Search

Machine learning models have demonstrated vulnerability to adversarial attacks, more specifically misclassification of adversarial examples. In this paper, we propose a one-off and attack-agnostic Feature Manipulation (FM)-Defense to detect and purify adversarial examples in an interpretable and efficient manner. The intuition is that the classification result of a normal image is generally resistant to non-significant intrinsic feature changes, e.g., varying thickness of handwritten digits. In contrast, adversarial examples are sensitive to such changes since the perturbation lacks transferability. To enable manipulation of features, a Combo-variational autoencoder is applied to learn disentangled latent codes that reveal semantic features. The resistance to classification change over the morphs, derived by varying and reconstructing latent codes, is used to detect suspicious inputs. Further, Combo-VAE is enhanced to purify the adversarial examples with good quality by considering class-shared and class-unique features. We empirically demonstrate the effectiveness of detection and the quality of purified instances. Our experiments on three datasets show that FM-Defense can detect nearly $100\%$ of adversarial examples produced by different state-of-the-art adversarial attacks. It achieves more than $99\%$ overall purification accuracy on the suspicious instances that close the manifold of clean examples.

Neural topic models (NTM) have shown their success in topic modelling with awide range of applications in text analysis. NTMs based on generative modelstend to have good refactoring capabilities in achieving good document represen-tations, rather than providing a topic representation that has an understandableassociation with documents and words. To bridge this gap, we have proposeda neural topic model that allows simultaneously learning topics from text dataas well as capturing the structural manifold in the original data space of doc-uments. Comprehensive experiments have shown that the proposed model canimprove learned topics in terms of intrinsic metrics (purity and coherence)and extrinsic downstream tasks (clustering, classification). Beyond that, themodel can provide interpretable low dimensional presentations of documents

A polysarcosine-modified NIR-II small-molecule-based polymer (TQF–PSar) was facilely prepared for efficient breast cancer pulmonary metastasis imaging.

Developing conjugated small molecules (CSM) with intense NIR-II (1000-1700 nm) absorption for phototheranostic is highly desirable but remains a tremendous challenge due to a lack of reliable design guidelines. This study reports a high-performance NIR-II CSM for phototheranostic by tailoring molecular planarity. A series of CSM show bathochromic absorption extended to the NIR-II region upon the increasing thiophene number, but an excessive number of thiophene results in decreased NIR-IIa (1300-1400 nm) brightness and photothermal effects. Further introduction of terminal nonconjugated alkyl chain can enhance NIR-II absorption coefficient, NIR-IIa brightness, and photothermal effects. Mechanism studies ascribe this overall enhancement to molecular planarity stemming from the collective contribution of donor/side-chain engineering. This finding directs the design of NIR-II CSM by rational manipulating molecular planarity to perform 1064 nm mediated phototheranostic at high efficiency.

Near-infrared II (NIR-II, 1000-1700 nm) fluorescent imaging (FI) has been reported to achieve optical images with higher resolution and deeper penetration. Among the organic NIR-II small molecules, donor-acceptor-donor (D-A-D) type fluorescent agents have shown superior photophysical and biocompatible properties for FI applications but have ongoing limitations, such as the difficulty in further modifying them with drug-carrying functional groups or prodrugs. In this work, three D-A-D type NIR-II fluorophores with electron acceptors of 4,8-bis(5-bromo-4-(2-octyldodecyl)thiophen-2-yl)-1H,3H-benzo[1,2-c:4,5-c']bis([1,2,5]thiadiazole) (BBT), 6,7-bis(4-(hexyloxy)phenyl)-4,9-di(thiophen-2-yl)-[1,2,5]thiadiazolo[3,4-g]quinoxaline (TTQ) and 4,6-bis(5-bromo-2-thienyl)thieno[3,4-c][1,2,5]thiadiazole (TTDT) have been successfully prepared. Their optical and imaging properties and stability were investigated via theoretical and experimental studies. The results demonstrated that TTDT-SF exhibited good NIR-II imaging ability. Importantly, TTDT-SF showed outstanding stability in an alkaline and redox environment. Subsequently, a stable atom transfer radical polymerization (ATRP) initiator, based on TTDT and its derivative water-soluble fluorescent polymer TTDT-TF-POEGMA, synthesized through ATRP, was successfully fabricated. It was demonstrated that TTDT-TF-POEGMA exhibited excellent fluorescence ability, great water solubility, effective light stability and great potential in tumor FI and image-guided surgery. In a word, this work has developed a new stable initiator with NIR-II fluorescent properties, which provides a platform for the development of water-soluble and multifunctional NIR-II fluorescent polymers for a broad range of applications.

