Recent Advances in Open Set Recognition: A Survey Geng, Chuanxing; Huang, Sheng-Jun; Chen, Songcan
IEEE transactions on pattern analysis and machine intelligence,
10/2021, Letnik:
43, Številka:
10
Journal Article
Recenzirano
In real-world recognition/classification tasks, limited by various objective factors, it is usually difficult to collect training samples to exhaust all classes when training a recognizer or ...classifier. A more realistic scenario is open set recognition (OSR), where incomplete knowledge of the world exists at training time, and unknown classes can be submitted to an algorithm during testing, requiring the classifiers to not only accurately classify the seen classes, but also effectively deal with unseen ones. This paper provides a comprehensive survey of existing open set recognition techniques covering various aspects ranging from related definitions, representations of models, datasets, evaluation criteria, and algorithm comparisons. Furthermore, we briefly analyze the relationships between OSR and its related tasks including zero-shot, one-shot (few-shot) recognition/learning techniques, classification with reject option, and so forth. Additionally, we also review the open world recognition which can be seen as a natural extension of OSR. Importantly, we highlight the limitations of existing approaches and point out some promising subsequent research directions in this field.
Superstructures have attracted extensive attention because of their potential applications in materials science and biology. Herein, we fabricate the first centimeter‐sized porous superstructure of ...carbon nanosheets (SCNS) by using metal–organic framework nanoparticles as a template and polyvinylpyrrolidone as an additional carbon source. The SCNS shows a honeycomb‐like morphology with wall‐sharing carbon cages, in each cavity of which a porous carbon sphere is encapsulated. A single piece of SCNS is directly used as the electrode for a two‐electrode symmetrical supercapacitor cell without any binders and supports, benefiting from its advantage in ultra‐large geometric size, and the Fe‐immobilized SCNS exhibits excellent catalytic performances for oxygen reduction reaction and in a Zn–air battery. This synthetic strategy presents a facile approach for preparing functional SCNS at centimetric scale with controllable morphologies and compositions favoring the fabrication of energy devices.
Nanosheets to honeycombs: Superstructures have attracted extensive attention in material science and biology. The first centimeter‐sized porous superstructure of carbon nanosheets was prepared by using MOF nanoparticles as a template, which exhibits a honeycomb‐like morphology with wall‐sharing carbon cages and shows great potential applications in electrochemical devices.
Multi‐component metal–organic frameworks (MOFs) with precisely controlled pore environments are highly desired owing to their potential applications in gas adsorption, separation, cooperative ...catalysis, and biomimetics. A series of multi‐component MOFs, namely PCN‐900(RE), were constructed from a combination of tetratopic porphyrinic linkers, linear linkers, and rare‐earth hexanuclear clusters (RE6) under the guidance of thermodynamics. These MOFs exhibit high surface areas (up to 2523 cm2 g−1) and unlimited tunability by modification of metal nodes and/or linker components. Post‐synthetic exchange of linear linkers and metalation of two organic linkers were realized, allowing the incorporation of a wide range of functional moieties. Two different metal sites were sequentially placed on the linear linker and the tetratopic porphyrinic linker, respectively, giving rise to an ideal platform for heterogeneous catalysis.
A series of multi‐component MOFs, namely PCN‐900(RE), were constructed from a combination of tetratopic porphyrinic linkers, linear linkers, and rare‐earth hexanuclear clusters (RE6) under the guidance of thermodynamics. Pore‐environment engineering with precisely placed metal sites was realized in these MOFs, which led to the formation of an efficient heterogeneous catalytic system.
Photocatalytic hydrogen evolution (PHE) over semiconductor photocatalysts is usually constrained by the limited light‐harvesting and separation of photogenerated electron–hole pairs. Most of the ...reported systems focusing on PHE are facilitated by consuming the photoinduced holes with organic sacrificial electron donors (SEDs). The introduction of the SEDs not only causes the environmental problem, but also increases the cost of the reaction. Herein, a dual‐functional photocatalyst is developed with the morphology of sandwiched‐like hollowed Pd@TiO2@ZnIn2S4 nanobox, which is synthesized by choosing microporous zeolites with sub‐nanometer‐sized Pd nanoparticles (Pd NPs) embedded as the sacrificial templates. The ternary Pd@TiO2@ZnIn2S4 photocatalyst exhibits a superior PHE rate (5.35 mmol g−1 h−1) and benzylamine oxidation conversion rate (>99%) simultaneously without adding any other SEDs. The PHE performance is superior to the reported composites of TiO2 and ZnIn2S4, which is attributed to the elevated light capture ability induced by the hollow structure, and the enhanced charge separation efficiency facilitated by the ultrasmall sized Pd NPs. The unique design presented here holds great potential for other highly efficient cooperative dual‐functional photocatalytic reactions.
A dual‐functional photocatalyst with the morphology of sandwiched‐like hollowed Pd@TiO2@ZnIn2S4 nanobox is developed, which is synthesized by choosing microporous zeolites with sub‐nanometer‐sized Pd nanoparticles embedded as the sacrificial templates. The ternary Pd@TiO2@ZnIn2S4 photocatalyst exhibits a superior photocatalytic hydrogen evolution rate and benzylamine oxidation conversion rate simultaneously without adding any other sacrificial electron donors.
The controlled synthesis of multicomponent metal–organic frameworks (MOFs) allows for the precise placement of multiple cooperative functional groups within a framework, leading to emergent ...synergistic effects. Herein, we demonstrate that turn‐on fluorescence sensors can be assembled by combining a fluorophore and a recognition moiety within a complex cavity of a multicomponent MOF. An anthracene‐based fluorescent linker and a hemicyanine‐containing CN−‐responsive linker were sequentially installed into the lattice of PCN‐700. The selective binding of CN− to hemicyanine inhibited the energy transfer between the two moieties, resulting in a fluorescence turn‐on effect. Taking advantage of the high tunability of the MOF platform, the ratio between anthracene and the hemicyanine moiety could be fine‐tuned in order to maximize the sensitivity of the overall framework. The optimized MOF‐sensor had a CN−‐detection limit of 0.05 μm, which is much lower than traditional CN− fluorescent sensors (about 0.2 μm).
