Heterogeneous Domains Research Articles

Efficient Large-Scale Nonstationary Spatial Covariance Function Estimation Using Convolutional Neural Networks

Spatial processes observed in various fields, such as climate and environmental science, often occur at large-scale and demonstrate spatial nonstationarity. However, fitting a Gaussian process with a nonstationary Matérn covariance is challenging, as it requires handling the complexity and computational demands associated with modeling the varying spatial dependencies over large and heterogeneous domains. Previous studies in the literature have tackled this challenge by employing spatial partitioning techniques to estimate the parameters that vary spatially in the covariance function. The selection of partitions is an important consideration, but it is often subjective and lacks a data-driven approach. To address this issue, in this study, we use the power of Convolutional Neural Networks (ConvNets) to derive subregions from the nonstationary data by employing a selection mechanism to identify subregions that exhibit similar behavior to stationary fields. We rely on the ExaGeoStat software for large-scale geospatial modeling to implement the nonstationary Matérn covariance for large scale exact computation of nonstationary Gaussian likelihood. We also assess the performance of the proposed method with synthetic and real datasets at large-scale. The results revealed enhanced accuracy in parameter estimations when relying on ConvNet-based partition compared to traditional user-defined approaches.

Pairpot: a database with real-time lasso-based analysis tailored for paired single-cell and spatial transcriptomics.

Paired single-cell and spatially resolved transcriptomics (SRT) data supplement each other, providing in-depth insights into biological processes and disease mechanisms. Previous SRT databases have limitations in curating sufficient single-cell and SRT pairs (SC-SP pairs) and providing real-time heuristic analysis, which hinder the effort to uncover potential biological insights. Here, we developed Pairpot (http://pairpot.bioxai.cn), a database tailored for paired single-cell and SRT data with real-time heuristic analysis. Pairpot curates 99 high-quality pairs including 1,425,656 spots from 299 datasets, and creates the association networks. It constructs the curated pairs by integrating multiple slices and establishing potential associations between single-cell and SRT data. On this basis, Pairpot adopts semi-supervised learning that enables real-time heuristic analysis for SC-SP pairs where Lasso-View refines the user-selected SRT domains within milliseconds, Pair-View infers cell proportions of spots based on user-selected cell types in real-time and Layer-View displays SRT slices using a 3D hierarchical layout. Experiments demonstrated Pairpot's efficiency in identifying heterogeneous domains and cell proportions.

A domain generalization network for imbalanced machinery fault diagnosis

Traditional models for Imbalanced Fault Diagnosis (IFD) face challenges in practical applications due to domain shifts caused by varying working conditions and machinery. Domain Generalization (DG) models provide an advantage over traditional approaches by learning class-discriminative and domain-invariant feature representations, allowing them to generalize to unseen target data. However, the scarcity of fault samples relative to healthy ones limits their application in real-world industrial scenarios. In this paper, we propose a Domain Mixed-Enhanced Domain Generalization Network (DEMDGN) that enhances IFD performance by utilizing mixup-based data augmentation and domain-based discrepancy metrics to align feature distributions across multiple heterogeneous source domains. By creating domain-invariant features, DEMDGN allows robust fault diagnosis under varying conditions. Extensive experiments on one marine machinery dataset and two bearing datasets demonstrate that the proposed method effectively addresses class imbalance and domain shift problems, achieving superior diagnostic performance.

Nanoscale Topography Dictates Residue Hydropathy in Proteins.

Proteins exhibit diverse structures, including pockets, cavities, channels, and bumps, which are crucial in determining their functions. This diversity in topography also introduces significant chemical heterogeneity, with polar and charged domains often juxtaposed with nonpolar domains in proximity. Consequently, accurately assessing the hydropathy of amino acid residues within the intricate nanoscale topology of proteins is essential. This study presents quantitative hydropathy data for 277,877 amino acid residues, computed using the Protocol for Assigning a Residue's Character on a Hydropathy (PARCH) scale. Leveraging this data set comprising 1000 structurally diverse proteins sourced from the Protein Data Bank, we examined residues situated in various nanoscale environments and analyzed hydropathy in relation to protein topography. Our findings indicate that the hydropathy of a residue is intricately linked to both its individual characteristics and the geometric features of its neighboring residues in response to water. Changes in the number and chemical identity of the neighbors, as well as the nanoscale topography surrounding a residue, are mirrored in its hydropathy profile. Our calculations reveal the intricate interplay of hydrophilic, hydroneutral, and hydrophobic residues distributed across the surface and core of proteins. Notably, we observe that protein surfaces can be ten times more hydrophilic than their cores.