The luminescence of traditional phosphorescence-based hypoxia probes is limited to the visible and first near-infrared wavelength regions (1000 nm), which has defects of higher light scattering and lower penetration depth in contrast with the second near-infrared wavelength window (NIR-II, 1000-1700 nm) for optical bioimaging. Herein, 5,15-bis(2,6-bis(dodecyloxy)phenyl)-porphyrin platinum(II) (PpyPt) with J-aggregation induced NIR-II phosphorescence is reported. J-aggregates of PpyPt are confirmed by the X-ray diffraction data in the crystalline state. Moreover, the emission and excitation spectra of PpyPt in the solid states reveal NIR-II luminescence feature of PpyPt in J-aggregates. More importantly, by preparation of water-soluble PpyPt nanoparticles (PpyPt NPs

Optical imaging-guided chemo-photothermal combination therapy of cancers has attracted considerable attention, because of its capacity for personalized, precision treatment and its synergistic chem...

The common existence of hypoxia within the tumor microenvironment severely restricts the efficacy of photodynamic therapy (PDT), which is attributed to the fact that the PDT process is strongly oxygen (O2) dependent. Here, a multifunctional composite (named CPCG), which combines polyethylene glycol (PEG) functionalized cerium oxide nanoparticles (CeO2) with photosensitizer chlorin e6 (Ce6) and glucose oxidase (GOx), is reported for generating O2 within the tumor microenvironment by the dual-path hydrogen peroxide (H2O2)-modulated ways to ameliorate hypoxia, thereby enhancing the PDT efficiency. This process is realized based on the dual enzyme-like activity of CeO2. The first modulated way is to transform the superoxide anion (O2˙-) into H2O2 by the superoxide dismutase-like activity of CeO2. The second modulated way is to decompose glucose into H2O2 through the catalysis of GOx. Subsequently, H2O2 generated from the above dual modulated ways can further produce O2via the catalase-like activity of CeO2. Additionally, the depletion of glucose could impede the nutrient supply to obtain starvation therapy. Both in vitro and in vivo experiments indicate that the CPCG composite could enhance the efficacy of photodynamic/starvation synergistic therapy. Therefore, this strategy offers great potential to modulate the O2 level in the tumor microenvironment for better therapeutic outcomes, and can act as a promising candidate in photodynamic/starvation synergistic therapy.

BackgroundThe association between hypernatremia and delirium after cardiac surgery has rarely been investigated. This study aimed to determine whether hypernatremia increases the risk of delirium after exposure.Materials and MethodsFrom April 2016 to June 2021, 7,831 consecutive patients receiving cardiac surgery were screened for potential enrollment. The primary outcome was postoperative delirium (POD). For the respective case of delirium, 10 controls were matched according to the index date within the nested case-control design. Hypernatremia exposure was defined as serum sodium > 145 mmol/L within 7 days before the index date. A generalized estimation equation was performed to assess excess risks for POD associated with hypernatremia, adjusted for demographics and clinical variables.ResultsAbout 7,277 patients were included in the final analyses. About 669 (9.2%) patients with POD were assigned to the case group, and 6,690 controls were identified from the whole population. About 66.5% of the cases and 36.3% of the controls had hypernatremia exposure. After being adjusted to certain well-recognized confounding factors, hypernatremia showed a significant correlation with increased risk of delirium after cardiac surgery (adjusted OR, 1.73; 95% CI, 1.41~2.12). An e-value analysis suggested the robustness to unmeasured confounding.ConclusionsHypernatremia was associated with an increased risk of delirium after cardiac surgery. This finding could have implications for risk stratification, early detection, and management of delirium in patients receiving cardiac surgery.

