arXiv subjects

Bo Zhu

Publications and source records attributed to Bo Zhu.

At least 91 records · Page 5Linked to original sources

arXiv2023

Learning Vortex Dynamics for Fluid Inference and Prediction

We propose a novel differentiable vortex particle (DVP) method to infer and predict fluid dynamics from a single video. Lying at its core is a particle-based latent space to encapsulate the hidden, Lagrangian vortical evolution underpinning the observable, Eulerian flow phenomena. Our differentiable vortex particles are coupled with a learnable, vortex-to-velocity dynamics mapping to effectively capture the complex flow features in a physically-constrained, low-dimensional space. This representation facilitates the learning of a fluid simulator tailored to the input video that can deliver robust, long-term future predictions. The value of our method is twofold: first, our learned simulator enables the inference of hidden physics quantities (e.g., velocity field) purely from visual observation; secondly, it also supports future prediction, constructing the input video's sequel along with its future dynamics evolution. We compare our method with a range of existing methods on both synthetic and real-world videos, demonstrating improved reconstruction quality, visual plausibility, and physical integrity.

cs.LG↗

arXiv2023

Neural Partial Differential Equations with Functional Convolution

We present a lightweighted neural PDE representation to discover the hidden structure and predict the solution of different nonlinear PDEs. Our key idea is to leverage the prior of ``translational similarity'' of numerical PDE differential operators to drastically reduce the scale of learning model and training data. We implemented three central network components, including a neural functional convolution operator, a Picard forward iterative procedure, and an adjoint backward gradient calculator. Our novel paradigm fully leverages the multifaceted priors that stem from the sparse and smooth nature of the physical PDE solution manifold and the various mature numerical techniques such as adjoint solver, linearization, and iterative procedure to accelerate the computation. We demonstrate the efficacy of our method by robustly discovering the model and accurately predicting the solutions of various types of PDEs with small-scale networks and training sets. We highlight that all the PDE examples we showed were trained with up to 8 data samples and within 325 network parameters.

cs.LG↗

arXiv2023

FluidLab: A Differentiable Environment for Benchmarking Complex Fluid Manipulation

Humans manipulate various kinds of fluids in their everyday life: creating latte art, scooping floating objects from water, rolling an ice cream cone, etc. Using robots to augment or replace human labors in these daily settings remain as a challenging task due to the multifaceted complexities of fluids. Previous research in robotic fluid manipulation mostly consider fluids governed by an ideal, Newtonian model in simple task settings (e.g., pouring). However, the vast majority of real-world fluid systems manifest their complexities in terms of the fluid's complex material behaviors and multi-component interactions, both of which were well beyond the scope of the current literature. To evaluate robot learning algorithms on understanding and interacting with such complex fluid systems, a comprehensive virtual platform with versatile simulation capabilities and well-established tasks is needed. In this work, we introduce FluidLab, a simulation environment with a diverse set of manipulation tasks involving complex fluid dynamics. These tasks address interactions between solid and fluid as well as among multiple fluids. At the heart of our platform is a fully differentiable physics simulator, FluidEngine, providing GPU-accelerated simulations and gradient calculations for various material types and their couplings. We identify several challenges for fluid manipulation learning by evaluating a set of reinforcement learning and trajectory optimization methods on our platform. To address these challenges, we propose several domain-specific optimization schemes coupled with differentiable physics, which are empirically shown to be effective in tackling optimization problems featured by fluid system's non-convex and non-smooth properties. Furthermore, we demonstrate reasonable sim-to-real transfer by deploying optimized trajectories in real-world settings.

cs.RO↗

arXiv2022

FSID: Fully Synthetic Image Denoising via Procedural Scene Generation

For low-level computer vision and image processing ML tasks, training on large datasets is critical for generalization. However, the standard practice of relying on real-world images primarily from the Internet comes with image quality, scalability, and privacy issues, especially in commercial contexts. To address this, we have developed a procedural synthetic data generation pipeline and dataset tailored to low-level vision tasks. Our Unreal engine-based synthetic data pipeline populates large scenes algorithmically with a combination of random 3D objects, materials, and geometric transformations. Then, we calibrate the camera noise profiles to synthesize the noisy images. From this pipeline, we generated a fully synthetic image denoising dataset (FSID) which consists of 175,000 noisy/clean image pairs. We then trained and validated a CNN-based denoising model, and demonstrated that the model trained on this synthetic data alone can achieve competitive denoising results when evaluated on real-world noisy images captured with smartphone cameras.