Acoustic pattern of a full scale simplified ship

Underwater radiated noise from ship propellers is one of the major contributors to the increase in overall noise levels in marine environments. The development of numerical tools for the computation of noise propagation is of paramount importance for the assessment of the noise impact on the environment. Hybrid approaches, based on the decoupling of the computation of the fluid flow and the successive evolution of the noise using an acoustic analogy, are typically conducted in open-water conditions, which do not correspond to a realistic environment. The Full Acoustic Analogy, which couples a propagation model based on the solution of the acoustic wave equation to the hybrid approach, enables the propagation of the noise generated by a marine propeller in a confined heterogeneous domain. The near-field scattering characteristics of a simplified hull positioned in front of a ship's propeller are analyzed in shallow and deep water, in the presence of the free surface, and compared to the open water case. The analysis determines the directivity of the propeller noise and how this may affect how the noise propagates in the marine environment.

Adaptive wavelet-VNet for single-sample test time adaptation in medical image segmentation.

In medical image segmentation, a domain gap often exists between training and testing datasets due to different scanners or imaging protocols, which leads to performance degradation in deep learning-based segmentation models. Given the high cost of manual labeling and the need for privacy protection, it is often challenging to annotate the testing (target) domain data for model fine-tuning or to collect data from different domains to train domain generalization models. Therefore, using only unlabeled target domain data for test-time adaptation (TTA) presents a more practical but challenging solution. To improve the segmentation accuracy of deep learning-based models on unseen datasets, and especially to enhance the efficiency and stability of TTA for individual samples from heterogeneous domains. In this study, we proposed to dynamically adapt a wavelet-VNet (WaVNet) to unseen target domains with a hybrid objective function, based on each unlabeled test sample during the test time. We embedded multiscale wavelet coefficients into a V-Net encoder and adaptively adjusted the spatial and spectral features according to the input, and the model parameters were optimized by three loss functions. We integrated a shape-aware loss to focus on the foreground segmentations, a Refine loss to correct the incomplete and noisy segmentations caused by domain shifts, and an entropy loss to promote the global consistency of the segmentations. We evaluated the proposed method on multidomain liver and prostate segmentation datasets to assess its advantages over other TTA methods. For the source domain model training of the liver dataset, we used 15 3D MR image samples for training and 5 for validation. Correspondingly, for the prostate dataset, we used 22 3D MR image samples for training and 7 for validation. In the target domain, we used a single 3D MR image sample for adaptation and testing. The total number of testing samples is 60 in the liver dataset (for 3 different domains) and 116 in the prostate dataset (for 6 different domains). The proposed method showed the highest segmentation accuracy among all methods, achieving a mean(±SD) Dice coefficient (DSC) of 78.10±5.23% and a mean 95th Hausdorff distance (HD95) of 15.52±5.84mm on the liver dataset; and a mean DSC of 80.02±3.89% and a mean HD95 of 9.18±3.47mm on the prostate dataset. The DSC is 11.67% (in absolute terms) and 15.27% higher than that of the baseline (no adaptation) method, for the liver and the prostate datasets, respectively. The proposed adaptive WaVNet enhanced the image segmentation accuracy from unseen domains during the test time via unsupervised learning and multi-objective optimization. It can benefit clinical applications where data are scarce or with changing data distributions, including online adaptive radiotherapy. The code will be released at: https://github.com/sanny1226/WaVNet.

Homogenization of quasi-periodic conformal architectured materials and applications to chiral lattices

In this study, we propose to extend asymptotic periodic homogenization for non-periodic continuous microstructured media, assuming that the non-periodic geometry (called quasi-periodic) can be designed by a conformal planar transformation of a periodic parent domain architectured media with periodically disposed unit cells. Conformal transformations are shown to play a privileged role in the design of circular macroscopic heterogeneous domains tessellated with non-periodic unit cells, obtained from a periodic parent domain architectured with these unit cells. The conditions for conformal invariance are established, leading to the general form of conformal transformation in their dependencies upon the periodic coordinates. It is shown that any conformal map can be decomposed into the product of an isotropic dilatation function of the first periodic spatial position of decreasing exponential type and a rotation characterized by an angular function linear in the second periodic position. A general theory of quasi-periodic homogenization in the framework of conformal transformations is established for the first time, leading to an expression of the tensor of quasi-periodic moduli which is fully evaluated from the solution of the elasticity boundary value problem posed over the periodic unit cell. The influence of microcurvature distortion of individual unit cells on their effective properties is evaluated. Closed-form solutions are confronted to numerical examples issued from the implementation of circular periodicity in a finite element solver, showing overall a good agreement with the identified homogenized moduli.