Single-component nanoplatforms combined with the second near-infrared optical window (NIR-II, 1000–1700 nm) fluorescence imaging (FI) and NIR-II photothermal therapy (PTT) have received increasing attention owing to their capacity for precise diagnosis, noninvasive therapy, and real-time monitoring of the therapeutic effects. However, most of the PTT treatments are performed in the NIR-I window (700–900 nm). Moreover, the design and development of conjugated polymers (P1, P2, and P3) with both bright NIR-II fluorescence and superior NIR-II photothermal effect remained a huge challenge. Therefore, three double-acceptor conjugated polymers were designed and developed by adjusting the molar ratios of two acceptors, TTQ and DPP. Subsequently, their corresponding nanoparticles were fabricated, and finally, nanoparticles based on the conjugated polymer P1 (P1 NPs) with both high NIR-II fluorescence intensity and superior NIR-II photothermal efficiency were selected and applied for NIR-II FI and NIR-II PTT. Importantly, the experiments of in vivo NIR-II FI and NIR-II PTT demonstrated that P1 NPs exhibited not only high accumulation in the tumour sites and high sign-to-background ratio (SBR) of vascular imaging, but also superior NIR-II PTT efficiency for tumour treatment.

In many walks of life, data recording/capture and its subsequent use can be of benefit to society and businesses alike. In many cases, users are fully willing to share data about themselves as part of getting access to some perceived benefit through this sharing, e.g. using the geo-location of their mobile device to obtain local information for route planning. However in other cases, information is captured on individuals where they have little choice. This might be CCTV images of pedestrians walking down the streets, or as presented in this paper: WiFi data from individuals at a given educational organisation. Individuals at the University of Melbourne often depend on having access to University wireless to undertake their courses/degrees or their jobs respectively. The terms and conditions that they agree to when they sign up to WiFi access include a description of how the data can be used by the University, e.g. to assist in understanding the use of the WiFi network and/or the physical University campus for space/infrastructure management. The University also has stringent privacy policies that prevent the tracking of individuals. In this context, the need to provide services that can use WiFi data for business purposes, but protect the information so that individuals cannot be re-identified is paramount. There are also many researchers that wish to access safe versions of this data for research purposes, e.g. way-finding. This paper explores case studies exploring this data based on WiFi data analytics. We show how (consenting) individuals can be re-identified with minimal external and seemingly innocuous (i.e. non-identity-revealing) data.

Anomaly detection aims to recognize samples with anomalous and unusual patterns with respect to a set of normal data. This is significant for numerous domain applications, such as industrial inspection, medical imaging, and security enforcement. There are two key research challenges associated with existing anomaly detection approaches: (1) many approaches perform well on low-dimensional problems however the performance on high-dimensional instances, such as images, is limited; (2) many approaches often rely on traditional supervised approaches and manual engineering of features, while the topic has not been fully explored yet using modern deep learning approaches, even when the well-label samples are limited. In this paper, we propose a One-for-all Image Anomaly Detection system (OIAD) based on disentangled learning using only clean samples. Our key insight is that the impact of small perturbation on the latent representation can be bounded for normal samples while anomaly images are usually outside such bounded intervals, referred to as structure consistency. We implement this idea and evaluate its performance for anomaly detection. Our experiments with three datasets show that OIAD can detect over $90\%$ of anomalies while maintaining a low false alarm rate. It can also detect suspicious samples from samples labeled as clean, coincided with what humans would deem unusual., arXiv admin note: text overlap with arXiv:1802.05983, arXiv:1909.02755, arXiv:1804.03599 by other authors

In this paper, we propose a practical privacy-preserving generative model for data sanitization and sharing, called Sanitizer-Variational Autoencoder (SVAE). We assume that the data consists of identification-relevant and irrelevant components. A variational autoencoder (VAE) based sanitization model is proposed to strip the identification-relevant features and only retain identification-irrelevant components in a privacy-preserving manner. The sanitization allows for task-relevant discrimination (utility) but minimizes the personal identification information leakage (privacy). We conduct extensive empirical evaluations on the real-world face, biometric signal and speech datasets, and validate the effectiveness of our proposed SVAE, as well as the robustness against the membership inference attack.

