Search

The ability to create high-quality 3D faces from a single image has become increasingly important with wide applications in video conferencing, AR/VR, and advanced video editing in movie industries. In this paper, we propose Face Diffusion NeRF (FDNeRF), a new generative method to reconstruct high-quality Face NeRFs from single images, complete with semantic editing and relighting capabilities. FDNeRF utilizes high-resolution 3D GAN inversion and expertly trained 2D latent-diffusion model, allowing users to manipulate and construct Face NeRFs in zero-shot learning without the need for explicit 3D data. With carefully designed illumination and identity preserving loss, as well as multi-modal pre-training, FD-NeRF offers users unparalleled control over the editing process enabling them to create and edit face NeRFs using just single-view images, text prompts, and explicit target lighting. The advanced features of FDNeRF have been designed to produce more impressive results than existing 2D editing approaches that rely on 2D segmentation maps for editable attributes. Experiments show that our FDNeRF achieves exceptionally realistic results and unprecedented flexibility in editing compared with state-of-the-art 3D face reconstruction and editing methods. Our code will be available at https://github.com/BillyXYB/FDNeRF.

Cross-lingual entity alignment is an important step in knowledge fusion, and its purpose is to determine whether two or more entities from different information sources refer to the same object in the real world. Recently, with the development of representation learning, many neural network-based entity alignment methods have been proposed. Existing embedding-based entity alignment methods embed entities from different information sources into a unified space, and then calculate the similarity between entity vector representations to find similar entities. Although the existing neural network-based methods have shown good performance on some public datasets, they also inherit some problems of neural networks, such as inexplicability and high model complexity, and most models are not suitable for the structural isomorphism of similar entity pairs is relatively high, which greatly limits the scalability of the model. Based on this, we propose our model on how to reduce the requirement for isomorphism—Multi-Depth Joint Entity Alignment Based on Text Information (MDJEA), which utilizes the name information and attribute information of entities, and simultaneously transforms the entity alignment problem into the allocation problem, which greatly improves the interpretability and robustness of the model, and can achieve alignment without training data. At the same time, we also combine the multi-depth information of the entity and combine the structural information at different depths to improve the alignment result. Experiments show that our method outperforms existing models by 2.7%-3.5% on each dataset.

Machine Learning (ML) research has focused on maximizing the accuracy of predictive tasks. ML models, however, are increasingly more complex, resource intensive, and costlier to deploy in resource-constrained environments. These issues are exacerbated for prediction tasks with sequential classification on progressively transitioned stages with ''happens-before'' relation between them.We argue that it is possible to ''unfold'' a monolithic single multi-class classifier, typically trained for all stages using all data, into a series of single-stage classifiers. Each single-stage classifier can be cascaded gradually from cheaper to more expensive binary classifiers that are trained using only the necessary data modalities or features required for that stage. UnfoldML is a cost-aware and uncertainty-based dynamic 2D prediction pipeline for multi-stage classification that enables (1) navigation of the accuracy/cost tradeoff space, (2) reducing the spatio-temporal cost of inference by orders of magnitude, and (3) early prediction on proceeding stages. UnfoldML achieves orders of magnitude better cost in clinical settings, while detecting multi-stage disease development in real time. It achieves within 0.1% accuracy from the highest-performing multi-class baseline, while saving close to 20X on spatio-temporal cost of inference and earlier (3.5hrs) disease onset prediction. We also show that UnfoldML generalizes to image classification, where it can predict different level of labels (from coarse to fine) given different level of abstractions of a image, saving close to 5X cost with as little as 0.4% accuracy reduction., To be published in NeurIPS'22

Recent advances like StyleGAN have promoted the growth of controllable facial editing. To address its core challenge of attribute decoupling in a single latent space, attempts have been made to adopt dual-space GAN for better disentanglement of style and content representations. Nonetheless, these methods are still incompetent to obtain plausible editing results with high controllability, especially for complicated attributes. In this study, we highlight the importance of interaction in a dual-space GAN for more controllable editing. We propose TransEditor, a novel Transformer-based framework to enhance such interaction. Besides, we develop a new dual-space editing and inversion strategy to provide additional editing flexibility. Extensive experiments demonstrate the superiority of the proposed framework in image quality and editing capability, suggesting the effectiveness of TransEditor for highly controllable facial editing., Comment: CVPR 2022. Code: https://github.com/BillyXYB/TransEditor Project page: https://billyxyb.github.io/TransEditor/