Domain-augmented meta ensemble learning for mechanical fault diagnosis from heterogeneous source domains to unseen target domains

Existing domain generalization (DG) fault diagnosis methods primarily use adversarial training to reduce shifts between source domains and learn domain-invariant features. However, such features are difficult to learn when dealing with substantial shifts between source domains. In addition, focusing only on domain-invariant and ignoring domain-specific information is not conducive to improving model generalization. Furthermore, these methods are typically developed under an assumption that source domains share the same label space. In situations where the source domains exhibit heterogeneity with inconsistent labels, as explored in this paper, aligning the source domains becomes more difficult due to severe class mismatches. To address the above challenges, this paper proposes a method named domain-augmented meta ensemble learning. Specifically, Dirichlet CutMix is developed to compensate for missing classes in a source domain by utilizing knowledge from other source domains. Moreover, a training strategy summarized as “learn to generalize to unseen domains through collaborative ensemble” is designed to balance the learning of domain-specific features with the ability to generalize across domains. The proposed method is applied to the bearing and gearbox fault diagnosis, and experimental results demonstrate the excellent generalization of the method from heterogeneous source domains to unseen target domains.

Modeling fluid flow in heterogeneous porous media with physics-informed neural networks: Weighting strategies for the mixed pressure head-velocity formulation

Physics-informed neural networks (PINNs) are receiving increased attention in modeling flow in porous media because they can surpass purely data-driven approaches. However, in heterogeneous domains, PINNs often face convergence challenges due to discontinuities in rock properties. A promising alternative is the mixed formulation of PINNs, which utilizes pressure head and velocity fields as primary variables. This formulation introduces a multi-term loss function whose terms must be carefully balanced to ensure effective convergence during training. The main goal of this work is to identify the most suitable weighting technique to overcome convergence issues and enhance the applicability of the mixed formulation of PINNs for modeling flow in heterogeneous porous media. Thus, we implement and adapt different global and local weighting techniques and evaluate their performance through multiple test scenarios, involving stochastic and block heterogeneity. The results reveal that the most appropriate weighting strategy is the max-average technique. In the case of stochastic heterogeneity, this technique allows for improving the convergence of the training algorithm. In the case of discontinuous heterogeneity, the max-average method is the only strategy that achieved convergence, highlighting its robustness. The results also show that under high heterogeneity, using an appropriate weighting technique becomes imperative because baseline PINN failed to converge. Implementing an optimal weighting strategy can improve convergence and yield accurate solutions with fewer learnable parameters, thereby enhancing overall model performance.

Fuzzy Rule-Based Heterogeneous Domain Adaptive: A Case Study of Particle Motion in Flow Field from Simulation to Experiment

Fuzzy Rule-Based Heterogeneous Domain Adaptive: A Case Study of Particle Motion in Flow Field from Simulation to Experiment

Heterogeneous cross domain coordinated control of ASV-AUV system for maritime search and rescue

This paper is concerned with a coordinated search and rescue problem of heterogeneous ocean vehicle cluster composed of autonomous surface vehicles (ASVs) and autonomous underwater vehicles (AUVs) subject to model uncertainties and ocean disturbances. Firstly, a three dimensions (3D) cross domain kinematic controller with auto-stop function at the rescue points is designed for the coordinated ASV-AUV system by using a line-of-sight guidance law. Secondly, an event-triggered based communication mechanism is constructed to ensure that the ASV-AUV system can broadcast information at its trigger moments except to the auto-stop points. Finally, an adaptive motion control law based on a low-frequency learning fuzzy predictor is designed, which is able to filter out the high-frequency components of uncertainties and disturbances. The stability of the closed-loop system is proven by employing input-to-state stability and cascade stability analysis. Simulation results and numerical analysis show the effectiveness of the proposed 3D heterogeneous cross domain coordinated path following controller with auto-stop function for marine search and rescue.