Transfer learning provides an effective solution for feasibly and fast customize accurate \textit{Student} models, by transferring the learned knowledge of pre-trained \textit{Teacher} models over large datasets via fine-tuning. Many pre-trained Teacher models used in transfer learning are publicly available and maintained by public platforms, increasing their vulnerability to backdoor attacks. In this paper, we demonstrate a backdoor threat to transfer learning tasks on both image and time-series data leveraging the knowledge of publicly accessible Teacher models, aimed at defeating three commonly-adopted defenses: \textit{pruning-based}, \textit{retraining-based} and \textit{input pre-processing-based defenses}. Specifically, (A) ranking-based selection mechanism to speed up the backdoor trigger generation and perturbation process while defeating \textit{pruning-based} and/or \textit{retraining-based defenses}. (B) autoencoder-powered trigger generation is proposed to produce a robust trigger that can defeat the \textit{input pre-processing-based defense}, while guaranteeing that selected neuron(s) can be significantly activated. (C) defense-aware retraining to generate the manipulated model using reverse-engineered model inputs. We launch effective misclassification attacks on Student models over real-world images, brain Magnetic Resonance Imaging (MRI) data and Electrocardiography (ECG) learning systems. The experiments reveal that our enhanced attack can maintain the $98.4\%$ and $97.2\%$ classification accuracy as the genuine model on clean image and time series inputs respectively while improving $27.9\%-100\%$ and $27.1\%-56.1\%$ attack success rate on trojaned image and time series inputs respectively in the presence of pruning-based and/or retraining-based defenses.

In this paper, we provide a practical privacy-preserving generative model for time series data augmentation and sharing, called PART-GAN. Our model enables the local data curator to provide a freely accessible public generative model derived from original data for cloud storage. Compared with existing approaches, PART-GAN has three key advantages: It enables the generation of an unlimited amount of synthetic time series data under the guidance of a given classification label and addresses the incomplete and temporal irregularity issues. It provides a robust privacy guarantee that satisfies differential privacy to time series data augmentation and sharing. It addresses the trade-offs between utility and privacy by applying optimization strategies. We evaluate and report the utility and efficacy of PART-GAN through extensive empirical evaluations of real-world health/medical datasets. Even at a higher level of privacy protection, our method outperforms GAN with ordinary perturbation. It achieves similar performance with GAN without perturbation in terms of inception score, machine learning score similarity, and distance-based evaluations.

Generally, suppose a wave is a linear combination of multiple basis(Not necessarily a sine or cosine waves, it could also be a wavelet, etc.), different types of waves may be similar on some basis, but vary greatly on a certain basis. To address this problem, we introduce a PCCA-based feature extraction method that extends canonical correlation analysis (CCA). The PCCA-based method can train efficient classifiers to rely on only a few samples for periodic signals with support for removing noisy signals. As a demonstration, an efficient system is implemented for the classification of electrocardiogram (ECG) signals by PCCA. The performance is measured using several normal and abnormal ECG signals from the real-world database. These are compared with three commonly-adopted feature extraction techniques using five classes classification tasks related to ECG heartbeats. The AUC(Area under the ROC curve) of the PCCA-based feature extraction technique with two-digits size train dataset for four ECG type-pairs we compared were 0.8805, 0.957, 0.8968 and 1.00 respectively. The experimental results demonstrate that the proposed feature extraction techniques achieve better performance compared to other features extraction techniques with small amount of well-labeled data.

