No abstract

In this paper, we present an effective method to craft text adversarial samples, revealing one important yet underestimated fact that DNN-based text classifiers are also prone to adversarial sample attack. Specifically, confronted with different adversarial scenarios, the text items that are important for classification are identified by computing the cost gradients of the input (white-box attack) or generating a series of occluded test samples (blackbox attack). Based on these items, we design three perturbation strategies, namely insertion, modification, and removal, to generate adversarial samples. The experiment results show that the adversarial samples generated by our method can successfully fool both state-of-the-art character-level and wordlevel DNN-based text classifiers. The adversarial samples can be perturbed to any desirable classes without compromising their utilities. At the same time, the introduced perturbation is difficult to be perceived.

Recently, many studies have demonstrated deep neural network (DNN) classifiers can be fooled by the adversarial example, which is crafted via introducing some perturbations into an original sample. Accordingly, some powerful defense techniques were proposed. However, existing defense techniques often require modifying the target model or depend on the prior knowledge of attacks. In this paper, we propose a straightforward method for detecting adversarial image examples, which can be directly deployed into unmodified off-the-shelf DNN models. We consider the perturbation to images as a kind of noise and introduce two classic image processing techniques, scalar quantization and smoothing spatial filter, to reduce its effect. The image entropy is employed as a metric to implement an adaptive noise reduction for different kinds of images. Consequently, the adversarial example can be effectively detected by comparing the classification results of a given sample and its denoised version, without referring to any prior knowledge of attacks. More than 20,000 adversarial examples against some state-of-the-art DNN models are used to evaluate the proposed method, which are crafted with different attack techniques. The experiments show that our detection method can achieve a high overall F1 score of 96.39% and certainly raises the bar for defense-aware attacks.

This work describes a strategy to produce circularly polarized thermally activated delayed fluorescence (CP-TADF). A set of two structurally similar organic emitters SFST and SFOT are constructed, whose spiro architectures containing asymmetric donors result in chirality. Upon grafting within the spiro frameworks, the donor and acceptor are fixed proximally in a face-to-face manner. This orientation allows intramolecular through-space charge transfer (TSCT) to occur in both emitters, leading to TADF properties. The donor units in SFST and SFOT have a sulfur and oxygen atom, respectively; such a subtle difference has great impacts on their photophysical, chiroptical, and electroluminescence (EL) properties. SFOT exhibits greatly enhanced EL performance in doped organic light-emitting diodes, with external quantum efficiency (EQE) up to 23.1%, owing to the concurrent manipulation of highly photoluminescent quantum efficiency (PLQY, ∼90%) and high exciton utilization. As a comparison, the relatively larger sulfur atom in SFST introduces heavy atom effects and leads to distortion of the molecular backbone that lengthens the donor–acceptor distance. SFST thus has lower PLQY and faster nonradiative decay rate. The collective consequence is that the EQE value of SFST, i.e., 12.5%, is much lower than that of SFOT. The chirality of these two spiro emitters results in circularly polarized luminescence. Because SFST has a more distorted molecular architecture than SFOT, the luminescence dissymmetry factor (|g lum|) of circularly polarized luminescence of one enantiomer of the former, namely, either (S)-SFST or (R)-SFST, is almost twice that of (S)-SFOT/(R)-SFOT. Moreover, the CP organic light-emitting diodes (CP-OLEDs) show obvious circularly polarized electroluminescence (CPEL) signals with g EL of 1.30 × 10–3 and 1.0 × 10–3 for (S)-SFST and (S)-SFOT, respectively.

We have measured the dielectric loss in SrTiO3 thin films grown on SrRuO3 electrode layers with thickness ranging from 25 nm to 2.5 μm. The loss depends strongly on the thickness but differently above and below T≈80 K: as the thickness increases, the loss decreases at high temperatures but becomes higher at low temperatures. Our result suggests that, in the high temperature regime, the interfacial dead layer effect dominates while, in the low temperature regime, the losses related to the structural phase transition and quantum fluctuations are important.