12023 research outputs found
Sort by
Density Fluctuations and Machine Learning in Active Matter
Get PDFActive matter is a class of materials that has constituents capable of self-propulsion through the conversion of energy into mechanical motion. The origin or specific details of their method of locomotion is a rich area of research, but is not important for understanding some aspects of their dynamics. Our interest is in the single-particle dynamics and the large-scale collective motion observed in these systems. Thus, we use the minimal active Brownian particle (ABP) model to model self-propulsion. The ABP model consists of particles of radius a that swim with an intrinsic speed U₀ in some direction q and reorients on a timescale τR. Active motion is persistent in that a particle will continue to swim in a direction until it reorients itself, giving rise to a swim or persistence length ℓ = U₀τR. This persistence leads to directed motion at short times but has been shown to become diffusive at long times with a diffusivity that originates from the random swim steps. I show that while these particles do become diffusive at steady state, they display wavelike dynamics while relaxing from an initial point source, which is shown by fluctuations in the number density. The strength of these fluctuations is determined by the ratio of the swim diffusivity to thermal diffusivity. Our resulting theory predicts these dynamics in other instances where spatial gradients in number density are present. This motivated me to look into fluctuations in interacting suspensions of ABPs by studying the "isothermal" compressibility. Our theoretical perspective and simulation results show that the compressibility behaves just like a thermodynamic response function, even though these suspensions are driven far from equilibrium. As such, the compressibility is capable of predicting the critical point for the motility induced phase separation (MIPS). We then developed a machine learning (ML) model to predict particle phase identity near the MIPS critical point in the coexistence region -- where fluctuations are large -- to recreate the binodal. Our successful recreation of the binodal and understanding of compressibility resulted in our attempt to define a Widom line (an extension of the coexistence line) for ABPs. The Widom line is the collection of points where density fluctuations are maximized and marks where supercritical behavior goes from being more gas-like to more liquid-like. I conclude by discussing current works in progress to further our understanding of the MIPS transition by using ML to identify particles in the interface of the coexisting phases. From the work on compressibility, we believe the interface plays a crucial role in describing active phase behavior.</p
Assuring Safety under Uncertainty in Learning-Based Control Systems
Get PDFLearning-based controllers have recently shown impressive results for different robotic tasks in well-defined environments, successfully solving a Rubiks cube and sorting objects in a bin. These advancements promise to enable a host of new capabilities for complex robotic systems. However, these learning-based controllers cannot yet be deployed in highly uncertain environments due to significant issues relating to learning reliability, robustness, and safety.
To overcome these issues, this thesis proposes new methods for integrating model information (e.g. model-based control priors) into the reinforcement learning framework, which is crucial to ensuring reliability and safety. I show, both empirically and theoretically, that this model information greatly reduces variance in learning and can effectively constrain the policy search space, thus enabling significant improvements in sample complexity for the underlying RL algorithms. Furthermore, by leveraging control barrier functions and Gaussian process uncertainty models, I show how system safety can be maintained under uncertainty without interfering with the learning process (e.g. distorting the policy gradients).
The last part of the thesis will discuss fundamental limitations that arise when utilizing machine learning to derive safety guarantees. In particular, I show that widely used uncertainty models can be highly inaccurate when predicting rare events, and examine the implications of this for safe learning. To overcome some of these limitations, a novel framework is developed based on assume-guarantee contracts in order to ensure safety in multi-agent human environments. The proposed approach utilizes contracts to impose loose responsibilities on agents in the environment, which are learned from data. Imposing these responsibilities on agents, rather than treating their uncertainty as a purely random process, allows us to achieve both safety and efficiency in interactions.</p
Intelligent Control for Fixed-Wing eVTOL Aircraft
Get PDFUrban Air Mobility (UAM) holds promise for personal air transportation by deploying "flying cars" over cities. As such, fixed-wing electric vertical take-off and landing (eVTOL) aircraft has gained popularity as they can swiftly traverse cluttered areas, while also efficiently covering longer distances. These modes of operation call for an enhanced level of precision, safety, and intelligence for flight control. The hybrid nature of these aircraft poses a unique challenge that stems from complex aerodynamic interactions between wings, rotors, and the environment. Thus accurate estimation of external forces is indispensable for a high performance flight. However, traditional methods that stitch together different control schemes often fall short during hybrid flight modes. On the other hand, learning-based approaches circumvent modeling complexities, but they often lack theoretical guarantees for stability.
