12023 research outputs found
Sort by
Molecular Pattern Recognition and Supervised Learning in DNA-Based Neural Networks
Get PDFAdaptation in nature begins at the subcellular, molecular level with the delicate interplay of biomolecule cascades orchestrating the myriad function of cells. The intermingling activity of these cells becomes expressions of complex behavior of multi-cellular system. Nature provides a dazzling array of examples showcasing the variations of intelligent functions. However, in the realm of synthetic construction, what systems have humans managed to engineer, and what are the boundaries of our technological power? In comparison to nature's repertoire, mankind's accomplishments appear rather modest. The intricate behaviors observed in intelligent organisms emerge from the collective interactions and feedback loops among their constituent elements, resulting in the emergence of novel properties and phenomena. To develop large-scale engineered systems exhibiting ever more brain-like, intelligent behaviors, we must first devise new molecular architectures and algorithms designed for adaptation and learning at the molecular scale. My research presented here is a humble step toward those goals. I will present the design of novel molecular systems made from DNA that exhibit complex neural computation and learning behaviors.
Chapter 2 covers my contribution to scaling up the computing power of DNA circuits. From bacteria following simple chemical gradients to the brain distinguishing complex odor information, the ability to recognize molecular patterns is essential for biological organisms. This type of information-processing function has been implemented using DNA-based neural networks. Winner-take-all computation has been suggested as a potential strategy for enhancing the capability of DNA-based neural networks. Compared to the linear-threshold circuits and Hopfield networks used previously, winner-take-all circuits are computationally more powerful, allow simpler molecular implementation, and are not constrained by coupling the number of patterns and their complexity, so both a large number of simple patterns and a small number of complex patterns can be recognized. Here, we report a systematic implementation of winner-take-all neural networks based on DNA-strand-displacement reactions. We use a previously developed seesaw DNA gate motif, extended to include a simple and robust component that facilitates the cooperative hybridization involved in selecting a ‘winner.' We show that with this extended seesaw motif, DNA-based neural networks can classify patterns into up to nine categories. Each of these patterns consists of 20 distinct DNA molecules chosen from the set of 100 that represents the 100 bits in 10x10 patterns, with the 20 DNA molecules selected tracing one of the handwritten digits ‘1’ to ‘9.' The network successfully classified test patterns with up to 30 of the 100 bits flipped relative to the digit patterns ‘remembered’ during training, suggesting that molecular circuits can robustly accomplish the sophisticated task of classifying highly complex and noisy information on the basis of similarity to a memory.
Chapter 3 investigates the development of a computational neural network model inspired by biological learning mechanisms, particularly focusing on the new mechanisms for learning in a WTA neural network. The study incorporates novel molecular motifs used in inhibited activators and inhibited weights, designed specifically for training from environmental input patterns. These motifs emulate biological systems by facilitating memory storage and retrieval within DNA-based neural networks, similar to synaptic connections and signal processing observed in living organisms. We assess the function of the individual molecular motifs and characterize their specificity in up to 18-species cross-talk experiments. Furthermore, we characterize the network's performance across a wide array of training and test patterns, mirroring the adaptive responses and diverse conditions encountered by biological systems. Additionally, we analyze the computational efficiency and speed of the learning system, comparing it with both the previous non-learning DNA-based WTA model and a direct weight activation model. By exploring the principles of molecular learning, particularly within winner-take-all neural networks, this study aims to advance computational systems by emulating adaptability and resilience observed in biological organisms using robust, new molecular motifs.</p
Essays on Rational Social Learning
Get PDFThis dissertation contains three essays, each contributing to the study of social learning among rational agents in various contexts.
In Chapter 1, I study whether individuals can learn the informativeness of their information technology through social learning. Building on the classic sequential social learning model, I introduce the possibility that a common source is completely uninformative. I then define asymptotic learning as the situation in which an outsider, who observes the actions of all agents, eventually distinguishes between uninformative and informative sources. I show that asymptotic learning in this setting is not guaranteed; it depends crucially on the relative tail distributions of private beliefs induced by uninformative and informative signals. Furthermore, I identify the phenomenon of perpetual disagreement as the cause of learning and provide a characterization of learning in the canonical Gaussian environment.
In Chapter 2, co-authored with Omer Tamuz and Philipp Strack, we study the asymptotic rate at which the probability of a group of long-lived, rational agents in a social network taking the correct action converges to one. In every period, after observing the past actions of his neighbors, each agent receives a private signal, and chooses an action whose payoff depends only on the state. Since equilibrium actions depend on higher-order beliefs, characterizing agents' behavior becomes difficult. Nevertheless, we show that the rate of learning in any equilibrium is bounded from above by a constant, regardless of the size and shape of the network, the utility function, and the patience of the agents. This bound only depends on the private signal distribution.
