Inverse Mapping Research Articles

Label Deconvolution for Node Representation Learning on Large-Scale Attributed Graphs Against Learning Bias.

Node representation learning on attributed graphs-whose nodes are associated with rich attributes (e.g., texts and protein sequences)-plays a crucial role in many important downstream tasks. To encode the attributes and graph structures simultaneously, recent studies integrate pre-trained models with graph neural networks (GNNs), where pre-trained models serve as node encoders (NEs) to encode the attributes. As jointly training large NEs and GNNs on large-scale graphs suffers from severe scalability issues, many methods propose to train NEs and GNNs separately. Consequently, they do not take feature convolutions in GNNs into consideration in the training phase of NEs, leading to a significant learning bias relative to the joint training. To address this challenge, we propose an efficient label regularization technique, namely Label Deconvolution (LD), to alleviate the learning bias by a novel and highly scalable approximation to the inverse mapping of GNNs. The inverse mapping leads to an objective function that is equivalent to that by the joint training, while it can effectively incorporate GNNs in the training phase of NEs against the learning bias. More importantly, we show that LD converges to the optimal objective function values by the joint training under mild assumptions. Experiments demonstrate LD significantly outperforms state-of-the-art methods on Open Graph Benchmark datasets.

Integrating deep learning and discrete cosine transform for surface waves full waveform inversion

Summary Accurate estimations of near-surface S-wave velocity (Vs) models hold particular significance in geological and engineering investigations. On the one hand, the popular Multichannel Analysis of Surface Waves (MASW) is limited to the 1D and the plane wave assumptions. On the other hand, the more advanced and computationally expensive full-waveform inversion (FWI) approach is often solved within a deterministic framework that hampers an accurate uncertainty assessment and makes the final predictions heavily reliant on the starting model. Here we combine deep learning with Discrete Cosine Transforms (DCT) to solve the FWI of surface waves and to efficiently estimate the inversion uncertainties. Our neural network approach effectively learns the inverse non-linear mapping between DCT-compressed seismograms and DCT-compressed S-velocity models. The incorporation of DCT into the deep learning framework provides several advantages: it notably reduces parameter space dimensionality and alleviates the ill-conditioning of the problem. Additionally, it decreases the complexity of the network architecture and the computational cost for the training phase compared to training in the full domain. A Monte Carlo simulation is also used to propagate the uncertainties from the data to the model space. We first test the implemented inversion method on synthetic data to showcase the generalization capabilities of the trained network and to explore the implications of incorrect noise assumptions in the recorded seismograms and inaccurate wavelet estimations. Further, we demonstrate the applicability of the implemented method to field data. In this case, available borehole information is used to validate our predictions. In both the synthetic and field applications, the predictions provided by the proposed method are compared with those of a deterministic FWI and the outcomes of a network trained in the full data and model spaces. Our experiments confirm that the implemented deep-learning inversion efficiently and successfully solves the FWI problem and yields more accurate and stable results than a network trained without the DCT compression. This opens the possibility to efficiently train a neural network that provides accurate instantaneous predictions of Vs near-surface models and related uncertainties.

Diffeomorphisms of Foliated Manifolds I

The set $Diff(M)$ of all diffeomorphisms of manifold $M$ onto itself is the group related to composition and inverse mapping. The group of diffeomorphisms of smooth manifolds is of great importance in differential geometry and analysis. It is known that the group $Diff(M)$ is topological group in compact open topology.In this paper we investigate the group $Diff_{F}(M)$ of diffeomorphisms foliated manifold $(M,F)$ with foliated compact open topology. In this paper we prove that if all leaves of the the foliation $F$ are closed subsets of $M$ then the foliated compact open topology of the group $Diff_{F}(M)$ coincides with compact open topology. In addition it is studied the question on the dimension of the group of isometries of foliated manifold is studied when foliation generated by riemannian submersion.