In the first part of this thesis, we study the theoretical benefits of these fixed-wing eVTOL aircraft, followed by the derivation of a novel unified control framework. It consists of nonlinear position and attitude controllers using forces and moments as inputs; and control allocation modules that determine desired attitudes and thruster signals. Next, we present a composite adaptation scheme for linear-in-parameter (LiP) dynamics models, which provides accurate realtime estimation for wing and rotor forces based on measurements from a three-dimensional airflow sensor. Then, we introduce a design method to optimize multirotor configuration that ensures a property of robustness against rotor failures.
In the second part of the thesis, we use deep neural networks (DNN) to learn part of unmodeled dynamics of the flight vehicles. Spectral normalization that regulates the Lipschitz constants of the neural network is applied for better generalization outside the training domain. The resultant network is utilized in a nonlinear feedback controller with a contraction mapping update, solving the nonaffine-in-control issue that arises. Next, we formulate general methods for designing and training DNN-based dynamics, controller, and observer. The general framework can theoretically handle any nonlinear dynamics with prior knowledge of its structure. Finally, we establish a delay compensation technique that transforms nominal controllers for an undelayed system into a sample-based predictive controller with numerical integration. The proposed method handles both first-order and transport delays in actuators and balances between numerical accuracy and computational efficiency to guarantee stability under strict hardware limitations.</p
Synthesis of 2D Quantum Materials for Nanoelectronic and Nanophotonic Applications
Get PDF2D materials have attracted tremendous attention for a variety of properties such as ultra-low body thickness, ultra-high mobility, and tunable bandgap. These unique merits of the 2D materials bring in the significant improvements and new perspectives in the digital CMOS scaling, analog performance, as well as the 3D integration of wafer stacking.
In this thesis, we explore van der Waals materials for future CMOS technologies. Chapter 2 introduces a compatible and a single-step method for synthesizing high-mobility monolayer graphene (MLG) in merely a few minutes by means of plasma-enhanced chemical vapor deposition (PECVD) techniques without the need of active heating. This environment enables graphene growth on different surfaces at relatively low temperatures, which paves ways to a CMOS-compatible approach to graphene synthesis. Chapter 3 describes the development of a synthesis method that controls the growth of large-area h-BN films from monolayer to 30 atomic layers, and summarizes the characterizations of the properties of these h-BN films that demonstrate the high-quality of these materials.
New degrees of freedom possess the immense potential and attract huge attentions as the imminent end of "Moore's Law". Compared with the traditional charge degree of freedom, spin and valley are the other two additional internal degree of freedom in solid-state electronics which enable the spintronic and valleytronic devices with high integration density, fast processing speed, low power dissipation, and non-volatility. Monolayer transition-metal dichalcogenides (TMDCs) in the 2H-phase are semiconductors promising for opto-valleytronic and opto-spintronic applications because of their strong spin-valley coupling. In chapter 4, we report detailed studies of opto-valleytronic properties of heterogeneous domains in CVD-grown monolayer WS₂ single crystals. By illuminating WS₂ with off-resonance circularly-polarized light and measuring the resulting spatially resolved circularly-polarized emission (Pcirc), we find large circular polarization increases significantly to nearly 90% at 80 K. In Chapter 5, it is reported that valley polarized PL of monolayer WS₂ can be efficiently tailored at room temperature (RT) through the surface plasmon-exciton interaction with plasmonic Archimedes spiral (PAS) nanostructures. The DVP of WS₂ using 2 turns (2T) and 4 turns (4T) of PAS can reach up to 40% and 50% at RT, respectively. Further enhancement and continuous control of excitonic valley polarization in electrostatically doped monolayer WS₂ are demonstrated. Under the circularly polarized light on WS₂-2TPAS heterostructure, 40% valley polarization of exciton without electrostatic doping is icreased to 70% by modulating the carrier doping via a backgate. This enhancement of valley polarization may be attributed to the screening of momentum-dependent long-range electron-hole exchange interactions. The demonstration of electrical tunability in the valley-polarized emission from WS₂-PAS heterostructures provides new strategies to harness valley excitons for application in ultrathin valleytronic devices.