cs.CV↗

arXiv2022

Cellular Topology Optimization on Differentiable Voronoi Diagrams

Cellular structures manifest their outstanding mechanical properties in many biological systems. One key challenge for designing and optimizing these geometrically complicated structures lies in devising an effective geometric representation to characterize the system's spatially varying cellular evolution driven by objective sensitivities. A conventional discrete cellular structure, e.g., a Voronoi diagram, whose representation relies on discrete Voronoi cells and faces, lacks its differentiability to facilitate large-scale, gradient-based topology optimizations. We propose a topology optimization algorithm based on a differentiable and generalized Voronoi representation that can evolve the cellular structure as a continuous field. The central piece of our method is a hybrid particle-grid representation to encode the previously discrete Voronoi diagram into a continuous density field defined in a Euclidean space. Based on this differentiable representation, we further extend it to tackle anisotropic cells, free boundaries, and functionally-graded cellular structures. Our differentiable Voronoi diagram enables the integration of an effective cellular representation into the state-of-the-art topology optimization pipelines, which defines a novel design space for cellular structures to explore design options effectively that were impractical for previous approaches. We showcase the efficacy of our approach by optimizing cellular structures with up to thousands of anisotropic cells, including femur bone and Odonata wing.

cs.GR↗

arXiv2022

Fluidic Topology Optimization with an Anisotropic Mixture Model

Fluidic devices are crucial components in many industrial applications involving fluid mechanics. Computational design of a high-performance fluidic system faces multifaceted challenges regarding its geometric representation and physical accuracy. We present a novel topology optimization method to design fluidic devices in a Stokes flow context. Our approach is featured by its capability in accommodating a broad spectrum of boundary conditions at the solid-fluid interface. Our key contribution is an anisotropic and differentiable constitutive model that unifies the representation of different phases and boundary conditions in a Stokes model, enabling a topology optimization method that can synthesize novel structures with accurate boundary conditions from a background grid discretization. We demonstrate the efficacy of our approach by conducting several fluidic system design tasks with over four million design parameters.

cs.GR↗

arXiv2022

Symplectically Integrated Symbolic Regression of Hamiltonian Dynamical Systems

Here we present Symplectically Integrated Symbolic Regression (SISR), a novel technique for learning physical governing equations from data. SISR employs a deep symbolic regression approach, using a multi-layer LSTM-RNN with mutation to probabilistically sample Hamiltonian symbolic expressions. Using symplectic neural networks, we develop a model-agnostic approach for extracting meaningful physical priors from the data that can be imposed on-the-fly into the RNN output, limiting its search space. Hamiltonians generated by the RNN are optimized and assessed using a fourth-order symplectic integration scheme; prediction performance is used to train the LSTM-RNN to generate increasingly better functions via a risk-seeking policy gradients approach. Employing these techniques, we extract correct governing equations from oscillator, pendulum, two-body, and three-body gravitational systems with noisy and extremely small datasets.

cs.LG↗

arXiv2022

On Real-time Image Reconstruction with Neural Networks for MRI-guided Radiotherapy

MRI-guidance techniques that dynamically adapt radiation beams to follow tumor motion in real-time will lead to more accurate cancer treatments and reduced collateral healthy tissue damage. The gold-standard for reconstruction of undersampled MR data is compressed sensing (CS) which is computationally slow and limits the rate that images can be available for real-time adaptation. Here, we demonstrate the use of automated transform by manifold approximation (AUTOMAP), a generalized framework that maps raw MR signal to the target image domain, to rapidly reconstruct images from undersampled radial k-space data. The AUTOMAP neural network was trained to reconstruct images from a golden-angle radial acquisition, a benchmark for motion-sensitive imaging, on lung cancer patient data and generic images from ImageNet. Model training was subsequently augmented with motion-encoded k-space data derived from videos in the YouTube-8M dataset to encourage motion robust reconstruction. We find that AUTOMAP-reconstructed radial k-space has equivalent accuracy to CS but with much shorter processing times after initial fine-tuning on retrospectively acquired lung cancer patient data. Validation of motion-trained models with a virtual dynamic lung tumor phantom showed that the generalized motion properties learned from YouTube lead to improved target tracking accuracy. Our work shows that AUTOMAP can achieve real-time, accurate reconstruction of radial data. These findings imply that neural-network-based reconstruction is potentially superior to existing approaches for real-time image guidance applications.

