Graph Regularized Autoencoder and its Application in Unsupervised Anomaly Detection

Ahmed, Imtiaz; Galoppo, Travis; Hu, Xia; Ding, Yu

doi:10.1109/tpami.2021.3066111

Cited by 18 publications

(13 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Such 3D information is of pivotal importance to predicting crystal structure outcomes, such as bond lengths, polyhedral distortions, and long-range bonding motifs. 3D graph neural networks (3D GNNs), that explicitly encode 3D positional information about molecular and materials composition and structure, or GNNs capable of learning low-dimensional embedding manifold in a highdimensional design space, 99 will enable reaction precursors to be mapped to intermediates and final productsthereby establishing proxies for the most probable reaction trajectories. The mathematical behavior of traditional material precursor− crystal−structure−property functions is inadequate to model far-from equilibrium materials.…”

Section: Prospects For Improved Navigation Of the Design Spacementioning

confidence: 99%

Lone but Not Alone: Precise Positioning of Lone Pairs for the Design of Photocatalytic Architectures

et al. 2022

Self Cite

View full text Add to dashboard Cite

With current economic growth and consumption trends projected to bring about a precipitous and rapid rise of the global temperature, the world stands at a crossroads with regards to climate change. The rate at which greenhouse gas emissions from fossil fuels, industry, and land-use is curtailed over the next decade will determine the trajectory of global warming for the rest of the century. It is increasingly apparent that far-reaching decarbonization of the transportation infrastructure will need to be supplemented by extensive carbon capture, storage, and utilization. Taking a leaf from Nature's playbook, photocatalytic architectures that can utilize water or CO 2 in conjunction with energy harvested from sunlight and store it in the form of energy-dense chemical bonds represent an attractive proposition. Harnessing solar irradiance, through solar energy conversion involving photovoltaics, as well as the photocatalytic generation of solar fuels, and the photocatalytic reduction of CO 2 have emerged as urgent imperatives for the energy transition. Functional photocatalysts must be capable of efficiently absorbing sunlight, effectively separating electronhole pairs, and ensuring they are delivered at appropriate potentials to catalytic sites to mediate redox reactions. Such photocatalytic architectures must further direct redox events down specific pathways to yield desired products, and ensure the transport of reactants between catalytic sites; all with high efficiency and minimal degradation. In this Perspective, we describe a palette of heterostructures designed to promote robust and efficient direct solar-driven water splitting and CO 2 reduction. The heterostructures comprise M x V 2 O 5 or M x M y ′V 2 O 5 , where M is a p-block cation, M′ is an s-, p-, or d-block cation, and V 2 O 5 represents one of multiple polymorphs of this composition interfaced with semiconductor quantum dots (QDs, binary or ternary II−VI or III−V QDs). The stereochemically active 5/6s 2 electron lone pairs of p-block cations in M x V 2 O 5 give rise to filled midgap electronic states that reside above the O 2p-derived valence band. Within heterostructures, the photoexcitation of QDs results in the transfer of holes to the midgap states of M x V 2 O 5 or M x M y ′V 2 O 5 on subpicosecond time scales. Ultrafast charge separation minimizes the photoanodic corrosion of QDs, which has historically been a major impediment to their use in photocatalysis, and enables charge transport and the subsequent redox reactions underpinning photocatalysis to compete with electron−hole recombination. The energy positioning and dispersion of lone pair states is tunable through multiple chemical and compositional levers accessible across the palette of M x V 2 O 5 or M x M y ′V 2 O 5 compounds: choice of lone-pair cation M and its stoichiometry x, atomic connectivity of V 2 O 5 polymorphs, cointercalation of M′ cations in "quaternary" vanadium oxide bronzes, anionic substitution, and alternative lone pair vanadate frameworks with altogether different c...

show abstract

Section: Prospects For Improved Navigation Of the Design Spacementioning

confidence: 99%

Lone but Not Alone: Precise Positioning of Lone Pairs for the Design of Photocatalytic Architectures

et al. 2022

Self Cite

View full text Add to dashboard Cite

show abstract

“…3. Given the training data matrix X ∈ R N ×m , according to (23) and ( 9), we have the system feature matrix T ∈ R N ×a and residual matrix X ∈ R N ×m . Set x ∈ R m to be a sample of X, its Hotelling's T 2 statistics and squared prediction error (SP E) statistics can be defined as follows:…”

Section: Dae-pca Based Nonlinear Fault Detectionmentioning

confidence: 99%

Learnable Faster Kernel-PCA for Nonlinear Fault Detection: Deep Autoencoder-Based Realization

Ren¹,

Yang²,

Jiang³

et al. 2021

Preprint

View full text Add to dashboard Cite

Kernel principal component analysis (KPCA) is a well-recognized nonlinear dimensionality reduction method that has been widely used in nonlinear fault detection tasks. As a kernel trick-based method, KPCA inherits two major problems. First, the form and the parameters of the kernel function are usually selected blindly, depending seriously on trial-and-error. As a result, there may be serious performance degradation in case of inappropriate selections. Second, at the online monitoring stage, KPCA has much computational burden and poor real-time performance, because the kernel method requires to leverage all the offline training data. In this work, to deal with the two drawbacks, a learnable faster realization of the conventional KPCA is proposed. The core idea is to parameterize all feasible kernel functions using the novel nonlinear DAE-FE (deep autoencoder based feature extraction) framework and propose DAE-PCA (deep autoencoder based principal component analysis) approach in detail. The proposed DAE-PCA method is proved to be equivalent to KPCA but has more advantage in terms of automatic searching of the most suitable nonlinear high-dimensional space according to the inputs. Furthermore, the online computational efficiency improves by approximately 100 times compared with the conventional KPCA. With the Tennessee Eastman (TE) process benchmark, the effectiveness and superiority of the proposed method is illustrated.

show abstract

“…As defined in (23), v is a latent variable for constructing process variables (u, y) in the nominal operation. The fact (24) reveals that v preserves the exact amount of information needed for constructing the process variables (u, y) in the nominal operation. As the latent variable, v is achieved by maximally compressed mapping of (u, y) and preserves as much as possible the information in (u, y) by its reconstruction.…”

Section: A Introduction Of the Basic Ideamentioning

confidence: 99%

Control theoretically explainable application of autoencoder methods to fault detection in nonlinear dynamic systems

Li¹,

Ding²,

Liang³

et al. 2022

Preprint

View full text Add to dashboard Cite

Graph Regularized Autoencoder and its Application in Unsupervised Anomaly Detection

Cited by 18 publications

References 44 publications

Lone but Not Alone: Precise Positioning of Lone Pairs for the Design of Photocatalytic Architectures

Lone but Not Alone: Precise Positioning of Lone Pairs for the Design of Photocatalytic Architectures

Learnable Faster Kernel-PCA for Nonlinear Fault Detection: Deep Autoencoder-Based Realization

Control theoretically explainable application of autoencoder methods to fault detection in nonlinear dynamic systems

Contact Info

Product

Resources

About