A classification model of road accumulated snow severity based on the fusion of DeepLabV3 + and U-Net

Treacherous driving conditions, especially accumulated snow on roads and bridges, pose a serious safety concern during the winter season. Developing accurate warning systems to classify the severity of snow accumulation in severe weather is crucial. To address this problem, this study investigated the severity of snow accumulation on roads using computer vision techniques. After discussing and analyzing the advantages and disadvantages of the DeepLabV3+ and U-Net models, this paper proposed a classification model for road snow severity based on the fusion of these two models in complex backgrounds. The recognition segmentation task was divided into two stages, with Inverse Perspective Mapping (IPM) added to the output stage of both models. Compared to traditional single-stage models, Pixel Accuracy improved from 76.88% to 97.44%, Mean Intersection over Union from 78.36% to 95.57%, and Class-wise Mean Pixel Accuracy from 79.57% to 96.01%.

Real-Time Aerodynamic Load Estimation for Hypersonics via Strain-Based Inverse Maps

This work develops an efficient real-time inverse formulation for inferring the aerodynamic surface pressures on a hypersonic vehicle from sparse measurements of the structural strain. The approach aims to provide real-time estimates of the aerodynamic loads acting on the vehicle for ground and flight testing, as well as guidance, navigation, and control applications. Specifically, the approach targets hypersonic flight conditions where direct measurement of the surface pressures is challenging due to the harsh aerothermal environment. For problems employing a linear elastic structural model, the inference problem can be posed as a least-squares problem with a linear constraint arising from a finite element discretization of the governing elasticity partial differential equation. Due to the linearity of the problem, an explicit solution is given by the normal equations. Precomputation of the resulting inverse map enables rapid evaluation of the surface pressure and corresponding integrated quantities, such as the force and moment coefficients. The inverse approach additionally allows for uncertainty quantification, providing insights for theoretical recoverability and robustness to sensor noise. Numerical studies demonstrate the estimator performance for reconstructing the surface pressure field, as well as the force and moment coefficients, for the Initial Concept 3.X (IC3X) conceptual hypersonic vehicle.

Agency at a distance: learning causal connections

AbstractIn a series of papers, we have argued that causal cognition has coevolved with the use of various tools. Animals use tools, but only as extensions of their own bodies, while humans use tools that act at a distance in space and time. This means that we must learn new types of causal mappings between causes and effects. The aim of this article is to account for what is required for such learning of causal relations. Following a proposal by Grush and Springle, we argue that learning of inverse mappings from effects to causes is central. Learning such mappings also involves constraints based on monotonicity, continuity and convexity. In order for causal thinking to extend beyond space and time, mental simulations are required that predict the effects of actions. More advanced forms of causal reasoning involve more complicated forms of simulations.

Stereo Bi-Telecentric Phase-Measuring Deflectometry.

Replacing the endocentric lenses in traditional Phase-Measuring Deflectometry (PMD) with bi-telecentric lenses can reduce the number of parameters to be optimized during the calibration process, which can effectively increase both measurement precision and efficiency. Consequently, the low distortion characteristics of bi-telecentric PMD contribute to improved measurement accuracy. However, the calibration of the extrinsic parameters of bi-telecentric lenses requires the help of a micro-positioning stage. Using a micro-positioning stage for the calibration of external parameters can result in an excessively cumbersome and time-consuming calibration process. Thus, this study proposes a holistic and flexible calibration solution for which only one flat mirror in three poses is needed. In order to obtain accurate measurement results, the calibration residuals are utilized to construct the inverse distortion map through bicubic Hermite interpolation in order to obtain accurate anchor positioning result. The calibrated stereo bi-telecentric PMD can achieve 3.5 μm (Peak-to-Valley value) accuracy within 100 mm (Width) × 100 mm (Height) × 200 mm (Depth) domain for various surfaces. This allows the obtaining of reliable measurement results without restricting the placement of the surface under test.

