Search

In breast cancer detection and diagnosis, the longitudinal analysis of mammogram images is crucial. Contemporary models excel in detecting temporal imaging feature changes, thus enhancing the learning process over sequential imaging exams. Yet, the resilience of these longitudinal models against adversarial attacks remains underexplored. In this study, we proposed a novel attack method that capitalizes on the feature-level relationship between two sequential mammogram exams of a longitudinal model, guided by both cross-entropy loss and distance metric learning, to achieve significant attack efficacy, as implemented using attack transferring in a black-box attacking manner. We performed experiments on a cohort of 590 breast cancer patients (each has two sequential mammogram exams) in a case-control setting. Results showed that our proposed method surpassed several state-of-the-art adversarial attacks in fooling the diagnosis models to give opposite outputs. Our method remained effective even if the model was trained with the common defending method of adversarial training.

Recently, there has been a growing availability of pre-trained text models on various model repositories. These models greatly reduce the cost of training new models from scratch as they can be fine-tuned for specific tasks or trained on large datasets. However, these datasets may not be publicly accessible due to the privacy, security, or intellectual property issues. In this paper, we aim to develop a lightweight student network that can learn from multiple teacher models without accessing their original training data. Hence, we investigate Data-Free Knowledge Amalgamation (DFKA), a knowledge-transfer task that combines insights from multiple pre-trained teacher models and transfers them effectively to a compact student network. To accomplish this, we propose STRATANET, a modeling framework comprising: (a) a steerable data generator that produces text data tailored to each teacher and (b) an amalgamation module that implements a self-regulative strategy using confidence estimates from the teachers' different layers to selectively integrate their knowledge and train a versatile student. We evaluate our method on three benchmark text classification datasets with varying labels or domains. Empirically, we demonstrate that the student model learned using our STRATANET outperforms several baselines significantly under data-driven and data-free constraints., Comment: 12 pages, 5 Figures, Proceedings of NAACL 2024

Circuit topology generation plays a crucial role in the design of electronic circuits, influencing the fundamental functionality of the circuit. In this paper, we introduce CIRCUITSYNTH, a novel approach that harnesses LLMs to facilitate the automated synthesis of valid circuit topologies. With a dataset comprising both valid and invalid circuit configurations, CIRCUITSYNTH employs a sophisticated two-phase methodology, comprising Circuit Topology Generation and Circuit Topology Refinement. Experimental results demonstrate the effectiveness of CIRCUITSYNTH compared to various fine-tuned LLM variants. Our approach lays the foundation for future research aimed at enhancing circuit efficiency and specifying output voltage, thus enabling the automated generation of circuit topologies with improved performance and adherence to design requirements., Comment: 6 pages, 1 figure, LAD'24

With the unprecedented advancements in Large Language Models (LLMs), their application domains have expanded to include code generation tasks across various programming languages. While significant progress has been made in enhancing LLMs for popular programming languages, there exists a notable gap in comprehensive evaluation frameworks tailored for Hardware Description Languages (HDLs), particularly VHDL. This paper addresses this gap by introducing a comprehensive evaluation framework designed specifically for assessing LLM performance in VHDL code generation task. We construct a dataset for evaluating LLMs on VHDL code generation task. This dataset is constructed by translating a collection of Verilog evaluation problems to VHDL and aggregating publicly available VHDL problems, resulting in a total of 202 problems. To assess the functional correctness of the generated VHDL code, we utilize a curated set of self-verifying testbenches specifically designed for those aggregated VHDL problem set. We conduct an initial evaluation of different LLMs and their variants, including zero-shot code generation, in-context learning (ICL), and Parameter-efficient fine-tuning (PEFT) methods. Our findings underscore the considerable challenges faced by existing LLMs in VHDL code generation, revealing significant scope for improvement. This study emphasizes the necessity of supervised fine-tuning code generation models specifically for VHDL, offering potential benefits to VHDL designers seeking efficient code generation solutions., Comment: 6 pages, 3 Figures, LAD'24

Adversarial data can lead to malfunction of deep learning applications. It is essential to develop deep learning models that are robust to adversarial data while accurate on standard, clean data. In this study, we proposed a novel adversarially robust feature learning (ARFL) method for a real-world application of breast cancer diagnosis. ARFL facilitates adversarial training using both standard data and adversarial data, where a feature correlation measure is incorporated as an objective function to encourage learning of robust features and restrain spurious features. To show the effects of ARFL in breast cancer diagnosis, we built and evaluated diagnosis models using two independent clinically collected breast imaging datasets, comprising a total of 9,548 mammogram images. We performed extensive experiments showing that our method outperformed several state-of-the-art methods and that our method can enhance safer breast cancer diagnosis against adversarial attacks in clinical settings.

