Unseen Distribution Research Articles

Improving deep learning-based automatic cranial defect reconstruction by heavy data augmentation: From image registration to latent diffusion models

Modeling and manufacturing of personalized cranial implants are important research areas that may decrease the waiting time for patients suffering from cranial damage. The modeling of personalized implants may be partially automated by the use of deep learning-based methods. However, this task suffers from difficulties with generalizability into data from previously unseen distributions that make it difficult to use the research outcomes in real clinical settings. Due to difficulties with acquiring ground-truth annotations, different techniques to improve the heterogeneity of datasets used for training the deep networks have to be considered and introduced. In this work, we present a large-scale study of several augmentation techniques, varying from classical geometric transformations, image registration, variational autoencoders, and generative adversarial networks, to the most recent advances in latent diffusion models. We show that the use of heavy data augmentation significantly increases both the quantitative and qualitative outcomes, resulting in an average Dice Score above 0.94 for the SkullBreak and above 0.96 for the SkullFix datasets. The results show that latent diffusion models combined with vector quantized variational autoencoder outperform other generative augmentation strategies. Moreover, we show that the synthetically augmented network successfully reconstructs real clinical defects, without the need to acquire costly and time-consuming annotations. The findings of the work will lead to easier, faster, and less expensive modeling of personalized cranial implants. This is beneficial to numerous people suffering from cranial injuries. The work constitutes a considerable contribution to the field of artificial intelligence in the automatic modeling of personalized cranial implants.

Enhancing time series data classification for structural damage detection through out-of-distribution representation learning

The analysis and classification of time series data, notably within Structural Health Monitoring (SHM) systems, face profound challenges due to the nonstationary nature of the data and changes in distribution over time. Additionally, the complexity of the SHM systems' time series datasets also relates to the presence of elusive latent distributions, making it challenging to pinpoint the dynamics of data distribution shifts. Therefore, this paper uses a Deep Learning (DL) model to address dynamic distribution changes and target elusive latent distributions in the data of SHM systems by adopting an out-of-distribution (OOD) representation learning method. The core idea is that by using adversarial training and domain-invariant feature learning, the used model enhances the analysis of time series data by dividing the data into distinct latent sub-domains, aiming to maximize the variance in distribution between each segment, facilitating a more unified analysis across different data segments. The effectiveness is demonstrated through testing on extensive datasets, including both real-world bridge data and laboratory data, showing a significant improvement over traditional DL models, especially in its ability to generalize to new, unseen distributions.

IOP-FL: Inside-Outside Personalization for Federated Medical Image Segmentation.

Federated learning (FL) allows multiple medical institutions to collaboratively learn a global model without centralizing client data. It is difficult, if possible at all, for such a global model to commonly achieve optimal performance for each individual client, due to the heterogeneity of medical images from various scanners and patient demographics. This problem becomes even more significant when deploying the global model to unseen clients outside the FL with unseen distributions not presented during federated training. To optimize the prediction accuracy of each individual client for medical imaging tasks, we propose a novel unified framework for both Inside and Outside model Personalization in FL (IOP-FL). Our inside personalization uses a lightweight gradient-based approach that exploits the local adapted model for each client, by accumulating both the global gradients for common knowledge and the local gradients for client-specific optimization. Moreover, and importantly, the obtained local personalized models and the global model can form a diverse and informative routing space to personalize an adapted model for outside FL clients. Hence, we design a new test-time routing scheme using the consistency loss with a shape constraint to dynamically incorporate the models, given the distribution information conveyed by the test data. Our extensive experimental results on two medical image segmentation tasks present significant improvements over SOTA methods on both inside and outside personalization, demonstrating the potential of our IOP-FL scheme for clinical practice. Code is available at https://github.com/med-air/IOP-FL.

Cation Disorder and Large Tetragonal Supercell Ordering in the Li-Rich Argyrodite Li7Zn0.5SiS6.

A tetragonal argyrodite with >7 mobile cations, Li7Zn0.5SiS6, is experimentally realized for the first time through solid state synthesis and exploration of the Li–Zn–Si–S phase diagram. The crystal structure of Li7Zn0.5SiS6 was solved ab initio from high-resolution X-ray and neutron powder diffraction data and supported by solid-state NMR. Li7Zn0.5SiS6 adopts a tetragonal I4 structure at room temperature with ordered Li and Zn positions and undergoes a transition above 411.1 K to a higher symmetry disordered F43m structure more typical of Li-containing argyrodites. Simultaneous occupation of four types of Li site (T5, T5a, T2, T4) at high temperature and five types of Li site (T5, T2, T4, T1, and a new trigonal planar T2a position) at room temperature is observed. This combination of sites forms interconnected Li pathways driven by the incorporation of Zn2+ into the Li sublattice and enables a range of possible jump processes. Zn2+ occupies the 48h T5 site in the high-temperature F43m structure, and a unique ordering pattern emerges in which only a subset of these T5 sites are occupied at room temperature in I4 Li7Zn0.5SiS6. The ionic conductivity, examined via AC impedance spectroscopy and VT-NMR, is 1.0(2) × 10–7 S cm–1 at room temperature and 4.3(4) × 10–4 S cm–1 at 503 K. The transition between the ordered I4 and disordered F43m structures is associated with a dramatic decrease in activation energy to 0.34(1) eV above 411 K. The incorporation of a small amount of Zn2+ exercises dramatic control of Li order in Li7Zn0.5SiS6 yielding a previously unseen distribution of Li sites, expanding our understanding of structure–property relationships in argyrodite materials.