In Chapter 3, I study how fads emerge from social learning in a changing environment. I consider a simple sequential learning model in which rational agents arrive in order, each acting only once, and the underlying unknown state is constantly evolving. Each agent receives a private signal, observes all past actions of others, and chooses an action to match the current state. Because the state changes over time, cascades cannot last forever, and actions also fluctuate. I show that despite the rise of temporary information cascades, in the long run, actions change more often than the state. This provides a theoretical foundation for faddish behavior in which people often change their actions more frequently than necessary.</p
Response of Earthquakes to Transient Stresses, in Laboratory and Nature
Get PDFEarthquake rates are known to fluctuate with time according to the changing state of stress in the Earth’s crust. Studying the response of earthquakes to transient stresses provides a unique insight into the mechanisms controlling the earthquake nucleation process. Common sources of transient stresses include stress changes from fault slip during large earthquakes, spontaneous slow fault slip, fluid pressure diffusion, seasonal changes of water mass and snowpacks related to hydrological cycles, tidal stresses from changes of gravitational forces of the Sun and the Moon, and anthropogenic fluid injection and extraction related to geoenergy production. In this
thesis, we first start in the laboratory-scale fault and conduct friction experiments to enhance our understanding of the underlying friction laws used for modeling earthquakes. We find that the traditional view of Coulomb friction, which postulates that there exists a threshold shear force called “static friction,” below which the frictional interface remains stationary, is incorrect. Our measurements have shown that such an interface is still sliding, albeit with extremely small decaying slip rates down to 10^{−12} m/s. This is consistent with a more recently developed friction law, which describes friction as dependent on slip rate and the state of the interface, e.g., time since the last earthquake. Next, we move beyond the laboratory and study natural faults. In one example, we study the response of earthquakes to transient stress induced by a spontaneous slow fault slip event that preceded the earthquake swarm
sequence by approximately half a day. In another example, we study the response of earthquakes to seasonal stress perturbations as a result of seasonal changes in groundwater mass and snowpack between wet and dry seasons, using California as a case study. In both examples, we find that earthquake nucleation is not an instantaneous process. Rather the earthquake rates lag after the stress rates. Such behavior cannot be described by Coulomb friction but can be quantitatively explained by the rate- and state-dependent friction. In the final example, we document bursts of fast propagating swarms of induced earthquakes at the Groningen gas field in the Netherlands. While transient stress must exist to drive the sequence, we cannot explicitly quantify the sources. Overall, our work provides key insights into the earthquake nucleation process, allowing us to better understand how to model the response of earthquakes to transient stress, including earthquakes that are induced by anthropogenic activities related to geoenergy production
Formal Methods for Test and Evaluation: Reasoning over Tests, Automated Test Synthesis, and System Diagnostics
Get PDFWith the integration of autonomous systems into our everyday lives edging closer to reality, ensuring the safety of these systems is paramount. Part of the safety verification process is a rigorous testing procedure, which currently does not exist for autonomous vehicles. In this thesis, we aim to provide approaches using formal methods to increase the efficiency of testing campaigns.
First, we provide a framework based on assume-guarantee contracts to specify tests in the form of a test structure. Using these test structures, we then show how to combine, split, and compare tests. Additionally, we characterize when tests can be combined and when the resulting test requires temporal constraints. Next, we demonstrate the approach on examples and find a strategy for a test agent using winning sets and Monte Carlo tree search.
Second, we present a framework to automatically synthesize a test environment, consisting of static and reactive obstacles, and dynamic test agents. We characterize the desired test behavior in a system and a test objective in the form of a linear temporal logic specification, consisting of sub-tasks commonly used for robotic missions. This test environment must ensure that the test is not impossible (i.e. a correct system can pass the test), but also that every test execution that satisfies the system objective also satisfies the test objective. We use tools from automata theory to construct the virtual product graph that represents all possible test executions, and the virtual system graph, which corresponds to the system's perspective.
We formulate this routing problem as a network flow optimization on the virtual product graph in the form of a mixed integer linear program for different test environments. We show that this routing problem is NP-hard. We propose a counterexample-guided search using GR(1) synthesis to find a strategy for a test agent. This framework is demonstrated in several examples in simulation and hardware.
Lastly, we present a framework to diagnose a system-level fault by identifying the component responsible for the failure. We make use of assume-guarantee contracts and Pacti, a tool for compositional system analysis and design, to construct a diagnostics map, which allows us to trace a system-level guarantee to possible causes. We show that this framework can reduce the number of statements that need to be checked in the diagnostics process. We illustrate this framework on several abstract examples and two examples inspired by a real-world autonomous system.</p
On Arithmetic Invariants of Special Families of K3-Type Surfaces
Get PDFThis thesis studies applications of Shimura varieties in positive characteristic to questions on arithmetic invariants of special families of K3-type surfaces.