Email is a widely used tool for business communication, and email marketing has emerged as a cost-effective strategy for enterprises. While previous studies have examined factors affecting email marketing performance, limited research has focused on understanding email response behavior by considering email content and metadata. This study proposes a Prototype-based Multi-view Network (PROMINET) that incorporates semantic and structural information from email data. By utilizing prototype learning, the PROMINET model generates latent exemplars, enabling interpretable email response prediction. The model maps learned semantic and structural exemplars to observed samples in the training data at different levels of granularity, such as document, sentence, or phrase. The approach is evaluated on two real-world email datasets: the Enron corpus and an in-house Email Marketing corpus. Experimental results demonstrate that the PROMINET model outperforms baseline models, achieving a ~3% improvement in F1 score on both datasets. Additionally, the model provides interpretability through prototypes at different granularity levels while maintaining comparable performance to non-interpretable models. The learned prototypes also show potential for generating suggestions to enhance email text editing and improve the likelihood of effective email responses. This research contributes to enhancing sender-receiver communication and customer engagement in email interactions., Comment: Accepted at EMNLP 2023 (industry)

Recently we found compelling evidence for a gravitational wave background with Hellings and Downs (HD) correlations in our 15-year data set. These correlations describe gravitational waves as predicted by general relativity, which has two transverse polarization modes. However, more general metric theories of gravity can have additional polarization modes which produce different interpulsar correlations. In this work we search the NANOGrav 15-year data set for evidence of a gravitational wave background with quadrupolar Hellings and Downs (HD) and Scalar Transverse (ST) correlations. We find that HD correlations are the best fit to the data, and no significant evidence in favor of ST correlations. While Bayes factors show strong evidence for a correlated signal, the data does not strongly prefer either correlation signature, with Bayes factors $\sim 2$ when comparing HD to ST correlations, and $\sim 1$ for HD plus ST correlations to HD correlations alone. However, when modeled alongside HD correlations, the amplitude and spectral index posteriors for ST correlations are uninformative, with the HD process accounting for the vast majority of the total signal. Using the optimal statistic, a frequentist technique that focuses on the pulsar-pair cross-correlations, we find median signal-to-noise-ratios of 5.0 for HD and 4.6 for ST correlations when fit for separately, and median signal-to-noise-ratios of 3.5 for HD and 3.0 for ST correlations when fit for simultaneously. While the signal-to-noise-ratios for each of the correlations are comparable, the estimated amplitude and spectral index for HD are a significantly better fit to the total signal, in agreement with our Bayesian analysis., Comment: 11 pages, 5 figures

The Australian, Chinese, European, Indian, and North American pulsar timing array (PTA) collaborations recently reported, at varying levels, evidence for the presence of a nanohertz gravitational wave background (GWB). Given that each PTA made different choices in modeling their data, we perform a comparison of the GWB and individual pulsar noise parameters across the results reported from the PTAs that constitute the International Pulsar Timing Array (IPTA). We show that despite making different modeling choices, there is no significant difference in the GWB parameters that are measured by the different PTAs, agreeing within $1\sigma$. The pulsar noise parameters are also consistent between different PTAs for the majority of the pulsars included in these analyses. We bridge the differences in modeling choices by adopting a standardized noise model for all pulsars and PTAs, finding that under this model there is a reduction in the tension in the pulsar noise parameters. As part of this reanalysis, we "extended" each PTA's data set by adding extra pulsars that were not timed by that PTA. Under these extensions, we find better constraints on the GWB amplitude and a higher signal-to-noise ratio for the Hellings and Downs correlations. These extensions serve as a prelude to the benefits offered by a full combination of data across all pulsars in the IPTA, i.e., the IPTA's Data Release 3, which will involve not just adding in additional pulsars, but also including data from all three PTAs where any given pulsar is timed by more than as single PTA., Comment: 21 pages, 9 figures, submitted to ApJ