Decode, Move and Speak! Self-supervised Learning of Speech Units, Gestures, and Sound Relationships Using Vocal Imitation

Abstract Speech learning encompasses mastering a complex motor system to produce speech sounds from articulatory gestures while simultaneously uncovering discrete units that provide entry to the linguistic system. Remarkably, children acquire these associations between speech sounds, articulatory gestures, and linguistic units in a weakly supervised manner, without the need for explicit labeling of auditory inputs or access to target articulatory gestures. This study uses self-supervised deep learning to investigate the respective roles of sounds, gestures, and linguistic units in speech acquisition and control. In a first experiment, we analyzed the quantized representations learned by vector-quantized variational autoencoders (VQ-VAE) from ground truth acoustic and articulatory data using ABX tests. We show an interesting complementarity between acoustic and articulatory modalities that may help in the discovery of phonemes. In a second experiment, we introduce a computational agent that repeats auditory speech inputs by controlling a virtual vocal apparatus. This agent integrates an articulatory synthesizer capable of reproducing diverse speech stimuli from interpretable parameters, along with two internal models implementing the articulatory-to-acoustic (forward) and acoustic-to-articulatory (inverse) mapping, respectively. Additionally, two inductive biases are used to regularize the ill-posed acoustic-to-articulatory inverse mapping. In line with the first experiment, we explore the complementarity between the auditory input and the articulatory parameters inferred by the agent. We also evaluate the impact of discretizing auditory inputs using VQ-VAE. While the majority of the agent’s productions are intelligible (according to perceptual evaluations), our analysis highlights inconsistencies in the underlying articulatory trajectories. In particular, we show that the agent’s productions only partially reproduce the complementarity between the auditory and articulatory modalities observed in humans.

Multi-Objective Design of DNA-Stabilized Nanoclusters Using Variational Autoencoders With Automatic Feature Extraction.

DNA-stabilized silver nanoclusters (AgN-DNAs) have sequence-tuned compositions and fluorescence colors. High-throughput experiments together with supervised machine learning models have recently enabled design of DNA templates that select for AgN-DNA properties, including near-infrared (NIR) emission that holds promise for deep tissue bioimaging. However, these existing models do not enable simultaneous selection of multiple AgN-DNA properties, and require significant expert input for feature engineering and class definitions. This work presents a model for multiobjective, continuous-property design of AgN-DNAs with automatic feature extraction, based on variational autoencoders (VAEs). This model is generative, i.e., it learns both the forward mapping from DNA sequence to AgN-DNA properties and the inverse mapping from properties to sequence, and is trained on an experimental data set of DNA sequences paired with AgN-DNA fluorescence properties. Experimental testing shows that the model enables effective design of AgN-DNA emission, including bright NIR AgN-DNAs with 4-fold greater abundance compared to training data. In addition, Shapley analysis is employed to discern learned nucleobase patterns that correspond to fluorescence color and brightness. This generative model can be adapted for a range of biomolecular systems with sequence-dependent properties, enabling precise design of emerging biomolecular nanomaterials.

Backward induction-based deep image search.

Conditional image retrieval (CIR), which involves retrieving images by a query image along with user-specified conditions, is essential in computer vision research for efficient image search and automated image analysis. The existing approaches, such as composed image retrieval (CoIR) methods, have been actively studied. However, these methods face challenges as they require either a triplet dataset or richly annotated image-text pairs, which are expensive to obtain. In this work, we demonstrate that CIR at the image-level concept can be achieved using an inverse mapping approach that explores the model's inductive knowledge. Our proposed CIR method, called Backward Search, updates the query embedding to conform to the condition. Specifically, the embedding of the query image is updated by predicting the probability of the label and minimizing the difference from the condition label. This enables CIR with image-level concepts while preserving the context of the query. In this paper, we introduce the Backward Search method that enables single and multi-conditional image retrieval. Moreover, we efficiently reduce the computation time by distilling the knowledge. We conduct experiments using the WikiArt, aPY, and CUB benchmark datasets. The proposed method achieves an average mAP@10 of 0.541 on the datasets, demonstrating a marked improvement compared to the CoIR methods in our comparative experiments. Furthermore, by employing knowledge distillation with the Backward Search model as the teacher, the student model achieves a significant reduction in computation time, up to 160 times faster with only a slight decrease in performance. The implementation of our method is available at the following URL: https://github.com/dhlee-work/BackwardSearch.