physics.med-ph↗

arXiv2022

Learning Spatio-Temporal Downsampling for Effective Video Upscaling

Downsampling is one of the most basic image processing operations. Improper spatio-temporal downsampling applied on videos can cause aliasing issues such as moiré patterns in space and the wagon-wheel effect in time. Consequently, the inverse task of upscaling a low-resolution, low frame-rate video in space and time becomes a challenging ill-posed problem due to information loss and aliasing artifacts. In this paper, we aim to solve the space-time aliasing problem by learning a spatio-temporal downsampler. Towards this goal, we propose a neural network framework that jointly learns spatio-temporal downsampling and upsampling. It enables the downsampler to retain the key patterns of the original video and maximizes the reconstruction performance of the upsampler. To make the downsamping results compatible with popular image and video storage formats, the downsampling results are encoded to uint8 with a differentiable quantization layer. To fully utilize the space-time correspondences, we propose two novel modules for explicit temporal propagation and space-time feature rearrangement. Experimental results show that our proposed method significantly boosts the space-time reconstruction quality by preserving spatial textures and motion patterns in both downsampling and upscaling. Moreover, our framework enables a variety of applications, including arbitrary video resampling, blurry frame reconstruction, and efficient video storage.

cs.CV↗

arXiv2022

Symplectic Neural Networks in Taylor Series Form for Hamiltonian Systems

We propose an effective and lightweight learning algorithm, Symplectic Taylor Neural Networks (Taylor-nets), to conduct continuous, long-term predictions of a complex Hamiltonian dynamic system based on sparse, short-term observations. At the heart of our algorithm is a novel neural network architecture consisting of two sub-networks. Both are embedded with terms in the form of Taylor series expansion designed with symmetric structure. The key mechanism underpinning our infrastructure is the strong expressiveness and special symmetric property of the Taylor series expansion, which naturally accommodate the numerical fitting process of the gradients of the Hamiltonian with respect to the generalized coordinates as well as preserve its symplectic structure. We further incorporate a fourth-order symplectic integrator in conjunction with neural ODEs' framework into our Taylor-net architecture to learn the continuous-time evolution of the target systems while simultaneously preserving their symplectic structures. We demonstrated the efficacy of our Taylor-net in predicting a broad spectrum of Hamiltonian dynamic systems, including the pendulum, the Lotka--Volterra, the Kepler, and the Hénon--Heiles systems. Our model exhibits unique computational merits by outperforming previous methods to a great extent regarding the prediction accuracy, the convergence rate, and the robustness despite using extremely small training data with a short training period (6000 times shorter than the predicting period), small sample sizes, and no intermediate data to train the networks.

cs.LG↗

arXiv2022

Nonseparable Symplectic Neural Networks

Predicting the behaviors of Hamiltonian systems has been drawing increasing attention in scientific machine learning. However, the vast majority of the literature was focused on predicting separable Hamiltonian systems with their kinematic and potential energy terms being explicitly decoupled while building data-driven paradigms to predict nonseparable Hamiltonian systems that are ubiquitous in fluid dynamics and quantum mechanics were rarely explored. The main computational challenge lies in the effective embedding of symplectic priors to describe the inherently coupled evolution of position and momentum, which typically exhibits intricate dynamics. To solve the problem, we propose a novel neural network architecture, Nonseparable Symplectic Neural Networks (NSSNNs), to uncover and embed the symplectic structure of a nonseparable Hamiltonian system from limited observation data. The enabling mechanics of our approach is an augmented symplectic time integrator to decouple the position and momentum energy terms and facilitate their evolution. We demonstrated the efficacy and versatility of our method by predicting a wide range of Hamiltonian systems, both separable and nonseparable, including chaotic vortical flows. We showed the unique computational merits of our approach to yield long-term, accurate, and robust predictions for large-scale Hamiltonian systems by rigorously enforcing symplectomorphism.

cs.LG↗

arXiv2022

Geometry of positive scalar curvature on complete manifold

In this paper, we study the interplay of geometry and positive scalar curvature on a complete, non-compact manifold with non-negative Ricci curvature. In three-dimensional manifold, we prove a minimal volume growth, an estimate of integral of scalar curvature and width. In higher dimensional manifold, we obtain a volume growth with a stronger condition.

math.DG↗

arXiv2021

Fractional strong matching preclusion for Cartesian product graphs

The strong matching preclusion number of a graph, introduced by Park and Ihm in 2011, is the minimum number of vertices and edges whose deletion results in a graph that has neither perfect matchings nor almost perfect matchings. As a generalization, the fractional strong matching preclusion number of a graph is the minimum number of edges and vertices whose deletion leaves the resulting graph without a fractional perfect matching. In this paper, we obtain the fractional strong matching preclusion number for Cartesian product graphs. As an application, the fractional strong matching preclusion number for torus networks is obtained.