We present the results of a Bayesian search for gravitational wave (GW) memory in the NANOGrav 12.5-yr data set. We find no convincing evidence for any gravitational wave memory signals in this data set (Bayes factor = 2.8). As such, we go on to place upper limits on the strain amplitude of GW memory events as a function of sky location and event epoch. These upper limits are computed using a signal model that assumes the existence of a common, spatially uncorrelated red noise in addition to a GW memory signal. The median strain upper limit as a function of sky position is approximately $3.3 \times 10^{-14}$. We also find that there are some differences in the upper limits as a function of sky position centered around PSR J0613$-$0200. This suggests that this pulsar has some excess noise which can be confounded with GW memory. Finally, the upper limits as a function of burst epoch continue to improve at later epochs. This improvement is attributable to the continued growth of the pulsar timing array., Comment: 29 pages, 5 figures

Long COVID is a general term of post-acute sequelae of COVID-19. Patients with long COVID can endure long-lasting symptoms including fatigue, headache, dyspnea and anosmia, etc. Identifying the cohorts with severe long-term complications in COVID-19 could benefit the treatment planning and resource arrangement. However, due to the heterogeneous phenotype presented in long COVID patients, it is difficult to predict their outcomes from their longitudinal data. In this study, we proposed a spatiotemporal attention mechanism to weigh feature importance jointly from the temporal dimension and feature space. Considering that medical examinations can have interchangeable orders in adjacent time points, we restricted the learning of short-term dependency with a Local-LSTM and the learning of long-term dependency with the joint spatiotemporal attention. We also compared the proposed method with several state-of-the-art methods and a method in clinical practice. The methods are evaluated on a hard-to-acquire clinical dataset of patients with long COVID. Experimental results show the Local-LSTM with joint spatiotemporal attention outperformed related methods in outcome prediction. The proposed method provides a clinical tool for the severity assessment of long COVID.

The NANOGrav 15-year data set shows evidence for the presence of a low-frequency gravitational-wave background (GWB). While many physical processes can source such low-frequency gravitational waves, here we analyze the signal as coming from a population of supermassive black hole (SMBH) binaries distributed throughout the Universe. We show that astrophysically motivated models of SMBH binary populations are able to reproduce both the amplitude and shape of the observed low-frequency gravitational-wave spectrum. While multiple model variations are able to reproduce the GWB spectrum at our current measurement precision, our results highlight the importance of accurately modeling binary evolution for producing realistic GWB spectra. Additionally, while reasonable parameters are able to reproduce the 15-year observations, the implied GWB amplitude necessitates either a large number of parameters to be at the edges of expected values, or a small number of parameters to be notably different from standard expectations. While we are not yet able to definitively establish the origin of the inferred GWB signal, the consistency of the signal with astrophysical expectations offers a tantalizing prospect for confirming that SMBH binaries are able to form, reach sub-parsec separations, and eventually coalesce. As the significance grows over time, higher-order features of the GWB spectrum will definitively determine the nature of the GWB and allow for novel constraints on SMBH populations., Comment: Accepted by Astrophysical Journal Letters as part of Focus on NANOGrav's 15-year Data Set and the Gravitational Wave Background. For questions or comments, please email comments@nanograv.org. Edited to fix two equation typos (Eq.13 & 21), and minor text typos

The 15-year pulsar timing data set collected by the North American Nanohertz Observatory for Gravitational Waves (NANOGrav) shows positive evidence for the presence of a low-frequency gravitational-wave (GW) background. In this paper, we investigate potential cosmological interpretations of this signal, specifically cosmic inflation, scalar-induced GWs, first-order phase transitions, cosmic strings, and domain walls. We find that, with the exception of stable cosmic strings of field theory origin, all these models can reproduce the observed signal. When compared to the standard interpretation in terms of inspiraling supermassive black hole binaries (SMBHBs), many cosmological models seem to provide a better fit resulting in Bayes factors in the range from 10 to 100. However, these results strongly depend on modeling assumptions about the cosmic SMBHB population and, at this stage, should not be regarded as evidence for new physics. Furthermore, we identify excluded parameter regions where the predicted GW signal from cosmological sources significantly exceeds the NANOGrav signal. These parameter constraints are independent of the origin of the NANOGrav signal and illustrate how pulsar timing data provide a new way to constrain the parameter space of these models. Finally, we search for deterministic signals produced by models of ultralight dark matter (ULDM) and dark matter substructures in the Milky Way. We find no evidence for either of these signals and thus report updated constraints on these models. In the case of ULDM, these constraints outperform torsion balance and atomic clock constraints for ULDM coupled to electrons, muons, or gluons., Comment: 74 pages, 31 figures, 4 tables; published in Astrophysical Journal Letters as part of Focus on NANOGrav's 15-year Data Set and the Gravitational Wave Background. For questions or comments, please email comments@nanograv.org