An anthracene‐based fluorescent linker and a hemicyanine‐containing CN−‐responsive linker were sequentially installed into the lattice of the metal–organic framework PCN‐700. The proximity and periodic arrangement of the anthracene and hemicyanine linkers facilitated efficient energy transfer between the two components, causing fluorescence quenching. The binding of CN− to hemicyanine inhibited this energy transfer, resulting in fluorescence.
Active learning reduces the labeling cost by iteratively selecting the most valuable data to query their labels. It has attracted a lot of interests given the abundance of unlabeled data and the high ...cost of labeling. Most active learning approaches select either informative or representative unlabeled instances to query their labels, which could significantly limit their performance. Although several active learning algorithms were proposed to combine the two query selection criteria, they are usually ad hoc in finding unlabeled instances that are both informative and representative. We address this limitation by developing a principled approach, termed QUIRE, based on the min-max view of active learning. The proposed approach provides a systematic way for measuring and combining the informativeness and representativeness of an unlabeled instance. Further, by incorporating the correlation among labels, we extend the QUIRE approach to multi-label learning by actively querying instance-label pairs. Extensive experimental results show that the proposed QUIRE approach outperforms several state-of-the-art active learning approaches in both single-label and multi-label learning.
Parkinson's disease (PD) is a progressively debilitating neurodegenerative condition that leads to motor and cognitive dysfunction. At present, clinical treatment can only improve symptoms, but ...cannot effectively protect dopaminergic neurons. Several reports have demonstrated that human umbilical cord mesenchymal stem cells (hucMSCs) afford neuroprotection, while their application is limited because of their uncontrollable differentiation and other reasons. Stem cells communicate with cells through secreted exosomes (Exos), the present study aimed to explore whether Exos secreted by hucMSCs could function instead of hucMSCs. hucMSCs were successfully isolated and characterized, and shown to contribute to 6-hydroxydopamine (6-OHDA)-stimulated SH-SY5Y cell proliferation; hucMSC-derived Exos were also involved in this process. The Exos were purified and identified, and then labeled with PKH 26, it was found that the Exos could be efficiently taken up by SH-SY5Y cells after 12 h of incubation. Pretreatment with Exos promoted 6-OHDA-stimulated SH-SY5Y cells to proliferate and inhibited apoptosis by inducing autophagy. Furthermore, Exos reached the substantia nigra through the blood-brain barrier (BBB) in vivo, relieved apomorphine-induced asymmetric rotation, reduced substantia nigra dopaminergic neuron loss and apoptosis, and upregulated the level of dopamine in the striatum. These results demonstrate that hucMSCs-Exos have a treatment capability for PD and can traverse the BBB, indicating their potential for the effective treatment of PD.
Metal–organic frameworks (MOFs) based on group 3 and 4 metals are considered as the most promising MOFs for varying practical applications including water adsorption, carbon conversion, and ...biomedical applications. The relatively strong coordination bonds and versatile coordination modes within these MOFs endow the framework with high chemical stability, diverse structures and topologies, and interesting properties and functions. Herein, the significant progress made on this series of MOFs since 2018 is summarized and an update on the current status and future trends on the structural design of robust MOFs with high connectivity is provided. Cluster chemistry involving Y, lanthanides (Ln, from La to Lu), actinides (An, from Ac to Lr), Ti, and Zr is initially introduced. This is followed by a review of recently developed MOFs based on group 3 and 4 metals with their structures discussed based on the types of inorganic or organic building blocks. The novel properties and arising applications of these MOFs in catalysis, adsorption and separation, delivery, and sensing are highlighted. Overall, this review is expected to provide a timely summary on MOFs based on group 3 and 4 metals, which shall guide the future discovery and development of stable and functional MOFs for practical applications.
Metal–organic frameworks (MOFs) based on group 3 and 4 metals with high chemical stability, structural diversity, and various interesting properties are reviewed. Cluster and framework chemistry of group 3 and 4 metals are expected to provide a timely summary on MOF development, which shall guide the discovery and development of stable and functional MOFs for practical applications.
Highly efficient, indium tin oxide (ITO)‐free and flexible organic photovoltaic cells are demonstrated with 10.4% power conversion efficiency (PCE) through the innovative integration of ultrathin ...metal film electrode, conductive fullerene interfacial layer, and efficient active blend into a microcavity‐embedded device. In addition, a high PCE of 7.21% is achieved for the upsized devices with an active area of 1 cm2.
Crystal engineering of metal-organic frameworks (MOFs) has allowed the construction of complex structures at atomic precision, but has yet to reach the same level of sophistication as organic ...synthesis. The synthesis of complex MOFs with multiple organic and/or inorganic components is ultimately limited by the lack of control over framework assembly in one-pot reactions. Herein, we demonstrate that multi-component MOFs with unprecedented complexity can be constructed in a predictable and stepwise manner under simple kinetic guidance, which conceptually mimics the retrosynthetic approach utilized to construct complicated organic molecules. Four multi-component MOFs were synthesized by the subsequent incorporation of organic linkers and inorganic clusters into the cavity of a mesoporous MOF, each composed of up to three different metals and two different linkers. Furthermore, we demonstrated the utility of such a retrosynthetic design through the construction of a cooperative bimetallic catalytic system with two collaborative metal sites for three-component Strecker reactions.