The first main result determines the Newton polygons and Artin invariants of 144 special families of K3-type surfaces. The second is a refinement of a conjecture of Serre for K3 surfaces over number field.</p
On the Complexity of Neural Network Representations
Get PDFThe evolution of the human brain was one of the milestones in the history of information after the emergence of life. The underlying biological, chemical, and physical processes of the brain have amazed scientists for a long time. It is still a mystery how the human brain computes a simple arithmetical operation like 2 + 2 = 4. This enigma has spurred investigations into understanding the intrinsic architecture of the brain.
This thesis delves into two primary models for brain architecture: Feedforward Neural Networks and Nearest Neighbor (NN) Representations. Both models are treated under the hypothesis that our brain does not work with "large" numbers and expressive power is derived from connectivity. Thus, when examining a network or, more precisely, a single neuron model, we strive to minimize the bit resolution of weights, potentially increasing depth or circuit complexity.
For the NN representations, the memory is defined by a set of vectors in Rⁿ (that we call anchors), computation is performed by convergence from an input vector to a nearest neighbor anchor, and the output is a label associated with an anchor. Limited bit resolution in the anchor entries may result in an increase of the size of the NN representation.
In the digital age, computers universally employ the binary numeral system, ensuring the enduring relevance of Boolean functions. This study specifically explores the trade-off between resolution and size for the computation models for Boolean functions. It is established that "low resolution" models may require a polynomial or even an exponential increase in the size complexity of the "high resolution" model, potentially making the practical implementation infeasible. Building upon prior research, our goal is to optimize these blow-ups by narrowing the gaps between theoretical upper and lower bounds under various constraints. Additionally, we aim to establish connections between NN representations and neural network models by providing explicit NN representations for well-known Boolean functions in Circuit Complexity Theory.</p
Leveraging the ΦX174 Protein Antibiotic to Study MraY Structure, Function, and Regulation
Get PDFThe overuse of antibiotics has escalated the prevalence of bacterial resistance to existing treatments, posing a significant threat to global health. This rise in antimicrobial resistance (AMR) has spurred research into innovative therapeutic approaches. Among the most promising strategies is the use of viruses of bacteria for 'phage therapy'. This thesis delves into the interplay between antibacterial resistance and peptidoglycan biosynthesis, highlighting the pivotal role of the membrane protein MraY. We present the first structure of MraY from a pathogenic species, revealing its inhibition by the lysis protein from the bacteriophage ΦX174, protein E. Additionally, we analyze lipidic interactions with MraY, proposing a previously unexplored allosteric feedback mechanism for regulating its enzymatic activity. Building on these insights, we expand the application of protein E to non-native hosts, offering new avenues for the development of targeted antibiotic interventions. This work not only advances our understanding of the structural and functional dynamics of MraY but also paves the way for novel antibacterial strategies
New Long End-Associative Polymers for Mist Control in I. Aqueous Solutions and II. Hydrocarbon Solvents
Get PDFUltralong linear polymers are well known to be useful in a variety of applications such as mist control, drag reduction, and agricultural spray drift control. However, the application of ultralong linear polymers is limited by shear degradation of the ultralong polymer chains that occurs under strong flow conditions. To overcome the issue of shear degradation, our group previously designed long end-associative polymers that can self-assemble into megasupramolecules (Mw > 2000 kg/mol) in low polarity solvents like jet fuel. The previously developed long end associative polymers had polycyclooctadiene backbones (1,4 polybutadiene) with carboxylic acid or tertiary amine end groups that associated via hydrogen bonding. They were shear degradation resistant and used as mist-control agents that provided fire-protection to jet fuel at concentrations as low as 0.3wt%.
Building upon the previous work, this thesis describes efforts toward expanding the applicability of long end-associative polymers. We first describe the design and synthesis of water-soluble long end-associative polymers for agricultural spray mist-control. We synthesized telechelic polyacrylamides using reversible addition fragmentation chain transfer (RAFT) polymerization. We explored two types of associations that can form stable supramolecules in water: host-guest interactions between adamantane and beta-cyclodextrin (Chapter 1), and metal-ligand association between terpyridine and transition metal ions (Chapter 2). Careful optimization of polymerization conditions allowed the synthesis of ultra-high molecular weight telechelic polyacrylamides with narrow polydispersity. We found that the terpyridine functionalized polyacrylamides with Mw 820 kg/mol could assemble into megasupramolecules (Mw > 2000 kg/mol) upon addition of Fe(II) or Ni(II).