We report multiple lines of evidence for a stochastic signal that is correlated among 67 pulsars from the 15-year pulsar-timing data set collected by the North American Nanohertz Observatory for Gravitational Waves. The correlations follow the Hellings-Downs pattern expected for a stochastic gravitational-wave background. The presence of such a gravitational-wave background with a power-law-spectrum is favored over a model with only independent pulsar noises with a Bayes factor in excess of $10^{14}$, and this same model is favored over an uncorrelated common power-law-spectrum model with Bayes factors of 200-1000, depending on spectral modeling choices. We have built a statistical background distribution for these latter Bayes factors using a method that removes inter-pulsar correlations from our data set, finding $p = 10^{-3}$ (approx. $3\sigma$) for the observed Bayes factors in the null no-correlation scenario. A frequentist test statistic built directly as a weighted sum of inter-pulsar correlations yields $p = 5 \times 10^{-5} - 1.9 \times 10^{-4}$ (approx. $3.5 - 4\sigma$). Assuming a fiducial $f^{-2/3}$ characteristic-strain spectrum, as appropriate for an ensemble of binary supermassive black-hole inspirals, the strain amplitude is $2.4^{+0.7}_{-0.6} \times 10^{-15}$ (median + 90% credible interval) at a reference frequency of 1/(1 yr). The inferred gravitational-wave background amplitude and spectrum are consistent with astrophysical expectations for a signal from a population of supermassive black-hole binaries, although more exotic cosmological and astrophysical sources cannot be excluded. The observation of Hellings-Downs correlations points to the gravitational-wave origin of this signal., Comment: 30 pages, 18 figures. Published in Astrophysical Journal Letters as part of Focus on NANOGrav's 15-year Data Set and the Gravitational Wave Background. For questions or comments, please email comments@nanograv.org

Optical coherence tomography (OCT) has seen widespread success as an in vivo clinical diagnostic 3D imaging modality, impacting areas including ophthalmology, cardiology, and gastroenterology. Despite its many advantages, such as high sensitivity, speed, and depth penetration, OCT suffers from several shortcomings that ultimately limit its utility as a 3D microscopy tool, such as its pervasive coherent speckle noise and poor lateral resolution required to maintain millimeter-scale imaging depths. Here, we present 3D optical coherence refraction tomography (OCRT), a computational extension of OCT which synthesizes an incoherent contrast mechanism by combining multiple OCT volumes, acquired across two rotation axes, to form a resolution-enhanced, speckle-reduced, refraction-corrected 3D reconstruction. Our label-free computational 3D microscope features a novel optical design incorporating a parabolic mirror to enable the capture of 5D plenoptic datasets, consisting of millimetric 3D fields of view over up to $\pm75^\circ$ without moving the sample. We demonstrate that 3D OCRT reveals 3D features unobserved by conventional OCT in fruit fly, zebrafish, and mouse samples.

We search NANOGrav's 12.5-year data set for evidence of a gravitational wave background (GWB) with all the spatial correlations allowed by general metric theories of gravity. We find no substantial evidence in favor of the existence of such correlations in our data. We find that scalar-transverse (ST) correlations yield signal-to-noise ratios and Bayes factors that are higher than quadrupolar (tensor transverse, TT) correlations. Specifically, we find ST correlations with a signal-to-noise ratio of 2.8 that are preferred over TT correlations (Hellings and Downs correlations) with Bayesian odds of about 20:1. However, the significance of ST correlations is reduced dramatically when we include modeling of the Solar System ephemeris systematics and/or remove pulsar J0030$+$0451 entirely from consideration. Even taking the nominal signal-to-noise ratios at face value, analyses of simulated data sets show that such values are not extremely unlikely to be observed in cases where only the usual TT modes are present in the GWB. In the absence of a detection of any polarization mode of gravity, we place upper limits on their amplitudes for a spectral index of $\gamma = 5$ and a reference frequency of $f_\text{yr} = 1 \text{yr}^{-1}$. Among the upper limits for eight general families of metric theories of gravity, we find the values of $A^{95\%}_{TT} = (9.7 \pm 0.4)\times 10^{-16}$ and $A^{95\%}_{ST} = (1.4 \pm 0.03)\times 10^{-15}$ for the family of metric spacetime theories that contain both TT and ST modes., Comment: 24 pages, 18 figures, 3 appendices. Please send any comments/questions to Nima Laal (laaln@oregonstate.edu)