math.CO↗

arXiv2021

Uryson width of three dimensional mean convex domain with non-negative Ricci curvature

We prove that for every three dimensional manifold with nonnegative Ricci curvature and strictly mean convex boundary, there exists a Morse function so that each connected component of its level sets has a uniform diameter bound, which depends only on the lower bound of mean curvature. This gives an upper bound of Uryson 1-width for those three manifolds with boundary.

math.DG↗

arXiv2021

Uncover band topology via quantized drift in two-dimensional Bloch oscillations

We propose to measure band topology via quantized drift of Bloch oscillations in a two-dimensional Harper-Hofstadter lattice subjected to tilted fields in both directions. When the difference between the two tilted fields is large, Bloch oscillations uniformly sample all momenta, and hence the displacement in each direction tends to be quantized at multiples of the overall period, regardless of any momentum of initial state. The quantized displacement is related to a reduced Chern number defined as a line integral of Berry curvature in each direction, providing an almost perfect measurement of Chern number. Our scheme can apply to detect Chern number and topological phase transitions not only for the energy-separable band, but also for energy-inseparable bands which cannot be achieved by conventional Thouless pumping or integer quantum Hall effect.

cond-mat.quant-gas↗

arXiv2021

On a dichotomy of the curvature decay of steady Ricci soliton

We establish a dichotomy on the curvature decay for four dimensional complete noncompact non Ricci flat steady gradient Ricci soliton with linear curvature decay and proper potential function. A similar dichotomy is also shown in higher dimensions under the additional assumption that the Ricci curvature is nonnegative outside a compact subset.

math.DG↗

arXiv2021

Edge states in quantum spin chains: the interplay among interaction, gradient magnetic field and Floquet engineering

We explore the edge defects induced by spin-spin interaction in a finite paradigmatic Heisenberg spin chain. By introducing a gradient magnetic field and a periodically-modulated spin-exchange strength, the resonance between modulation frequency and magnetic field gradient can also induce asymmetric defects at the edges, dubbed as Floquet-Wannier-Zeeman edge defects. The interplay between these two types of edge defects allows us to manipulate the magnon edge states from an isolated band into a continuum one. In the high-frequency regime, we analytically derive effective models to interpret the formation mechanisms for edge states by employing the multiscale perturbation analysis. Our study offers new insights to understand and control the magnon edge states governed by the interplay between edge defects induced by the spin-spin interaction and the Floquet-Wannier-Zeeman manipulation.

quant-ph↗

arXiv2021

An End-to-End AI-Based Framework for Automated Discovery of CEST/MT MR Fingerprinting Acquisition Protocols and Quantitative Deep Reconstruction (AutoCEST)

Purpose: To develop an automated machine-learning-based method for the discovery of rapid and quantitative chemical exchange saturation transfer (CEST) MR fingerprinting acquisition and reconstruction protocols. Methods: An MR physics governed AI system was trained to generate optimized acquisition schedules and the corresponding quantitative reconstruction neural-network. The system (termed AutoCEST) is composed of a CEST saturation block, a spin dynamics module, and a deep reconstruction network, all differentiable and jointly connected. The method was validated using a variety of chemical exchange phantoms and an in-vivo mouse brain at 9.4T. Results: The acquisition times for AutoCEST optimized schedules ranged from 35-71s, with a quantitative image reconstruction time of only 29 ms. The resulting exchangeable proton concentration maps for the phantoms were in good agreement with the known solute concentrations for AutoCEST sequences (mean absolute error = 2.42 mM; Pearson's r=0.992 , p$<$0.0001), but not for an unoptimized sequence (mean absolute error = 65.19 mM; Pearson's r=-0.161, p=0.522). Similarly, improved exchange rate agreement was observed between AutoCEST and quantification of exchange using saturation power (QUESP) methods (mean absolute error: 35.8 Hz, Pearson's r=0.971, p$<$0.0001) compared to an unoptimized schedule and QUESP (mean absolute error = 58.2 Hz; Pearson's r=0.959, p$<$0.0001). The AutoCEST in-vivo mouse brain semi-solid proton volume-fractions were lower in the cortex (12.21$\pm$1.37%) compared to the white-matter (19.73 $\pm$ 3.30%), as expected, and the amide proton volume-fraction and exchange rates agreed with previous reports. Conclusion: AutoCEST can automatically generate optimized CEST/MT acquisition protocols that can be rapidly reconstructed into quantitative exchange parameter maps.

physics.med-ph↗