185 research outputs found
Spinodal decomposition in the inverse cascade of two-dimensional, binary-fluid turbulence
No full textWe study spinodal decomposition in the inverse-cascade regime of two dimensional turbulence in symmetric, binary fluid mixtures. We show that turbulence leads to break up of domains whose size, in the inverse cascade regime, is proportional to the Hinze scale. Even more strikingly, we show that the inverse cascade of energy is blocked by the formation of domains
Particles and Fields in Superfluids: Insights from the Two-dimensional Gross-Pitaevskii Equation
No full textWe study the dynamics of active particles in two-dimensional superfluids at temperature , for a variety of initial configurations, by carrying out extensive direct-numerical-simulations of the two-dimensional, Galerkin-truncated Gross-Pitaevskii equation. Our study elucidates the interplay of particles and fields, in both simple and turbulent flows. We show that particle collisions can be inelastic, if the repulsive interactions between particles is weak, and elastic otherwise. We show that assemblies of many particles and vortices yield turbulent spatiotemporal evolutions
Multifractal Droplet Dynamics in Two-Dimensional, binary-fluid turbulence
No full textWe present the most extensive direct numerical simulations, attempted so far, of statistically steady, homogeneous, isotropic turbulence in two-dimensional, binary-fluid mixtures with air-drag-induced friction. We model this mixture by using the Cahn-Hilliard-Navier-Stokes equations and choose parameters, e.g., the surface tension, such that we have a droplet of the minority phase moving inside a turbulent background of the majority phase. Our study reveals that a single droplet, whose mean radius lies in the inertial range of scales, (a) enhances the the forward-cascade part of the energy spectrum of two-dimensional turbulence and (b) stretches the tails of the PDF of the Okubo-Weiss parameter . We show that the dynamics of the droplet is affected significantly by the turbulence in the fluid. In particular, the PDFs of the components of the acceleration shows wide, non-Guassian tails. We characterize the time dependence of the deformation of the droplet and show that it exhibits multifractality
Real-space Manifestations of Bottlenecks in Turbulence Spectra
No full textAn energy-spectrum bottleneck, a bump in the turbulence spectrum between the inertial and dissipation ranges, is shown to occur in the non-turbulent, one-dimensional, hyperviscous Burgers equation and found to be the Fourier-space signature of oscillations in the real-space velocity, which are explained by boundary-layer-expansion techniques. Pseudospectral simulations are used to show that such oscillations occur in velocity correlation functions in one- and three-dimensional hyperviscous hydrodynamical equations that display genuine turbulence
Universal Statistical Properties of Inertial-particle Trajectories in Three-dimensional, Homogeneous, Isotropic, Fluid Turbulence
No full textWe obtain new universal statistical properties of heavy-particle trajectories in three-dimensional, statistically steady, homogeneous, and isotropic turbulent flows by direct numerical simulations. We show that the probability distribution functions (PDFs) P(Φ), of the angle Φ between the Eulerian velocity u and the particle velocity v, at a point and time, scales as P(Φ) ∼Φ−, with a new universal exponent ≃ 4
Mutual-Friction Coefficients in Two-Dimensional Superfluids: From the Gross-Pitaevskii equation to the Hall-Vinen-Bekharevich-Khalatnikov Two-fluid Model
No full textWe start from the two-dimensional Gross-Pitaevskii equation (GPE) and develop algorithms for the ab-initio determination of the temperature (T) dependence of the mutual-friction coefficients, α and α, and the normal-fluid density Pn, which appear as parameters in the Hall-Vinen-Bekharevich-Khalatnikov (HVBK) two-fluid model for a superfluid. In the second part of our study, we elucidate the statistical properties of two-dimensional, homogeneous, isotropic superfluid turbulence in the simplified HVBK model, with values for the mutual-friction coefficients that are comparable to those we obtain from the first part of our study
How does trade impact agricultural productivity?
