Lack of transparency in deep neural networks (DNNs) make them susceptible to backdoor attacks, where hidden associations or triggers override normal classification to produce unexpected results. For example, a model with a backdoor always identifies a face as Bill Gates if a specific symbol is present in the input. Backdoors can stay hidden indefinitely until activated by an input, and present a serious security risk to many security or safety related applications, e.g., biometric authentication systems or self-driving cars.We present the first robust and generalizable detection and mitigation system for DNN backdoor attacks. Our techniques identify backdoors and reconstruct possible triggers. We identify multiple mitigation techniques via input filters, neuron pruning and unlearning. We demonstrate their efficacy via extensive experiments on a variety of DNNs, against two types of backdoor injection methods identified by prior work. Our techniques also prove robust against a number of variants of the backdoor attack.

As a new class of luminescent nanomaterials, carbon dots (CDs) have aroused significant interest because of their fascinating photoluminescence properties and potential applications in biological, optoelectronic, and energy‐related fields. Strikingly, embedding CDs in host matrices endow them with intriguing luminescent properties, in particular, room temperature phosphorescence and thermally activated delayed fluorescence, due to the confinement effect of the host matrix and the H‐bonding interactions between CDs and the matrix. Here, the state‐of‐the‐art strategies for introducing CDs in various host matrices are summarized, such as nanoporous materials, polyvinyl alcohol, polyurethane, potash alum, layered double hydroxides, amorphous silica, etc. The resultant luminescent properties of the composites and their emission mechanisms are discussed. Their applications in bioimaging, drug delivery/release, sensing, and anticounterfeiting are also presented. Finally, current problems and challenges of CDs‐based composites are noted for future development of such luminescent materials.

Carbon nanotubes (CNTs) are functionalized with nitrogen atoms for reduction of carbon dioxide (CO2 ). The investigation explores the origin of the catalyst's activity and the role of nitrogen chemical states therein. The catalysts show excellent performances, with about 90 % current efficiency for CO formation and stability over 60 hours. The Tafel analyses and density functional theory calculations suggest that the reduction of CO2 proceeds through an initial rate-determining transfer of one electron to CO2 , which leads to the formation of carbon dioxide radical anion (CO2 (.-) ). The initial reduction barrier is too high on pristine CNTs, resulting in a very high overpotentials at which the hydrogen evolution reaction dominates over CO2 reduction. The doped nitrogen atoms stabilize the radical anion, thereby lowering the initial reduction barrier and improving the intrinsic activity. The most efficient nitrogen chemical state for this reaction is quaternary nitrogen, followed by pyridinic and pyrrolic nitrogen.

Carbon dots (CDs) with red-emitting room-temperature phosphorescence (RTP) are rarely reported because of the increasing nonradiative decay of the excited states and the decreasing energy gap between the excited states and ground states. Herein, we demonstrate a facile strategy for modulating the RTP properties of CDs in terms of donor–acceptor energy transfer (EnT) in the CDs-in-zeolite system. Upon tuning of the heteroatoms (Zn 2+ , Mn 2+ ) doped in the aluminophosphate zeolite frameworks, CDs@zeolite composites with green and red phosphorescence have been prepared via in situ hydrothermal synthesis. In such composites, the zeolite matrix provides an efficient confinement role in stabilizing the triplet states of CDs. Significantly, the Mn-doped zeolite could act as an energy acceptor allowing EnT from excitons of CDs to the dopant in the host matrix, generating the intriguing red RTP behavior. This work provides an effective strategy for developing CD-based composite materials with special RTP emissions as well as new fields for applications.

Biaxial deformation of suspended membranes widely exists and is used in nanoindentation to probe elastic properties of structurally isotropic two-dimensional (2D) materials. However, the elastic properties and, in particular, the fracture behaviors of anisotropic 2D materials remain largely unclarified in the case of biaxial deformation. MoTe2 is a polymorphic 2D material with both isotropic (2H) and anisotropic (1T′ and T d) phases and, therefore, an ideal system of single-stoichiometric materials with which to study these critical issues. Here, we report the elastic properties and fracture behaviors of biaxially deformed, polymorphic MoTe2 by combining temperature-variant nanoindentation and first-principles calculations. It is found that due to similar atomic bonding, the effective moduli of the three phases deviate by less than 15%. However, the breaking strengths of distorted 1T′ and T d phases are only half the value of 2H phase due to their uneven distribution of bonding strengths. Fractures of both isotropic 2H and anisotropic 1T′ phases obey the theorem of minimum energy, forming triangular and linear fracture patterns, respectively, along the orientations parallel to Mo–Mo zigzag chains. Our findings not only provide a reference database for the elastic behaviors of versatile MoTe2 phases but also illuminate a general strategy for the mechanical investigation of any isotropic and anisotropic 2D materials.