Unbalanced Feature Transport for Exemplar-based Image Translation

Despite the great success of GANs in images translation with different conditioned inputs such as semantic segmentation and edge maps, generating high-fidelity realistic images with reference styles remains a grand challenge in conditional image-to-image translation. This paper presents a general image translation framework that incorporates optimal transport for feature alignment between conditional inputs and style exemplars in image translation. The introduction of optimal transport mitigates the constraint of many-to-one feature matching significantly while building up accurate semantic correspondences between conditional inputs and exemplars. We design a novel unbalanced optimaltransport to address the transport between features with deviational distributions which exists widely between conditional inputs and exemplars. In addition, we design a semantic-activation normalization scheme that injects stylefeatures of exemplars into the image translation process successfully. Extensive experiments over multiple image translation tasks show that our method achieves superior image translation qualitatively and quantitatively as comparedwith the state-of-the-art.

Representative Learning via Span-Based Mutual Information for PolSAR Image Classification

The optimal parameters of polarimetric scattering decomposition are critical to classify the pixels in polarimetric synthetic aperture radar (PolSAR) images by utilizing the method of machine learning. Therefore, span-based mutual information (Sp-MI) is proposed to lighten the dependence on labeling information, and then a heuristic representative learning scheme is also given by artificial neural network (ANN) to classify parameters separately with the increasing sequence according to the values of Sp-MI. Furthermore, an innovative method of using the sine function is presented to map the parameters of angular, and a min-max scaling method is applied to complete the procedure of normalization. Except for the support vector machine, three ANN-based classifiers are implemented to verify the rationality and effectiveness of the proposed representative learning and normalization scheme. Meanwhile, the classification method is compared with four similar comparison methods on three real PolSAR images. Finally, the classification results show the effectiveness of the proposed Sp-MI and the validation of the representative learning scheme in the aspect of classification overall accuracy and visual effect.

Instance-level contrastive learning yields human brain-like representation without category-supervision

Humans learn object categories without millions of labels, but to date the models with the highest correspondence to primate visual systems are all category-supervised. This paper introduces a new self-supervised learning framework: instance-prototype contrastive learning (IPCL), and compares the internal representations learned by this model and other instance-level contrastive learning systems to the structure of human brain responses. We present the first evidence to date showing that self-supervised systems can show more brain-like representation than category-supervised models. Further, we find that recent substantial gains in top-1 accuracy from instance-wise contrastive learning models do not result in more brain-like representation—instead we find the architecture and normalization scheme are critical. Finally, this dataset reveals substantial representational structure in intermediate and late stages of the human visual system that is not accounted for by any model, whether self-supervised or category-supervised. Considering both neuroscience and machine vision perspectives, these results provide promise for instance-level representation as a key objective of visual system encoding, and highlight the room to grow towards more robust, efficient, human-like object representation.

Robust classification of Immune Subtypes in Cancer

As part of the ‘immune landscape of cancer’, six immune subtypes were defined which describe a categorization of tumor-immune states. A number of phenotypic variables were found to associate with immune subtypes, such as nonsilent mutation rates, regulation of immunomodulator genes, and cytokine network structures. An ensemble classifier based on XGBoost is introduced with the goal of classifying tumor samples into one of six immune subtypes. Robust performance was accomplished through feature engineering; quartile-levels, binary gene-pair features, and gene-set-pair features were computed for each sample independently. The classifier is robust to software pipeline and normalization scheme, making it applicable to any expression data format from raw count data to TPMs since the classification is essentially based on simple binary gene-gene level comparisons within a given sample. The classifier is available as an R package or part of the CRI iAtlas portal.Code / Tool availabilitySource Code https://github.com/Gibbsdavidl/ImmuneSubtypeClassifierWeb App Tool https://www.cri-iatlas.org/

A Comparison of Regularization Techniques in Deep Neural Networks

Artificial neural networks (ANN) have attracted significant attention from researchers because many complex problems can be solved by training them. If enough data are provided during the training process, ANNs are capable of achieving good performance results. However, if training data are not enough, the predefined neural network model suffers from overfitting and underfitting problems. To solve these problems, several regularization techniques have been devised and widely applied to applications and data analysis. However, it is difficult for developers to choose the most suitable scheme for a developing application because there is no information regarding the performance of each scheme. This paper describes comparative research on regularization techniques by evaluating the training and validation errors in a deep neural network model, using a weather dataset. For comparisons, each algorithm was implemented using a recent neural network library of TensorFlow. The experiment results showed that an autoencoder had the worst performance among schemes. When the prediction accuracy was compared, data augmentation and the batch normalization scheme showed better performance than the others.

Use of Modified Temperature-Composition PDF Formulation in Modeling of Flame Dynamics in Diesel Engine Combustion

In the present work, the modified temperature-composition (MT-C) PDF formulation was embedded in the KIVA to study the characteristics of flame development and emissions in a diesel engine. The model uses a time scale defined by an energy balance on the flame surface and a new normalization scheme exploiting the maximum attainable mass fractions of progress variables. Development of the latter in the {\rm{T}} - {{\xi }} parameter space regulates the flame progress in the physical space and, thus, the approach presents some potential to capture the local flame extinction. The interactions of the swirl and spray penetration and their influence in the mixing process, combustion and emissions are also evaluated. Analyses of the temporal evolution of mixture fraction and temperature show that the swirl motion forms a homogeneous mixture on the lee sides of the spray jets where the ignition actually starts. Since the local time scales are considered in the model, the chemistry-controlled premixed combustion developing there is well predicted.

Prediction of Thin-Walled Areas of Unruptured Cerebral Aneurysms through Comparison of Normalized Hemodynamic Parameters and Intraoperative Images

Object. Rupture of a cerebral aneurysm occurs mainly in a thin-walled area (TWA). Prediction of TWAs would help to assess the risk of rupture and select appropriate treatment strategy. There are several limitations of current prediction techniques for TWAs. To predict TWAs more accurately, HP should be normalized to minimize the influence of analysis conditions, and the effectiveness of normalized, combined hemodynamic parameters (CHPs) should be investigated with help of the quantitative color analysis of intraoperative images. Methods. A total of 21 unruptured cerebral aneurysms in 19 patients were analyzed. A normalized CHP was newly suggested as a weighted average of normalized wall shear stress (WSS) and normalized oscillatory shear index (OSI). Delta E from International Commission on Illumination was used to more objectively quantify color differences in intraoperative images. Results. CFD analysis results indicated that WSS and OSI were more predictive of TWAs than pressure (P<.001, P=.187, P=.970, respectively); these two parameters were selected to define the normalized CHP. The normalized CHP became more statistically significant (P<.001) as the weighting factor of normalized WSS increased and that of normalized OSI decreased. Locations with high CHP values corresponded well to those with high Delta E values (P<.001). Predicted TWAs based on the normalized CHP showed a relatively good agreement with intraoperative images (17 in 21 cases, 81.0%). Conclusion. 100% weighting on the normalized WSS produced the most statistically significant result. The normalization scheme for WSS and OSI suggested in this work was validated using quantitative color analyses, rather than subjective judgments, of intraoperative images, and it might be clinically useful for predicting TWAs of unruptured cerebral aneurysms. The normalization scheme would also be integrated into further fluid-structure interaction analysis for more reliable estimation of the risk of aneurysm rupture.