed much obvious bone formation in the ICA-coated group by histological staining and double flu

Cyclophilin 40 (Cyp40) is a member of the immunophilin family that acts as a peptidyl-prolyl-isomerase enzyme and binds to the heat shock protein 90 (Hsp90). Its structure comprises an N-terminal cyclophilin domain and a C-terminal tetratricopeptide (TPR) domain. Cyp40 is over-expressed in prostate cancer and certain T-cell lymphomas. The groove for Hsp90 binding on the TPR domain includes residues Lys227 and Lys308, referred to as the carboxylate clamp, and is essential for Cyp40-Hsp90 binding. In this study, the effect of two mutations, K227A and K308A, and their combinative mutant was investigated by performing a total of 5.76 mu s of all-atom molecular dynamics (MD) simulations in explicit solvent. All simulations, except the K308A mutant, were found to adopt two distinct (extended or compact) conformers defined by different cyclophilin-TPR interdomain distances. The K308A mutant was only observed in the extended form which is observed in the Cyp40 X-ray structure. The wild-type, K227A, and combined mutant also showed bimodal distributions. The experimental melting temperature, T-m, values of the mutants correlate with the degree of compactness with the K308A extended mutant having a marginally lower melting temperature. Another novel measure of compactness determined from the MD data, the "coordination shell volume," also shows a direct correlation with Tm. In addition, the MD simulations show an allosteric effect with the mutations in the remote TPR domain having a pronounced effect on the molecular motions of the enzymatic cyclophilin domain which helps rationalise the experimentally observed increase in enzyme activity measured for all three mutations., Istanbul Technical University Scientific Research Projects (BAP) Unit; British Council UK-Turkey Partnership Programme; Edinburgh University Protein Production Facility (EPPF); Wellcome Trust

Lasiocarpine and riddelliine are pyrrolizidine alkaloids (PAs) present in food and able to cause liver toxicity. The aim of this study was to investigate whether physiologically based kinetic (PBK) modelling-facilitated reverse dosimetry can adequately translate in vitro concentration-response curves for toxicity of lasiocarpine and riddelliine to in vivo liver toxicity data for the rat. To this purpose, PBK models were developed for lasiocarpine and riddelliine, and predicted blood concentrations were compared to available literature data to evaluate the models. Concentration-response curves obtained from in vitro cytotoxicity assays in primary rat hepatocytes were converted to in vivo dose-response curves from which points of departure (PODs) were derived and that were compared to available literature data on in vivo liver toxicity. The results showed that the predicted PODs fall well within the range of PODs derived from available in vivo toxicity data. To conclude, this study shows the proof-of-principle for a method to predict in vivo liver toxicity for PAs by an alternative testing strategy integrating in vitro cytotoxicity assays with in silico PBK modelling-facilitated reverse dosimetry. The approach may facilitate prediction of acute liver toxicity for the large number of PAs for which in vivo toxicity data are lacking.

Considering the rapid developments in food safety in the past decade in China, it is of importance to obtain insight into what extent safety and risk assessments of chemicals performed for the Caucasian population apply to the Chinese population. The aim of the present study was to determine physiologically based kinetic (PBK) modeling-based predictions for differences between Chinese and Caucasians in terms of metabolic bioactivation and detoxification of the food-borne genotoxic carcinogen estragole. The PBK models were defined based on kinetic constants for hepatic metabolism derived from in vitro incubations using liver fractions of the two ethnic groups, and used to evaluate the inter-ethnic differences in metabolic activation and detoxification of estragole. The models predicted that at realistic dietary intake levels, only 0.02% of the dose was converted to the ultimate carcinogenic metabolite 1′-sulfooxyestragole in Chinese subjects, whereas this amounted to 0.09% of the dose in Caucasian subjects. Detoxification of 1′-hydroxyestragole, mainly via conversion to 1′-oxoestragole, was similar within the two ethnic groups. The 4.5-fold variation in formation of the ultimate carcinogenic metabolite of estragole accompanied by similar rates of detoxification may indicate a lower risk of estragole for the Chinese population at similar levels of exposure. The study provides a proof of principle for how PBK modeling can identify differences in ethnic sensitivity and provide a more refined risk assessment for a specific ethnic group for a compound of concern.

Cyclophilin 40 (Cyp40) comprises an N-terminal cyclophilin domain with peptidyl-prolyl isomerase (PPIase) activity and a C-terminal tetratricopeptide repeat (TPR) domain that binds to the C-terminal -EEVD sequence common to both heat shock protein 70 (Hsp70) and Hsp90. We show in the present study that binding of peptides containing the MEEVD motif reduces the PPIase activity by similar to 30%. CD and fluorescence assays show that the TPR domain is less stable than the cyclophilin domain and is stabilized by peptide binding. Isothermal titration calorimetry (ITC) shows that the affinity for the -MEEVD peptide is temperature sensitive in the physiological temperature range. Results from these biophysical studies fit with the MD simulations of the apo and holo (peptide-bound) structures which show a significant reduction in root mean square (RMS) fluctuation in both TPR and cyclophilin domains when -MEEVD is bound. The MD simulations of the apo-protein also highlight strong anti-correlated motions between residues around the PPIase-active site and a band of residues running across four of the seven helices in the TPR domain. Peptide binding leads to a distortion in the shape of the active site and a significant reduction in these strongly anti-correlated motions, providing an explanation for the allosteric effect of ligand binding and loss of PPIase activity. Together the experimental and MD results suggest that on heat shock, dissociation of Cyp40 from complexes mediated by the TPR domain leads to an increased pool of free Cyp40 capable of acting as an isomerase/chaperone in conditions of cellular stress., British Council UK-Turkey Partnership Programme; Wellcome Trust; Biotechnology and Biological Sciences Research Council; Wellcome-UoE ISSF award; Scottish Universities Life Science Alliance (SULSA)