Apatite and pyroxene as records of magmatic–hydrothermal processes and platinum-group mineral (PGM) formation in the Wengeqi mafic–ultramafic intrusion, Inner Mongolia, China: The role of oxidized, H2O-rich magmas

The Devonian Wengeqi mafic–ultramafic intrusion, situated on the northern margin of the North China Craton (NCC) — an Andean-style convergent margin during the Paleozoic — hosts many platinum-group minerals (PGMs), especially within the S-deficient, magnetite-rich clinopyroxenite zones. These PGMs occur in two distinct associations: Pt-rich PGMs (e.g., sperrylite) closely associated with primary magnetite and Pd-rich PGMs (e.g., sudburyite, kotuskite, arsenopalladinite) associated with secondary minerals (e.g., actinolite, pyrite). The mechanisms underlying the formation of PGMs in S-deficient rocks and the spatial decoupling of various PGM types remain elusive. The relationship of PGM formation and characteristics of magmas at convergent margins is also not well understood. In this contribution, we utilized apatite and clinopyroxene textures and chemistry alongside whole-rock SrNd isotopes to characterize i) the nature of the parental magma from which the Wengeqi intrusion crystallized, and ii) the magmatic–hydrothermal processes that operated to generate the Pt- and Pd-rich PGM. Based on the composition of clinopyroxene and clinopyroxene–melt partition coefficients, the parental magma of the Wengeqi intrusion is estimated to have an arc-like affinity on a primitive mantle-normalized trace-element diagram. This, together with the EMI-like SrNd isotope signature, implies that the Wengeqi magma was sourced from metasomatized SCLM beneath the NCC with an EMI-like composition. Three types of apatite (Ap1, Ap2, and Ap3) were identified within the intrusion. Ap1 and Ap2 are magmatic in origin, and appear as isolated grains, whereas Ap3 occurs as stringers or heterogeneous domains within Ap1 and Ap2, and likely resulted from hydrothermal modification of magmatic apatite. Notably, the abundant Ap2 exhibits higher VS contents and Eu/Eu* ratios than the less common Ap1, and contains S predominantly in the form of S6+, suggesting that the Wengeqi magma was relatively oxidized, with fO2 > FMQ + 1.2. Based on the chemistry of magmatic apatite, the Wengeqi magma had ∼5 wt% H2O, which, together with its high fO2, facilitated crystallization of magnetite, causing reduction of the magma by removal of Fe3+. This reduction process promoted PGM nucleation by decreasing PGE solubility in the magma, leading to the association of PGM with magnetite. Additionally, the oxidized, H2O-rich magma likely released oxidizing fluids, selectively mobilizing Pd rather than Pt, separating Pt-rich PGMs from Pd-rich PGMs. The budget of PGE in the parental magma could have been increased by i) metasomatism of the SCLM source (e.g., by carbonatitic fluids), which would have elevated the concentration of PGE in the mantle source, and ii) the high fO2 and H2O levels of the magma, which would have delayed sulfide saturation. Accordingly, mantle metasomatism and high fO2–H2O are deemed favorable for enrichment of PGE in magmas within convergent margins.

Weighted Heterogeneous Domain Adaptation for GIS Insulation Fault Diagnosis Based on SF6 Decomposition Products

Weighted Heterogeneous Domain Adaptation for GIS Insulation Fault Diagnosis Based on SF₆ Decomposition Products

Heterogeneous domain adaptation via incremental discriminative knowledge consistency

Heterogeneous domain adaptation is a challenging problem in transfer learning since samples from the source and target domains reside in different feature spaces with different feature dimensions. The key problem is how to minimize some gaps (e.g., data distribution mismatch) presented in two heterogeneous domains and produce highly discriminative representations for the target domain. In this paper, we attempt to address these challenges with the proposed incremental discriminative knowledge consistency (IDKC) method, which integrates cross-domain mapping, distribution matching, discriminative knowledge preservation, and domain-specific geometry structure consistency into a unified learning model. Specifically, we attempt to learn a domain-specific projection to project original samples into a common subspace in which the marginal distribution is well aligned and the discriminative knowledge consistency is preserved by leveraging the labeled samples from both domains. Moreover, domain-specific structure consistency is enforced to preserve the data manifold from the original space to the common feature space in each domain. Meanwhile, we further apply pseudo labeling to unlabeled target samples based on the feature correlation and retain pseudo labels with high feature correlation coefficients for the next iterative learning. Our pseudo-labeling strategy expands the number of labeled target samples in each category and thus enforces class-discriminative knowledge consistency to produce more discriminative feature representations for the target domain. Extensive experiments on several standard benchmarks for object recognition, cross-language text classification, and digit classification tasks verify the effectiveness of our method.