Image based information hiding via minimization of entropy and randomization

In this paper, a new approach that can effectively and securely hide information into color images with significantly improved security and hiding capacity is proposed. The proposed approach performs information hiding in three major steps. As the first step, two binary sequences are constructed from the least significant bits in the pixels of a cover image and the information that needs to be embedded, the information entropies of both sequences are minimized with a dynamic programming method. In the second step, the resulting sequences are randomly reshuffled into randomized sequences with mappings based on a set of one-dimensional chaotic systems, a single binary sequence can be obtained by a matching operation performed between the two randomized sequences. Finally, an inverse mapping is applied to the sequence obtained in the second step, and the transformed sequence is embedded into the least significant bits in the pixels of the cover image. Both analysis and experiments show that the proposed approach can achieve guaranteed performance in both security and capacity for long binary sequences. In addition, a comparison with other state-of-the-art methods for image-based information hiding suggests that the proposed approach can achieve significantly improved performance and is promising for practical applications.

An optimal grid scheduling method considering coupling degree of nodes and duality of reciprocal inverse mapping

The optimal scheduling method of the power grid considering the node coupling degree and the duality of mutual inverse mapping is studied to effectively establish the random dynamic scene of the power grid and improve its optimal scheduling effect. Considering the duality of the mutual inverse mapping, a random dynamic scene of the power grid is generated. The optimization model of the grid random dynamic scene partition was established considering the coupling degree of the nodes in the region. The anti-prey particle swarm optimization algorithm was used to solve the optimization model, and the scene partition results were obtained. To construct an optimization scheduling model for the regional power grid with the objective functions of minimizing energy abandonment rate and grid loss and the constraints of power flow, output, and climbing power. A predator-prey particle swarm optimization algorithm is used to solve the optimization scheduling model, and the minimum energy abandonment rate, minimum grid loss, and corresponding optimal scheduling strategies are obtained. The experimental results show that this method can effectively generate random dynamic scenes of a power grid and has a better scene partitioning effect. Under different working conditions, this method can achieve optimal dispatching of the power grid and reduce the power grid energy abandonment rate and network loss.

EVALUATING PEDESTRIAN CROSSING ATTRIBUTES AT INTERSECTIONS IN KABUL CITY, AFGHANISTAN: A COMPUTER VISION APPROACH

This study aims to evaluate pedestrian crossing attributes in heterogeneous traffic environments using computer vision. For this purpose, YoloV8 models were optimised to detect pedestrian crossing attributes. Moreover, an OpenCV-based Python programme was developed to track pedestrian trajectories manually. For accuracy, the inverse perspective mapping method is applied to obtain a bird's eye view. Finally, a heatmap of pedestrian trajectories was provided to visualise the pedestrian crossing attributes. The results show that more than three-quarters of pedestrians are engaging in noncompliance crossing behaviours at major intersections in Kabul City, Afghanistan. In addition, pedestrians tend to walk longer, more frequent routes at corners and outside of crosswalks. Furthermore, statistical analysis reveals that pedestrian crossing speed decreased by 5.8% when disobeying crossing rules, indicating the significant effect of pedestrian attributes on crossing speed. In conclusion, this study contributes to a better understanding of pedestrian behaviour in heterogeneous traffic environments using computer vision. The results would provide insightful information to traffic engineers and planners for traffic management.

The generalized metric property in strongly topological gyrogroups

A topological gyrogroup is a gyrogroup endowed with a topology such that the binary operation is jointly continuous and the inverse mapping is also continuous. It is shown that a strongly topological gyrogroup is a q-space if and only if it is an M-space. Then a characterization about weakly feathered strongly topological gyrogroups is given, that is, a strongly topological gyrogroup G is weakly feathered if and only if it contains a compact strong subgyrogroup H such that the quotient space G/H is submetrizable.

