اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدWhere Does the Robustness Come from? A Study of the Transformation-based Ensemble Defence
136
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
This paper aims to provide a thorough study on the effectiveness of the transformation-based ensemble defence for image classification and its reasons. It has been empirically shown that they can enhance the robustness against evasion attacks, while there is little analysis on the reasons. In particular, it is not clear whether the robustness improvement is a result of transformation or ensemble. In this paper, we design two adaptive attacks to better evaluate the transformation-based ensemble defence. We conduct experiments to show that 1) the transferability of adversarial examples exists among the models trained on data records after different reversible transformations; 2) the robustness gained through transformation-based ensemble is limited; 3) this limited robustness is mainly from the irreversible transformations rather than the ensemble of a number of models; and 4) blindly increasing the number of sub-models in a transformation-based ensemble does not bring extra robustness gain.
قيم البحث
اقرأ أيضاً
Recent studies show that deep neural networks (DNN) are vulnerable to adversarial examples, which aim to mislead DNNs by adding perturbations with small magnitude. To defend against such attacks, both empirical and theoretical defense approaches have
been extensively studied for a single ML model. In this work, we aim to analyze and provide the certified robustness for ensemble ML models, together with the sufficient and necessary conditions of robustness for different ensemble protocols. Although ensemble models are shown more robust than a single model empirically; surprisingly, we find that in terms of the certified robustness the standard ensemble models only achieve marginal improvement compared to a single model. Thus, to explore the conditions that guarantee to provide certifiably robust ensemble ML models, we first prove that diversified gradient and large confidence margin are sufficient and necessary conditions for certifiably robust ensemble models under the model-smoothness assumption. We then provide the bounded model-smoothness analysis based on the proposed Ensemble-before-Smoothing strategy. We also prove that an ensemble model can always achieve higher certified robustness than a single base model under mild conditions. Inspired by the theoretical findings, we propose the lightweight Diversity Regularized Training (DRT) to train certifiably robust ensemble ML models. Extensive experiments show that our DRT enhanced ensembles can consistently achieve higher certified robustness than existing single and ensemble ML models, demonstrating the state-of-the-art certified L2-robustness on MNIST, CIFAR-10, and ImageNet datasets.
The [CII] 158$mu$m emission line can arise in all phases of the ISM, therefore being able to disentangle the different contributions is an important yet unresolved problem when undertaking galaxy-wide, integrated [CII] observations. We present a new
multi-phase 3D radiative transfer interface that couples Starburst99, a stellar spectrophotometric code, with the photoionisation and astrochemistry codes Mocassin and 3D-PDR. We model entire star forming regions, including the ionised, atomic and molecular phases of the ISM, and apply a Bayesian inference methodology to parametrise how the fraction of the [CII] emission originating from molecular regions, $f_{[CII],mol}$, varies as a function of typical integrated properties of galaxies in the local Universe. The main parameters responsible for the variations of $f_{[CII],mol}$ are specific star formation rate (sSFR), gas phase metallicity, HII region electron number density ($n_e$), and dust mass fraction. For example, $f_{[CII],mol}$ can increase from 60% to 80% when either $n_e$ increases from 10$^{1.5}$ to 10$^{2.5}$cm$^{-3}$, or SSFR decreases from $10^{-9.6}$ to $10^{-10.6}$ yr$^{-1}$. Our model predicts for the Milky Way that $f_{[CII],mol}$$=75.8pm5.9$%, in agreement with the measured value of 75%. When applying the new prescription to a complete sample of galaxies from the Herschel Reference Survey (HRS), we find that anywhere from 60 to 80% of the total integrated [CII] emission arises from molecular regions.
The abundance of data collected by sensors in Internet of Things (IoT) devices, and the success of deep neural networks in uncovering hidden patterns in time series data have led to mounting privacy concerns. This is because private and sensitive inf
ormation can be potentially learned from sensor data by applications that have access to this data. In this paper, we aim to examine the tradeoff between utility and privacy loss by learning low-dimensional representations that are useful for data obfuscation. We propose deterministic and probabilistic transformations in the latent space of a variational autoencoder to synthesize time series data such that intrusive inferences are prevented while desired inferences can still be made with sufficient accuracy. In the deterministic case, we use a linear transformation to move the representation of input data in the latent space such that the reconstructed data is likely to have the same public attribute but a different private attribute than the original input data. In the probabilistic case, we apply the linear transformation to the latent representation of input data with some probability. We compare our technique with autoencoder-based anonymization techniques and additionally show that it can anonymize data in real time on resource-constrained edge devices.
Adversarial training can considerably robustify deep neural networks to resist adversarial attacks. However, some works suggested that adversarial training might comprise the privacy-preserving and generalization abilities. This paper establishes and
quantifies the privacy-robustness trade-off and generalization-robustness trade-off in adversarial training from both theoretical and empirical aspects. We first define a notion, {it robustified intensity} to measure the robustness of an adversarial training algorithm. This measure can be approximate empirically by an asymptotically consistent empirical estimator, {it empirical robustified intensity}. Based on the robustified intensity, we prove that (1) adversarial training is $(varepsilon, delta)$-differentially private, where the magnitude of the differential privacy has a positive correlation with the robustified intensity; and (2) the generalization error of adversarial training can be upper bounded by an $mathcal O(sqrt{log N}/N)$ on-average bound and an $mathcal O(1/sqrt{N})$ high-probability bound, both of which have positive correlations with the robustified intensity. Additionally, our generalization bounds do not explicitly rely on the parameter size which would be prohibitively large in deep learning. Systematic experiments on standard datasets, CIFAR-10 and CIFAR-100, are in full agreement with our theories. The source code package is available at url{https://github.com/fshp971/RPG}.
This paper proposes Characteristic Examples for effectively fingerprinting deep neural networks, featuring high-robustness to the base model against model pruning as well as low-transferability to unassociated models. This is the first work taking bo
th robustness and transferability into consideration for generating realistic fingerprints, whereas current methods lack practical assumptions and may incur large false positive rates. To achieve better trade-off between robustness and transferability, we propose three kinds of characteristic examples: vanilla C-examples, RC-examples, and LTRC-example, to derive fingerprints from the original base model. To fairly characterize the trade-off between robustness and transferability, we propose Uniqueness Score, a comprehensive metric that measures the difference between robustness and transferability, which also serves as an indicator to the false alarm problem.
الأسئلة المقترحة
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