Unsupervised heterogeneous domain adaptation for EEG classification

Objective. Domain adaptation has been recognized as a potent solution to the challenge of limited training data for electroencephalography (EEG) classification tasks. Existing studies primarily focus on homogeneous environments, however, the heterogeneous properties of EEG data arising from device diversity cannot be overlooked. This motivates the development of heterogeneous domain adaptation methods that can fully exploit the knowledge from an auxiliary heterogeneous domain for EEG classification. Approach. In this article, we propose a novel model named informative representation fusion (IRF) to tackle the problem of unsupervised heterogeneous domain adaptation in the context of EEG data. In IRF, we consider different perspectives of data, i.e. independent identically distributed (iid) and non-iid, to learn different representations. Specifically, from the non-iid perspective, IRF models high-order correlations among data by hypergraphs and develops hypergraph encoders to obtain data representations of each domain. From the non-iid perspective, by applying multi-layer perceptron networks to the source and target domain data, we achieve another type of representation for both domains. Subsequently, an attention mechanism is used to fuse these two types of representations to yield informative features. To learn transferable representations, the maximum mean discrepancy is utilized to align the distributions of the source and target domains based on the fused features. Main results. Experimental results on several real-world datasets demonstrate the effectiveness of the proposed model. Significance. This article handles an EEG classification situation where the source and target EEG data lie in different spaces, and what’s more, under an unsupervised learning setting. This situation is practical in the real world but barely studied in the literature. The proposed model achieves high classification accuracy, and this study is important for the commercial applications of EEG-based BCIs.

L p regularity for elliptic transmission problems with two spectral parameters

ABSTRACT We study a family of elliptic transmission problems on heterogeneous unbounded domains posed in UMD spaces. In order to determine the strict solution, this work investigates the effect and relationships between two spectral parameters; the first parameter is in the differential equation, and the second is in the permeability condition on the interface. Necessary and sufficient conditions on the data are established to prove the maximal regularity of the solution by using semigroup theory and the properties of fractional powers of linear operators

Noise-based correction for electrical impedance tomography

Objective. Noisy measurements frequently cause noisy and inaccurate images in impedance imaging. No post-processing technique exists to calculate the propagation of measurement noise and use this to suppress noise in the image. The objectives of this work were (1) to develop a post-processing method for noise-based correction (NBC) in impedance tomography, (2) to test whether NBC improves image quality in electrical impedance tomography (EIT), (3) to determine whether it is preferable to use correlated or uncorrelated noise for NBC, (4) to test whether NBC works with in vivo data and (5) to test whether NBC is stable across model and perturbation geometries. Approach. EIT was performed in silico in a 2D homogeneous circular domain and an anatomically realistic, heterogeneous 3D human head domain for four perturbations and 25 noise levels in each case. This was validated by performing EIT for four perturbations in a circular, saline tank in 2D as well as a human head-shaped saline tank with a realistic skull-like layer in 3D. Images were assessed on the error in the weighted spatial variance (WSV) with respect to the true, target image. The effect of NBC was also tested for in vivo EIT data of lung ventilation in a human thorax and cortical activity in a rat brain. Main results. On visual inspection, NBC maintained or increased image quality for all perturbations and noise levels in 2D and 3D, both experimentally and in silico. Analysis of the WSV showed that NBC significantly improved the WSV in nearly all cases. When the WSV was inferior with NBC, this was either visually imperceptible or a transformation between noisy reconstructions. For in vivo data, NBC improved image quality in all cases and preserved the expected shape of the reconstructed perturbation. Significance. In practice, uncorrelated NBC performed better than correlated NBC and is recommended as a general-use post-processing technique in EIT.

BAYESIAN NESTED LATENT CLASS MODELS FOR CAUSE-OF-DEATH ASSIGNMENT USING VERBAL AUTOPSIES ACROSS MULTIPLE DOMAINS.