We experimentally measure the static stress at the bottom of a granular chains column with a precise and reproducible method. The relation, between the filling mass and the apparent mass converted from the bottom stress, is investigated on various chain lengths. Our measurements reconfirm that the scaling behavior of the stress saturation curves is in accord with the theoretical expectation of the Janssen model. Additionally, the saturation mass is displayed as a nonmonotonic function of the chain length, where a distinguishing transition of the saturation mass is found at the persistence length of the granular chain. We repeat the measurement with another measuring methodology and a silo with different size, respectively, the position of the peak maintains robust. In order to understand the transition of the saturation mass, the friction coefficient and the volume fraction of granular chains are also measured, from which Janssen parameter can be calculated. Finally, we preliminarily measure the bottom stress for two distinct packing structures of long chains, find the effect of the entanglements on the bottom stress, and argue that the entanglements might be responsible for the transition of the saturation mass.

Secure message dissemination is an important issue in vehicular networks, especially considering the vulnerability of vehicle to vehicle message dissemination to malicious attacks. Traditional security mechanisms, largely based on message encryption and key management, can only guarantee secure message exchanges between known source and destination pairs. In vehicular networks however, every vehicle may learn its surrounding environment and contributes as a source, while in the meantime act as a destination or a relay of information from other vehicles, message exchanges often occur between "stranger" vehicles. For secure message dissemination in vehicular networks against insider attackers, who may tamper the content of the disseminated messages, ensuring the consistency and integrity of the transmitted messages becomes a major concern that traditional message encryption and key management based approaches fall short to provide. In this paper, by incorporating the underlying network topology information, we propose an optimal decision algorithm that is able to maximize the chance of making a correct decision on the message content, assuming the prior knowledge of the percentage of malicious vehicles in the network. Furthermore, a novel heuristic decision algorithm is proposed that can make decisions without the aforementioned knowledge of the percentage of malicious vehicles. Simulations are conducted to compare the security performance achieved by our proposed decision algorithms with that achieved by existing ones that do not consider or only partially consider the topological information, to verify the effectiveness of the algorithms. Our results show that by incorporating the network topology information, the security performance can be much improved. This work shed light on the optimum algorithm design for secure message dissemination.

Capacity of vehicular networks with infrastructure support is both an interesting and challenging problem as the capacity is determined by the inter-play of multiple factors including vehicle-to-infrastructure (V2I) communications, vehicle-to-vehicle (V2V) communications, density and mobility of vehicles, and cooperation among vehicles and infrastructure. In this paper, we consider a typical delay-tolerant application scenario with a subset of vehicles, termed Vehicles of Interest (VoIs), having download requests. Each VoI downloads a distinct large-size file from the Internet and other vehicles without download requests assist the delivery of the files to the VoIs. A cooperative communication strategy is proposed that explores the combined use of V2I communications, V2V communications, mobility of vehicles and cooperation among vehicles and infrastructure to improve the capacity of vehicular networks. An analytical framework is developed to model the data dissemination process using this strategy, and a closed form expression of the achievable capacity is obtained, which reveals the relationship between the capacity and its major performance-impacting parameters such as inter-infrastructure distance, radio ranges of infrastructure and vehicles, sensing range of vehicles, transmission rates of V2I and V2V communications, vehicular density and proportion of VoIs. Numerical result shows that the proposed cooperative communication strategy significantly boosts the capacity of vehicular networks, especially when the proportion of VoIs is low. Our results provide guidance on the optimum deployment of vehicular network infrastructure and the design of cooperative communication strategy to maximize the capacity.