Non-invasive estimation of the powder size distribution from a single speckle image

Non-invasive characterization of powders may take one of two approaches: imaging and counting individual particles; or relying on scattered light to estimate the particle size distribution (PSD) of the ensemble. The former approach runs into practical difficulties, as the system must conform to the working distance and other restrictions of the imaging optics. The latter approach requires an inverse map from the speckle autocorrelation to the particle sizes. The principle relies on the pupil function determining the basic sidelobe shape, whereas the particle size spread modulates the sidelobe intensity. We recently showed that it is feasible to invert the speckle autocorrelation and obtain the PSD using a neural network, trained efficiently through a physics-informed semi-generative approach. In this work, we eliminate one of the most time-consuming steps of our previous method by engineering the pupil function. By judiciously blocking portions of the pupil, we sacrifice some photons but in return we achieve much enhanced sidelobes and, hence, higher sensitivity to the change of the size distribution. The result is a 60 × reduction in total acquisition and processing time, or 0.25 seconds per frame in our implementation. Almost real-time operation in our system is not only more appealing toward rapid industrial adoption, it also paves the way for quantitative characterization of complex spatial or temporal dynamics in drying, blending, and other chemical and pharmaceutical manufacturing processes.

Mutual information-based feature selection for inverse mapping parameter updating of dynamical systems

AbstractA digital twin should be and remain an accurate model representation of a physical system throughout its operational life. To this end, we aim to update (physically interpretable) parameters of such a model in an online fashion. Hereto, we employ the inverse mapping parameter updating (IMPU) method that uses an artificial neural network (ANN) to map features, extracted from measurement data, to parameter estimates. This is achieved by training the ANN offline on simulated data, i.e., pairs of known parameter value sets and sets of features extracted from corresponding simulations. Since a plethora of features (and feature types) can be extracted from simulated time domain data, feature selection (FS) strategies are investigated. These strategies employ the mutual information between features and parameters to select an informative subset of features. Hereby, accuracy of the parameters estimated by the ANN is increased and, at the same time, ANN training and inference computation times are decreased. Additionally, Bayesian search-based hyperparameter tuning is employed to enhance performance of the ANNs and to optimize the ANN structure for various FS strategies. Finally, the IMPU method is applied to a high-tech industrial use case of a semi-conductor machine, for which measurements are performed in closed-loop on the controlled physical system. This system is modeled as a nonlinear multibody model in the Simscape multibody environment. It is shown that the model updated using the IMPU method simulates the measured system more accurately than a reference model of which the parameter values have been determined manually.

Manifold learning-based reduced-order model for full speed flow field

Reduced-order models (ROMs) can effectively balance the accuracy and efficiency of computational fluid dynamics (CFD). The nonlinear flow field characteristics cannot be captured accurately by traditional ROMs, such as proper orthogonal decomposition (POD). Combining isometric mapping (ISOMAP) and local linear embedding (LLE), a novel manifold learning method (ISOMAP-Local LLE) is proposed, performing global and accurate reconstruction of the nonlinear flow field. First, the nonlinear dimensionality reduction is derived by considering the global isometric characteristic and the local reconstruction relationship simultaneously. Then, the local interpolation model is constructed to improve the interpolation accuracy introduced by the global interpolation model. Finally, the flow field reconstruction is accomplished based on the inverse mapping. Furthermore, the criteria of hyperparameters have been established to achieve high-precision prediction. Several examples covering full speed flow field are carried out to demonstrate the accuracy and efficiency of ISOMAP-Local LLE. The proposed manifold learning-based ROM achieves prediction accuracy that surpasses that of other ROMs (such as POD, ISOMAP, LLE, etc.), resulting in significant time cost savings for CFD simulations.