Understanding cause-specific mortality rates is crucial for monitoring population health and designing public health interventions. Worldwide, two-thirds of deaths do not have a cause assigned. Verbal autopsy (VA) is a well-established tool to collect information describing deaths outside of hospitals by conducting surveys to caregivers of a deceased person. It is routinely implemented in many low- and middle-income countries. Statistical algorithms to assign cause of death using VAs are typically vulnerable to the distribution shift between the data used to train the model and the target population. This presents a major challenge for analyzing VAs, as labeled data are usually unavailable in the target population. This article proposes a latent class model framework for VA data (LCVA) that jointly models VAs collected over multiple heterogeneous domains, assigns causes of death for out-of-domain observations and estimates cause-specific mortality fractions for a new domain. We introduce a parsimonious representation of the joint distribution of the collected symptoms using nested latent class models and develop a computationally efficient algorithm for posterior inference. We demonstrate that LCVA outperforms existing methods in predictive performance and scalability. Supplementary Material and reproducible analysis codes are available online. The R package LCVA implementing the method is available on GitHub (https://github.com/richardli/LCVA).

Fog-based deep learning framework for real-time pandemic screening in smart cities from multi-site tomographies

The quick proliferation of pandemic diseases has been imposing many concerns on the international health infrastructure. To combat pandemic diseases in smart cities, Artificial Intelligence of Things (AIoT) technology, based on the integration of artificial intelligence (AI) with the Internet of Things (IoT), is commonly used to promote efficient control and diagnosis during the outbreak, thereby minimizing possible losses. However, the presence of multi-source institutional data remains one of the major challenges hindering the practical usage of AIoT solutions for pandemic disease diagnosis. This paper presents a novel framework that utilizes multi-site data fusion to boost the accurateness of pandemic disease diagnosis. In particular, we focus on a case study of COVID-19 lesion segmentation, a crucial task for understanding disease progression and optimizing treatment strategies. In this study, we propose a novel multi-decoder segmentation network for efficient segmentation of infections from cross-domain CT scans in smart cities. The multi-decoder segmentation network leverages data from heterogeneous domains and utilizes strong learning representations to accurately segment infections. Performance evaluation of the multi-decoder segmentation network was conducted on three publicly accessible datasets, demonstrating robust results with an average dice score of 89.9% and an average surface dice of 86.87%. To address scalability and latency issues associated with centralized cloud systems, fog computing (FC) emerges as a viable solution. FC brings resources closer to the operator, offering low latency and energy-efficient data management and processing. In this context, we propose a unique FC technique called PANDFOG to deploy the multi-decoder segmentation network on edge nodes for practical and clinical applications of automated COVID-19 pneumonia analysis. The results of this study highlight the efficacy of the multi-decoder segmentation network in accurately segmenting infections from cross-domain CT scans. Moreover, the proposed PANDFOG system demonstrates the practical deployment of the multi-decoder segmentation network on edge nodes, providing real-time access to COVID-19 segmentation findings for improved patient monitoring and clinical decision-making.

Cooperative roles of twinning, transformation and strain partitioning on the excellent cryogenic strength-toughness synergy of dual-heterogeneous metastable high entropy alloys

In a metastable Fe50Mn30Co10Cr10 high entropy alloy (HEA), a dual-heterogeneous structure, encompassing both grain size and dislocation heterogeneity, has been constructed to obtain excellent cryogenic mechanical performance by varying the annealing temperature (630 and 740 °C) following severe deformation. The HS-630 sample, consisting of recovered coarse-grains (CG), recrystallized fine-grains (FGs) and un-recrystallized ultrafine-grains (UFGs), possesses higher yield strength (YS) of ∼965 MPa, ultimate tensile strength (UTS) of ∼1680 MPa and uniform elongation (UEL) of ∼45 % when compared to the fully-recrystallized HS-740 sample at liquid nitrogen temperature (LNT). The difference in YS is primarily caused by the grain boundary and dislocation strengthening of UFGs, whereas the higher ductility of the HS-630 sample is associated with the hierarchical activation of transformation- and twinning-induced plasticity (TRIP and TWIP) effects, and compatible strain partitioning between heterogeneous domains. The dual-heterogeneity results in different critical stresses of ε-phase nucleation in heterogeneous domains, and promotes sustainable TRIP effect in CGs and the emergence of TWIP effect in UFGs with strain progressing. Moreover, the strain of the ductile transformed ε-phase is mainly accommodated by non-basal slip, which contributes to the strain hardening in the later deformation stage. Furthermore, the higher cryogenic impact toughness of the HS-630 sample can be attributed to the synergistic effects of the fracture mode, the transformation characteristics, and the heterogeneous strain partitioning.