At present, immunotherapy is in full swing in the treatment of non-small cell lung cancer, melanoma, bladder cancer and other tumor species, and its treatment methods are also diverse, including tumor vaccine treatment, adoptive T cell therapy, immune checkpoint inhibitor treatment, etc, but current immunotherapy for colorectal cancer mainly focuses on immune checkpoint inhibitors (PD-1/PD-L1 and CTLA-4 inhibitors). Since PD-1/PD-L1 inhibitors have shown amazing efficacy in patients with dMMR/MSI-H advanced colorectal cancer, immune checkpoint inhibitors have gained more and more attention in the field of colorectal cancer, and the treatment options have gradually shifted from late line treatment to first-line treatment or neoadjuvant therapy, and have achieved success. This paper summarizes the research progress of immune checkpoint therapy on colorectal cancer.

An estimated 70-80% of cases of colorectal cancer liver metastasis (CRLM) are defined as initially unresectable. “Converting” to no evidence of disease (NED) status may prolong survival. The current study aimed to develop a novel scoring system that predicts the conversion outcome for initially unresectable CRLM. A total of 215 consecutive CRLM patients who received first-line systemic therapy from December 2012 to January 2020 at Sun Yat-sen University Cancer Center were enrolled in the internal cohort. Forty CRLM patients from the database of the Chinese Colorectal Cancer Multidisciplinary Team Alliance were enrolled in the external cohort. A logistic regression model was applied to identify risk factors associated with the conversion outcome. The tumor-to-liver volume ratio (TLVR) was calculated as the total tumor volume divided by the total liver volume, and its cutoff value was 0.23. Three predictors of conversion failure were identified in the internal cohort and incorporated into the C-NED score: poor tumor differentiation (1 point), number of liver metastases > 8 (1 point) and TLVR ≥ 0.23 (1 point). The conversion rate was significantly negatively associated with the C-NED score (P < 0.001). The C-indexes of the C-NED score for predicting successful conversion outcome in the internal cohort and external cohort were 0.734 (95% confidence interval (CI), 0.668-0.800) and 0.736 (95% CIs, 0.566-0.907), respectively. Median progression-free survival (PFS) time (P = 0.001) and overall survival (OS) time (P = 0.003) were statistically significant different among different C-NED score groups. Our study demonstrated that the C-NED score is an effective scoring system that indicates the actual conversion probability for initially unresectable CRLM patients before treatment, which can serve as a tool that guides optimal first-line management strategies.

Counterfactual prediction is about predicting outcome of the unobserved situation from the data. For example, given patient is on drug A, what would be the outcome if she switch to drug B. Most of existing works focus on modeling counterfactual outcome based on static data. However, many applications have time-varying confounding effects such as multiple treatments over time. How to model such time-varying effects from longitudinal observational data? How to model complex high-dimensional dependency in the data? To address these challenges, we propose Deep Recurrent Inverse TreatmEnt weighting (DeepRite) by incorporating recurrent neural networks into two-phase adjustments for the existence of time-varying confounding in modern longitudinal data. In phase I cohort reweighting we fit one network for emitting time dependent inverse probabilities of treatment, use them to generate a pseudo balanced cohort. In phase II outcome progression, we input the adjusted data to the subsequent predictive network for making counterfactual predictions. We evaluate DeepRite on both synthetic data and a real data collected from sepsis patients in the intensive care units. DeepRite is shown to recover the ground truth from synthetic data, and estimate unbiased treatment effects from real data that can be better aligned with the standard guidelines for management of sepsis thanks to its applicability to create balanced cohorts.

In view of power electronic devices need to handle higher voltage grade and capacity grade, multilevel converter has become the key to solve the problem. Cascaded inverter type is widely used because of the low harmonic pollution, high input power factor, as well as good output waveform. This paper adopts the method of model design to study the five-level inverter PWM control method systematically, and better output waveforms are gotten. The results show that the method of model-based design has higher quality and efficiency, and it is suitable for the multilevel inverter.