In contrast to future optical switch applications, in Chpater 6, it is reported that Ternary tellurides based on alloying different 2D transition metal dichalcogenides can result in interesting new 2D materials with tunable optical and electrical properties. Additionally, such alloys can provide opportunities for significantly improving the electrical contact properties at the metal-semiconductor interface. In particular, realization of practical devices based on the 2D materials will require overcoming the typical Fermi-level pinning limitations of the electrical contacts at the metal-semiconductor interface and ultimately approaching the ideal Schottky-Mott limit. In this work, we develop a simple method of stacking 3D/2D electrical metal contacts onto dangling-bond-free 2D semiconductors in order to surmount the typical issue of Fermi-level pinning. Specifically, contacts of Au, graphene/Au, and WTe₂/Au are transferred onto WS1.94Te0.06 alloy-based devices via a new transfer method. The WS1.94Te0.06 field-effect transistors (FETs) with WTe₂/Au contacts reveal a field-effect mobility of 25 cm²V⁻¹s⁻¹, an on/off current ratio of 10⁶, and extremely low contact resistance of 8 kΩ μm. These electrical properties are far more superior to similar devices with either Au or graphene/Au contacts, which may be attributed to the fact that the work function of WTe₂ is close to the band edge of the WS1.94Te0.06 alloy so that the resulting metal-semiconductor interface of the FETs are free from Fermi-level pinning. The Schottky barrier heights of the WS1.94Te0.06-FETs with WTe₂/Au contacts also follow the general trend of the Schottky-Mott limit, implying high-quality electrical contacts. Finally, in Chapter 7, several promising opportunities were proposed for future CMOS integrated circuits based on monolayer semiconductors.</p
New Method and Analysis of Proximity Trajectory-Only Learned Dynamics for Small Body Gravity Fields
Get PDFRecent missions to small bodies in the past decade (e.g., Rosetta, Hayabusa 2, and OSIRIS-REx) have reshaped our understanding of small bodies and inspired new, more-capable future missions. Despite the high demand for more missions, large uncertainties in small body properties make missions challenging. Recent work in stochastic optimal control can ensure safety in the face of uncertainty in state, constraints, and dynamics. These stochastic optimal controllers require a model of the underlying dynamics, which is difficult for proximity maneuvers and landing around small bodies. Shape models and finite element-like models are the state-of-the-art for high-fidelity gravity models, but they are computationally expensive and do not readily incorporate onboard data. No gravity model yet exists that can use short-horizon position and acceleration data from recent trajectories onboard in safety-critical autonomous proximity maneuvers and landing. Therefore, we propose a new trajectory-only learning-based method to develop a gravity model. We consider three learning frameworks: Gaussian Process Models, Neural Networks, and Physics-Informed Neural Networks. For each framework, we assess the benefits, computational costs, and limitations of the framework. We found that the Gaussian Process Model generally outperforms the other frameworks in cases of moderate uncertainty. As the uncertainty declines or the data is sufficiently filtered, Neural Networks with spectral normalization provide more accurate gravity models and are computationally cheaper to evaluate. Lastly, we reflect on the methods in this thesis and recommend possible problem reformulations for future research.</p
Plus Ultra: Genome-Wide Spatial Transcriptomics with RNA seqFISH+
Get PDFVisualizing single cells and their organization in intact tissue is crucial to understanding their governing biological function. Even though single cell RNA sequencing has provided many insights into the heterogeneity and gene expression profiles across many tissue types, the dissociation process which loses the spatial information is hindering our deeper understanding of how these transcriptional distinct cell types are organized and interacting in their native tissue environment.