Fluorescence imaging (FI) in the second near-infrared region (NIR-II, 1000–1700 nm) has attracted great attention for brain tumor imaging due to its deep penetration and high resolution. However, traditional NIR-II organic fluorescent nanoparticles (NPs) are usually hindered by uncontrolled large size (~30–100 nm), marked aggregation-caused quenching (ACQ) effect, and limited blood circulation (~1–3 h), which have great impact on efficient NIR-II FI of deep brain tumors. Herein, starlike polymer brush-based ultrasmall TQFP-10 NPs, with bright NIR-II fluorescence, prolonged blood circulation, and enhanced tumor accumulation, are facilely prepared for efficient orthotopic glioblastoma (GBM) imaging. Compared with traditional method prepared NPs (physically coated TQF@NPs and PEG modified TQF-PEG5K NPs), the ultrasmall (~8 nm) TQFP-10 NPs display a higher NIR-II fluorescence QY (1.9%), which is 2.1- and 3.8-fold higher than TQF@NPs (0.9%) and TQF-PEG5K NPs (0.5%), respectively. In addition, TQFP-10 NPs present a 10.6-fold higher blood circulation half-life (t1/2 = 8.5 h) than that of TQF-PEG5K NPs. Consequently, TQFP-10 NPs exhibit 4.2- and 33-fold higher maximal tumor to normal tissue ratio in subcutaneous and in situ NIR-II FI of GBM, respectively, than TQF@NPs and TQF-PEG5K NPs, attractively realizing GBM imaging. This work provides a general strategy for constructing ultrasmall NIR-II fluorescent NPs with simultaneously improved NIR-II fluorescence and blood circulation for efficient brain tumor imaging.

The advancement of deep models poses great challenges to real-world deployment because of the limited computational ability and storage space on edge devices. To solve this problem, existing works have made progress to compress deep models by pruning or quantization. However, most existing methods rely on a large amount of training data and a pre-trained model in the same domain. When only limited in-domain training data is available, these methods fail to perform well. This prompts the idea of transferring knowledge from a resource-rich source domain to a target domain with limited data to perform model compression. In this paper, we propose a method to perform cross-domain pruning by cooperatively training in both domains: taking advantage of data and a pre-trained model from the source domain to assist pruning in the target domain. Specifically, source and target pruned models are trained simultaneously and interactively, with source information transferred through the construction of a cooperative pruning mask. Our method significantly improves pruning quality in the target domain, and shed light to model compression in the cross-domain setting.

αII-spectrin breakdown products are regarded as potential biomarkers for traumatic brain injury (TBI). The aim of the present study was to further evaluate these biomarkers by assessing their clinical utility in predicting the severity of injury and clinical outcome of patients with TBI. Eligible patients with acute TBI (n=17), defined by a Glasgow Coma Scale (GCS) score of ≤8, were enrolled. Ventricular cerebrospinal fluid (CSF) was sampled from each patient at 24, 72 and 120 h following TBI. An immunoblot assay was used to determine the concentrations of SBDPs in the CSF samples. The concentrations of SBDPs combined with the GCS score at 24 h after injury and the Glasgow Outcome Score (GOS) at 30 days after injury were compared and analyzed. The levels of SBDPs in CSF were markedly increased following acute TBI in comparison with those in the control group. In the early period after TBI, the levels of SBDPs were closely associated with GCS score. Comparisons of the SBDP levels with the severity of injury revealed significant differences between patients with the most severe brain injury and patients with severe brain injury in the first 24 h post-injury (P

East-ADL is an architectural description language dedicated to safety-critical automotive embedded system design. We have previously modified East-adl to include energy constraints and transformed energy-aware real-time behavioral constraints in East-adl into analyzable Uppaal models. In this paper, we extend our previous work by including support for Stateflow, which is used to design event-driven systems via hierarchical state machines and flow charts. However, Stateflow provides limited support for formal analysis and often suffers from incomplete coverage issues since it was originally designed for the simulation of designs and as such does not provide a model amenable to formal verification. We tackle that shortcoming by transforming Stateflow models into verifiable Uppaal models with stochastic semantics and integrating the translation with formal statistical analysis techniques: a flattening strategy and a set of mapping rules are proposed to facilitate the guarantee of translation. The analysis techniques, including the flattening and mapping strategy, are validated and demonstrated on the Fault-Tolerant Fuel Control case study.

Torchless is a procedurally generated competitive dungeon crawling game where players have distinct asymmetric roles while sharing the same screen. Utilizing asymmetry, Torchless explores the relationship between players by experimenting with available information, player power, and player objectives.