We study the problem of estimating the continuous response over time to interventions using observational time series---a retrospective dataset where the policy by which the data are generated is unknown to the learner. We are motivated by applications where response varies by individuals and therefore, estimating responses at the individual-level is valuable for personalizing decision-making. We refer to this as the problem of estimating individualized treatment response (ITR) curves. In statistics, G-computation formula (Robins, 1986) has been commonly used for estimating treatment responses from observational data containing sequential treatment assignments. However, past studies have focused predominantly on obtaining point-in-time estimates at the population level. We leverage the G-computation formula and develop a novel Bayesian nonparametric (BNP) method that can flexibly model functional data and provide posterior inference over the treatment response curves at both the individual and population level. On a challenging dataset containing time series from patients admitted to a hospital, we estimate responses to treatments used in managing kidney function and show that the resulting fits are more accurate than alternative approaches. Accurate methods for obtaining ITRs from observational data can dramatically accelerate the pace at which personalized treatment plans become possible.

The approval of transgenic Bacillus thuringiensis rice(ab.,Bt rice) with double-stacked traits(including a rice restorer line of cv.Bt Huahui-1 and a hybrid rice line of cv.Bt Shanyou-63,both expressing a fused Cry1Ab+Cry1Ac gene) by China′s Ministry of Agriculture(MOA) on 27 November 2009 was a momentous day for biotech crops worldwide.To date,some experiments have been carried out in labs or under controlled environmental conditions,which indicate that Bt rice can perform remarkably well against two targeted lepidopteran pests(e.g.,the stem borer,Chilo suppressalis,and leaf-folder,Cnaphalocrocis medinalis),and simultaneously offers: 1) the potential to generate economic benefits due to yield increases,and 2) overall decreases in insecticide use in China.In this study,the field experiments were conducted in order to examine the population dynamics and damage characteristics of two stem borers(i.e.,target pest,C.suppressalis and a sub-target pest,Sesamia inferens) feeding on the Bt rice with a fused Cry1Ab+Cry1Ac gene(cv.,Bt Huahui-1) and its parent line of nontransgenic rice(cv.,Minghui-63).The results indicate that no significant differences were found in the number of egg masses per 100 plants for C.suppressalis or S.inferens occurring on Bt rice in contrast to nontransgenic rice.In addition,no significant differences in egg mass densities between C.suppressalis and S.inferens were found when fed on nontransgenic rice,whereasas a significant increase in egg mass were observed for C.suppressalis relative to S.inferens fed on Bt rice.Compared with nontransgenic rice,Bt rice significantly reduced the population density(i.e.,individuals per 100 plants;84.9%—100%) of C.suppressalis larvae,while the transgenicity did not significantly influence the population abundance for S.inferens.Moreover,C.suppressalis had significantly higher larval abundances than S.inferens when they fed on nontransgenic rice,while no significant differences were observed when fed on Bt rice.Simultaneously,significantly lower percentages of dead heart or white head plants(30.8%—98.3%),and damaged plants(11.4%—96.6%) resulted from C.suppressalis,while no significant differences were detected in the damage percentages for S.inferens,as they fed on the Bt rice relative to the nontransgenic rice cultivar.Therefore,Bt rice appears to have a higher resistance to the targeted pest,C.suppressalis,while lower resistance was observed for the sub-target pest,S.inferens.Furthermore,significantly higher damage percentages were caused by C.suppressalis relative to S.inferens,regardless of the transgenicity of the rice cultivar.Thus,the stem borer,C.suppressalis,remains the key insect pest in the paddyfield.While it is presumed,owing to the significant inhibition of Bt rice on the occurrence of C.suppressalis relative to S.inferens under the commercialization usage of Bt rice,the inter-species competition and population succession may be occurring between the target and sub-target pests of stem borers.Furthermore,additional field research work needs to be conducted,especially during the rice growing season when high densities of C.suppressalis and S.inferens are both available on Bt rice to assess ecological risks of the inter-species competition and population successions between the target and sub-target insect pests.