The thesis begins by giving a background on how single cell RNA sequencing has transformed biology and the emergence of spatial technology such as sequential fluorescence in situ hybridization (seqFISH). While spatial methods are useful for mapping the cell types identified from single cell RNA sequencing, the need for turning spatial technology such as seqFISH, which has high detection efficiency of the transcriptome with spatial information, into an in situ discovery tool is discussed as the scientific community’s goal heads towards building spatial atlases for every human tissues and organs such as the brain.
While seqFISH has high detection efficiency, it is still limited in the number of genes capable of profiling at once. The major obstacle is the optical crowding problems when more RNA species are targeted and imaged using a fluorescence microscope. In Chapter 2, we first investigated, if the RNA molecules are instead captured on a coverslip and profiled with sequential barcoding strategy, the FISH-based method will reliably characterize the transcriptome when molecular crowding is not an issue.
Finally, in Chapter 3, we demonstrate the barcoding strategy to break through the molecular crowding limit of multiplexed FISH. From being able to profile hundreds to a thousand genes by various multiplexed FISH methods at that time in the field, we succeeded in profiling 10,000 genes by RNA seqFISH+, an evolved version of seqFISH, in various intact tissue sections, turning seqFISH+ into a spatial discovery technology with its genome-wide coverage and high detection efficiency. The work described in this part of the thesis is highlighted in Nature Method’s Method of The Year 2020- Spatially-resolved Transcriptomic article.</p
Geometric Langlands for Hypergeometric Sheaves
Get PDFGeneralized hypergeometric sheaves are rigid local systems on the punctured projective line with remarkable properties. In this paper, we construct the Hecke eigensheaves whose eigenvalues are the irreducible hypergeometric local systems in the setting of geometric Langlands program. We
work in the framework of rigid automorphic data that is mainly due to Zhiwei Yun. The key point is to choose a proper collection of level structures and compute the space of automorphic forms that are equivariant with respect to these level structures. For those hypergeometric sheaves with wild ramification, we also generalize the construction of Hecke eigensheaves to other classical groups. We study part of their eigenvalues in the de Rham setting by giving an alternative construction of their Hecke eigen D-module using quantization of Hitchin system.</p
Learning Patterns with Kernels and Learning Kernels from Patterns
Get PDFA major technique in learning involves the identification of patterns and their use to make predictions. In this work, we examine the symbiotic relationship between patterns and Gaussian process regression (GPR), which is mathematically equivalent to kernel interpolation. We introduce techniques where GPR can be used to learn patterns in denoising and mode (signal) decomposition. Additionally, we present the kernel flow (KF) algorithm which learns a kernels from patterns in the data with methodology inspired by cross validation. We further show how the KF algorithm can be applied to artificial neural networks (ANNs) to make improvements to learning patterns in images.
In our denoising and mode decomposition examples, we show how kernels can be constructed to estimate patterns that may be hidden due to data corruption. In other words, we demonstrate how to learn patterns with kernels. Donoho and Johnstone proposed a near-minimax method for reconstructing an unknown smooth function u from noisy data u + ζ by translating the empirical wavelet coefficients of u + ζ towards zero. We consider the situation where the prior information on the unknown function u may not be the regularity of u, but that of ℒu where ℒ is a linear operator, such as a partial differential equation (PDE) or a graph Laplacian. We show that a near-minimax approximation of u can be obtained by truncating the ℒ-gamblet (operator-adapted wavelet) coefficients of u + ζ. The recovery of u can be seen to be precisely a Gaussian conditioning of u + ζ on measurement functions with length scale dependent on the signal-to-noise ratio.
We next introduce kernel mode decomposition (KMD), which has been designed to learn the modes vi = ai(t)yi(θi(t)) of a (possibly noisy) signal Σivi when the amplitudes ai, instantaneous phases θi, and periodic waveforms yi may all be unknown. GPR with Gabor wavelet-inspired kernels is used to estimate ai, θi, and yi. We show near machine precision recovery under regularity and separation assumptions on the instantaneous amplitudes ai and frequencies ˙θi.