One-Class Knowledge Distillation for Face Presentation Attack Detection

Face presentation attack detection (PAD) has been extensively studied by research communities to enhance the security of face recognition systems. Although existing methods have achieved good performance on testing data with similar distribution as the training data, their performance degrades severely in application scenarios with data of unseen distributions. In situations where the training and testing data are drawn from different domains, a typical approach is to apply domain adaptation techniques to improve face PAD performance with the help of target domain data. However, it has always been a non-trivial challenge to collect sufficient data samples in the target domain, especially for attack samples. This paper introduces a teacher-student framework to improve the cross-domain performance of face PAD with one-class domain adaptation. In addition to the source domain data, the framework utilizes only a few genuine face samples of the target domain. Under this framework, a teacher network is trained with source domain samples to provide discriminative feature representations for face PAD. Student networks are trained to mimic the teacher network and learn similar representations for genuine face samples of the target domain. In the test phase, the similarity score between the representations of the teacher and student networks is used to distinguish attacks from genuine ones. To evaluate the proposed framework under one-class domain adaptation settings, we devised two new protocols and conducted extensive experiments. The experimental results show that our method outperforms baselines under one-class domain adaptation settings and even state-of-the-art methods with unsupervised domain adaptation.

Generalizing to Unseen Domains: A Survey on Domain Generalization

Machine learning systems generally assume that the training and testing distributions are the same. To this end, a key requirement is to develop models that can generalize to unseen distributions. Domain generalization (DG), i.e., out-of-distribution generalization, has attracted increasing interests in recent years. Domain generalization deals with a challenging setting where one or several different but related domain(s) are given, and the goal is to learn a model that can generalize to an unseen test domain. Great progress has been made in the area of domain generalization for years. This paper presents the first review of recent advances in this area. First, we provide a formal definition of domain generalization and discuss several related fields. We then thoroughly review the theories related to domain generalization and carefully analyze the theory behind generalization. We categorize recent algorithms into three classes: data manipulation, representation learning, and learning strategy, and present several popular algorithms in detail for each category. Third, we introduce the commonly used datasets, applications, and our open-sourced codebase for fair evaluation. Finally, we summarize existing literature and present some potential research topics for the future.

ROBUST WHITE MATTER HYPERINTENSITY SEGMENTATION ON UNSEEN DOMAIN.

Typical machine learning frameworks heavily rely on an underlying assumption that training and test data follow the same distribution. In medical imaging which increasingly begun acquiring datasets from multiple sites or scanners, this identical distribution assumption often fails to hold due to systematic variability induced by site or scanner dependent factors. Therefore, we cannot simply expect a model trained on a given dataset to consistently work well, or generalize, on a dataset from another distribution. In this work, we address this problem, investigating the application of machine learning models to unseen medical imaging data. Specifically, we consider the challenging case of Domain Generalization (DG) where we train a model without any knowledge about the testing distribution. That is, we train on samples from a set of distributions (sources) and test on samples from a new, unseen distribution (target). We focus on the task of white matter hyperintensity (WMH) prediction using the multi-site WMH Segmentation Challenge dataset and our local in-house dataset. We identify how two mechanically distinct DG approaches, namely domain adversarial learning and mix-up, have theoretical synergy. Then, we show drastic improvements of WMH prediction on an unseen target domain.

Random Access Memories: A New Paradigm for Target Detection in High Resolution Aerial Remote Sensing Images.

We propose a new paradigm for target detection in high resolution aerial remote sensing images under small target priors. Previous remote sensing target detection methods frame the detection as learning of detection model + inference of class-label and bounding-box coordinates. Instead, we formulate it from a Bayesian view that at inference stage, the detection model is adaptively updated to maximize its posterior that is determined by both training and observation. We call this paradigm "random access memories (RAM)." In this paradigm, "Memories" can be interpreted as any model distribution learned from training data and "random access" means accessing memories and randomly adjusting the model at detection phase to obtain better adaptivity to any unseen distribution of test data. By leveraging some latest detection techniques e.g., deep Convolutional Neural Networks and multi-scale anchors, experimental results on a public remote sensing target detection data set show our method outperforms several other state of the art methods. We also introduce a new data set "LEarning, VIsion and Remote sensing laboratory (LEVIR)", which is one order of magnitude larger than other data sets of this field. LEVIR consists of a large set of Google Earth images, with over 22 k images and 10 k independently labeled targets. RAM gives noticeable upgrade of accuracy (an mean average precision improvement of 1% ~ 4%) of our baseline detectors with acceptable computational overhead.

Computing Human-Understandable Strategies: Deducing Fundamental Rules of Poker Strategy

Algorithms for equilibrium computation generally make no attempt to ensure that the computed strategies are understandable by humans. For instance the strategies for the strongest poker agents are represented as massive binary files. In many situations, we would like to compute strategies that can actually be implemented by humans, who may have computational limitations and may only be able to remember a small number of features or components of the strategies that have been computed. For example, a human poker player or military leader may not have access to large precomputed tables when making real-time strategic decisions. We study poker games where private information distributions can be arbitrary (i.e., players are dealt cards from different distributions, which depicts the phenomenon in large real poker games where at some points in the hand players have different distribution of hand strength by applying Bayes’ rule given the history of play in the hand thus far). We create a large training set of game instances and solutions, by randomly selecting the information probabilities, and present algorithms that learn from the training instances to perform well in games with unseen distributions. We are able to conclude several new fundamental rules about poker strategy that can be easily implemented by humans.