In Chapters 3 and 4, we return to long end-associative polymers in hydrocarbon solvents. In Chapter 3, we tackle the issue of poor solubility of the previous generation of associative PCODs in a highly nonpolar solvent, PAO. In Chapter 4, we develop a new generation of end groups for improved solubility and ease of synthesis.
Finally in Chapter 5, we discuss polyDODT (poly(3,6-dioxa-1,8-octanedithiol)), a polydisulfide that tends to form a mixture of linear and cyclic species. We present new methods to detect the presence of linear species in the mixture, as well as an alternative synthesis route for synthesis of polyDODT.</p
Degradation of Ceramic Surfaces and its Mitigation: From Electric Propulsion to Cultural Heritage
Get PDFCeramics have played an evolving role throughout human history, with the earliest known fired clay figurines dating back to 29,000-25,000 BCE in what is today the Czech Republic. Such examples of porous, low-fired potteries transitioned into vitreous and heat-resistant porcelain with the advancements of kiln technologies, which steadily grew to reach temperatures up to 1270 °C by the time of the Song Dynasty (960-1279 CE). In the modern-age, furnaces technologies have become exceedingly more sophisticated, reaching ultra-high temperatures of up to 3000 °C with dedicated calibration systems. With the advancement of firing technologies came the advancement of materials processing methods, which have transformed the role of ceramics in everyday applications: from ceramic fibers used in tennis racquets, to ceramics used in space shuttle tiles and even in artificial joints. Indeed, ceramics are candidate solutions to the most stringent material problems faced today, including high-temperature applications like electric propulsion and turbine engine systems. However, in order to improve the long-term use and sustainability of ceramics, it is essential to evaluate both their performance and their eventual degradation from mechanical wear and chemical erosion.
This work explores the processing-microstructure-performance relationship of ceramics to better understand the performance and degradation mechanisms of ceramic surfaces. This relationship is investigated using a series of material case studies, including (i) advanced high-temperature ceramics composed of h-BN rich composites, and (ii) historic ceramics, ranging from low-fired pottery to porcelain. Details are provided for the design and manufacturing of novel high-performing ceramics, while simultaneously referring to ceramics of the past to understand how their surfaces have altered over their lifetime. Highlights of this work include the innovative use of carbothermic reactions to create h-BN surface layers for electric propulsion; self-healing strategies from AlN/BN composites; DFT-supported analysis of pottery corrosion in acidic soils; and nanoscale processing of historic porcelain glazes. These analyses provide us with an opportunity to learn from materials of the past to create more sustainable materials for the future, with an emphasis on ways of mitigating degradation by controlling processing conditions and environmental exposures.</p
Experimental Studies of Flow Control Techniques for Future Aircraft
Get PDFFrom the signing of the Paris Agreement to the COVID-19 outbreak, the past decade has truly challenged the aviation industry to adapt. New technologies need to be developed constantly to meet the increasing commercial and defense demands for more efficient, quiet, safe, and agile aircraft. To keep up with these rapidly changing times, an approach that marries a fundamental understanding of aerodynamics with systems design and optimization is necessary. This thesis explores two promising concepts for controlling flow over next-generation aircraft: active control on a swept wing for airplane applications, and passive control on a rotating blade for drone applications. In each, force measurements are combined with advanced flow visualization techniques to create a research framework that is both data-driven and physics-informed.
In Part I, a comprehensive wind tunnel campaign is carried out on a swept wing model of modular geometry equipped with an array of sweeping jet actuators, which have demonstrated tremendous promise for flow control authority in both laboratory settings and full-scale flight tests. The flow physics and performance of the wing is investigated first without actuation, revealing separation behaviors at both the leading and trailing edges that are crucial to consider when flow control is applied. This paves the way for an optimization study in a newly proposed framework that relies on fluid power coefficients rather than the momentum coefficient that has been the accepted parameter of choice for characterizing blowing systems over the past seven decades of active flow control research.
Part II explores the feasibility of a "prop-shroud" concept for small-scale aerial vehicles, in which the shroud is directly attached to the blade tips and thus co-rotates with the propeller. Such a configuration has the potential to provide the various aerodynamic and engineering benefits of a shrouded propeller without the associated costs and complexities of its installation. The hover efficiency of a prop-shroud is shown to be comparable to commercially available drone propellers, even without a rigorous optimization of its geometry. The effect of the co-rotating shroud is then analyzed in detail on the time-averaged, phase-averaged, and unsteady features of the flow field. A model based on vortex formation time is developed, laying out a foundation for future research and understanding.</p