GPR and kernel interpolation require the selection of an appropriate kernel modeling the data. We present the KF algorithm, which is a numerical-approximation approach to this selection. The main principle the method utilizes is that a "good" kernel is able to make accurate predictions with small subsets of a training set. In this way, we learn a kernel from patterns. In image classification, we show that the learned kernels are able to classify accurately using only one training image per class and show signs of unsupervised learning. Furthermore, we introduce the combination of the KF algorithm with conventional neural-network training. This combination is able to train the intermediate-layer outputs of the network simultaneously with the final-layer output. We test the proposed method on Convolutional Neural Networks (CNNs) and Wide Residual Networks (WRNs) without alteration of their structure or their output classifier. We report reduced test errors, decreased generalization gaps, and increased robustness to distribution shift without significant increase in computational complexity relative to standard CNN and WRN training (with Drop Out and Batch Normalization).
As a whole, this work highlights the interplay between kernel techniques with pattern recognition and numerical approximation.</p
Mitochondrial Dynamics and Mitophagy during Male Germline Development
Get PDFMitochondrial fusion and fission (mitochondrial dynamics) and mitophagy are well-established mitochondrial quality control mechanisms that safeguard cellular homeostasis. However, their role during development remains poorly understood. In this thesis, we establish the role of mitochondrial dynamics and mitophagy during the development of the male germline (spermatogenesis).
Spermatogenesis is one of biology’s most complex and lengthy differentiation processes, transforming spermatogonial stem cells into highly specialized sperm cells capable of fertilization. This elaborate differentiation program requires multiple transitions in mitochondrial morphology and extensive degradation of mitochondria, making it an attractive model system for investigating mitochondrial dynamics and mitophagy in vivo. Indeed, the field of mitochondrial dynamics has a long history with spermatogenesis. The first mitochondrial dynamics gene, Fuzzy onions (Fzo), was discovered in 1997 to mediate mitochondrial fusion during Drosophila spermatogenesis. However, the role of mitochondrial dynamics during mammalian spermatogenesis remained unknown for nearly two decades after discovery of Fzo. To address this gap in knowledge, we investigate mitochondrial dynamics and mitophagy during mammalian spermatogenesis. We uncover essential roles for mitochondrial fusion (Chapter 2), mitochondrial fission (Chapter 3), and mitophagy (Chapter 4) during spermatogenesis and show that each of these mitochondrial quality control mechanisms regulates a distinct stage of germ cell development. Our analyses reveal requirements for mitochondrial fusion, fission, and mitophagy that correspond to the mitochondrial and metabolic needs of the developing germ cells.
We also investigate the role of mitochondrial fusion and fission in regulating subcellular mitochondrial domains upon fusion of a skeletal muscle stem cell with a myofiber (Chapter 5). Thus, the work presented in this thesis characterizes the in vivo role of mitochondrial dynamics in two systems: male germline development and skeletal muscle regeneration. However, we focus on the role of mitochondrial dynamics and mitophagy during male germline development.</p
High-Aspect Ratio Structures in Light-Absorbers and Electrocatalysts for Solar Fuels Devices
Get PDFSolar fuels devices produce hydrogen fuel from water and sunlight and address a critical societal need for inexpensive, long-duration energy storage. Such devices are prepared from combinations of light-absorbing semiconductors and catalysts to sunlight to drive thermodynamically uphill reactions. This dissertation puts forth strategies for controlling the three-dimensional structure of semiconductors, electrocatalysts, and the film of gas bubbles evolved on the top and bottom of a solar fuels device. High-aspect ratio features led to unexpected effects in semiconductor/electrocatalyst assemblies. Optical losses were decoupled from the mass-loading of cobalt phosphide and copper electrocatalysts integrated onto silicon microwire photocathodes for the photoelectrochemical generation of hydrogen and hydrocarbons, respectively. Anti-reflective silicon microcone arrays were patterned with continuous films of Pt or CoP particles with minimal reflection losses due to the catalyst films. Transparent metal films were prepared from nanostructured metal phosphides, a class of earth-abundant hydrogen evolution catalysts. Silicon microwire array (photo)electrode surfaces were used to force bubbles away from electrocatalyst surfaces, even when oriented against gravity, leading to sustained operation in the absence of external convection.</p