No full textThe student, Akshay Pandit, submitted this Thesis for approval on 2020-07-22 at 15:29.This Thesis was approved for publication on 2020-07-23 at 10:50.DSpace SAF Submission Ingestion Package generated from Vireo submission #15729 on 2020-10-02 at 15:34:07Made available in DSpace on 2020-10-07T22:44:48Z (GMT). No. of bitstreams: 2
PANDIT-THESIS-2020.pdf: 10275210 bytes, checksum: bdf6f32a4714aaadf246aa27560ec60f (MD5)
LICENSE.txt: 4210 bytes, checksum: ad7b57595833966ecb91704e689e58e5 (MD5)
Previous issue date: 2020-07-23"Agricultural production has faced increased demands over the last half century from an expanding economy and population. We live in a globalized world, in which agriculture is deeply intertwined in international markets and trade. In this paper, we address the overarching research question: ""What is the impact of trade on agricultural productivity?''. To this end, we present a comprehensive statistical and econometric analysis on the relationship between international trade and agricultural production. We use national-scale data on crop yield, area harvested, production, and trade for the last half century (1961-2016) from the Food and Agricultural Organization of the United Nations. We introduce novel weighting and decomposition analyses to explore the relationship between trade and crop productivity. To determine the causal impact of trade on agriculture we implement instrumental variable (IV) econometric methods. We find that trade has led to an increase in global agricultural productivity over time (e.g. through increased productivity, the intensive margin). Global productivity gains have accrued primarily through the participation of more countries in global trade (e.g. expanding the area of contribution, the extensive margin). Additionally, we find that trade has enabled global crop consumption to increase. These findings indicate that trade openness leads to greater productivity in agriculture in general. This work highlights that trade can help to achieve productivity gains in agriculture and potentially help the world to address remaining yield gaps."Submission published under a 24 month embargo labeled 'U of I Access', the embargo will last until 2022-08-01The student, Akshay Pandit, accepted the attached license on 2020-07-22 at 15:28.Embargo set by: Seth Robbins for item 116267
Lift date: 2022-10-07T22:44:53Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemAuthor requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemU of I Onl
Scalable Continual Learning Using Cascading Hypernetworks and Cellular Automata
No full textAddressing the critical challenges of catastrophic forgetting and scalability is paramount for developing continual learning systems capable of robust operation in real-world environments. This dissertation presents a novel framework that achieves state-of-the-art performance by integrating two core innovations: a cascading hypernetwork architecture for adaptive model generation and a specialized mini-cellular automata (mini-CA) system for automatic task identification.
The proposed cascading hypernetworks employ a hierarchical structure, where parent hypernetworks can generate the weights of other hypernetworks and eventually task networks, enabling efficient learning across an expanding task set. This is coupled with an auto-generative replay mechanism that reconstructs network samples and replays them to the same hypernetwork, significantly mitigating catastrophic forgetting without reliance on large memory buffers.
For precise task identification, we introduce mini-CA, an efficient and scalable cellular automata variant tailored for complex spatio-temporal data. By utilizing smaller, specialized units with enhanced parallelism and input scalability, mini-CA surpasses the limitations of traditional CA. The system is adeptly extended to process colored images and sequences and integrates seamlessly with deep learning architectures for multi-channel input.
In the unified framework, mini-CA first identifies the incoming task, which then directs the cascading hypernetwork to dynamically generate a tailored, task-specific network. Rigorous evaluations on reinforcement learning and image classification benchmarks demonstrate superior accuracy, latency, and scalability. This research underscores the transformative potential of combining hierarchical hypernetworks with specialized CA to build truly adaptive and continuously learning systems for complex applications.Electrical and Computer Engineerin
Relational Neurogenesis for Lifelong Learning Agents
No full textReinforcement learning systems have shown tremendous potential in being able to model meritorious behavior in virtual agents and robots. The ability to learn through continuous reinforcement and interaction with an environment negates the requirement of painstakingly curated datasets and hand crafted features. However, the ability to learn multiple tasks in a sequential manner, referred to as lifelong or continual learning, remains unresolved.
The search for lifelong learning algorithms creates the foundation for this work. While there has been much research conducted in supervised learning domains under lifelong learning, the reinforced lifelong learning domain remains open for much exploration. Furthermore, current implementations either concentrate on preserving information in fixed capacity networks, or propose incrementally growing networks which randomly search through an unconstrained solution space.
In order to develop a comprehensive lifelong learning algorithm, it seems essential to amalgamate these approaches into a condensed algorithm which can perform both neuroevolution and constrict network growth automatically.
This thesis proposes a novel algorithm for continual learning using neurogenesis in reinforcement learning agents. It builds upon existing neuroevolutionary techniques, and incorporates several new mechanisms for limiting the memory resources while expanding neural network learning capacity. The algorithm is tested on a custom set of sequential virtual environments which emulate several meaningful scenarios for intellectually down-scaled species and autonomous robots.
Additionally, a library for connecting an unconstrained range of machine learning tools, in a variety of programming languages to the Unity3D simulation engine for the development of future learning algorithms and environments, is also proposed
- …