Inverse mapping of quantum properties to structures for chemical space of small organic molecules

Computer-driven molecular design combines the principles of chemistry, physics, and artificial intelligence to identify chemical compounds with tailored properties. While quantum-mechanical (QM) methods, coupled with machine learning, already offer a direct mapping from 3D molecular structures to their properties, effective methodologies for the inverse mapping in chemical space remain elusive. We address this challenge by demonstrating the possibility of parametrizing a chemical space with a finite set of QM properties. Our proof-of-concept implementation achieves an approximate property-to-structure mapping, the QIM model (which stands for “Quantum Inverse Mapping”), by forcing a variational auto-encoder with a property encoder to obtain a common internal representation for both structures and properties. After validating this mapping for small drug-like molecules, we illustrate its capabilities with an explainability study as well as by the generation of de novo molecular structures with targeted properties and transition pathways between conformational isomers. Our findings thus provide a proof-of-principle demonstration aiming to enable the inverse property-to-structure design in diverse chemical spaces.

COBLAH: A chaotic OBL initialized hybrid algebraic-heuristic algorithm for optimal S-box construction

The Substitution box (S-box) is the main nonlinear component responsible for the cryptographic strength of any Substitution-Permutation Network (SPN) based block cipher. Generating the S-box with optimal cryptographic properties is one of cryptography's most challenging combinatorial problems because of its enormous search space, lack of guidance, and conflicting performance criteria. This paper introduces a novel Chaotic Opposition-based Learning Initialized Hybrid Algebraic-Heuristic (COBLAH) algorithm, combining the favorable traits of Algebraic and heuristics methods based on Galois field inversion, affine mapping, and Genetic Algorithm (GA). The Galois field inversion and affine mapping are used to construct the S-box, while the GA guides the algebraic construction to find the best bit-matrix and additive vector based on any irreducible polynomial for GF(28). GA initializes with a random population generated using a newly constructed cosine-cubic map incorporated with binarization and Opposition-based Learning (OBL). Further, Multi-Objective Optimization Ratio Analysis (MOORA) is utilized to identify the best S-box from the final optimized population. The performance of the proposed algorithm is evaluated by comparing the generated COBLAH S-box with more than twenty state-of-the-art S-boxes, including Advanced Encryption Standard (AES), Skipjack, Gray, and Affine Power Affine (APA). The COBLAH S-box has nonlinearity 112, Strict Avalanche Criterion (SAC) offset 0.0202, Distance to SAC (DSAC) 332, Differential Approximation Probability (DP) 0.0625, Linear Approximation Probability (LP) 0.0156, Bit Independence Criterion-Strict Avalanche Criterion (BIC-SAC) 0.50006, and Bit Independence Criterion-Nonlinearity (BIC-NL) 112, which stands as the optimal observed thus far. The absence of fixed and opposite fixed points and the fact that it adheres to a single cycle aligns the COBLAH S-box with an ideal S-box. In addition, an image encryption mechanism is utilized to encrypt and decrypt the different images sourced from the standard USC-SIPI image dataset using COBLAH S-box and compared against different state-of-the-art S-boxes based on various image characteristics.

Modeling Medium-Term Debris Cloud of Satellite Breakup via Probabilistic Method

Accurately modeling the evolution of debris clouds from satellite breakups is necessary for assessing the collision risk to space assets. Generally, the evolution of debris clouds can be divided into short-, medium-, and long-term phases. Compared to the short- and long-term phases, the study on the medium-term phase was insufficient. This paper proposes a probabilistic method to model the medium-term evolution of a debris cloud. First, a satellite breakup event can be characterized by a probabilistic density distribution of the initial velocity vectors of all fragments. Then, the distribution in the initial velocity can be transformed to that in three classic orbital elements, i.e., semimajor axis, eccentricity, and inclination. Given that the driving force of medium-term evolution is the J2 perturbation and that J2 has no secular contribution to these three classic orbital elements, the distribution of classic orbital elements remains stationary during the evolution. This feature can be exploited to express the probabilistic density distribution of any given target position in terms of the density distribution of the classic orbital elements by changing variables. Finally, 16 one-variable equations with analytical first derivatives were formulated to find the proper inverse mapping from the given position to the classic orbital elements, and the Jacobian matrix of the inverse mapping was derived. Medium-term debris clouds resulting from two typical breakup velocity distributions were tested using the proposed method. Results show that the proposed method can accurately simulate the evolution of the clouds’ geometrical extent and density flow.