Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “computational complexity”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 559 records · Page 31
All sourcesNASA NTRSDOE OSTI
DOE OSTI2013

HPC Colony II: FAST_OS II: Operating Systems and Runtime Systems at Extreme Scale

HPC Colony II has been a 36-month project focused on providing portable performance for leadership class machines—a task made difficult by the emerging variety of more complex computer architectures. The project attempts to move the burden of portable performance to adaptive system software, thereby allowing domain scientists to concentrate on their field rather than the fine details of a new leadership class machine. To accomplish our goals, we focused on adding intelligence into the system software stack. Our revised components include: new techniques to address OS jitter; new techniques to dynamically address load imbalances; new techniques to map resources according to architectural subtleties and application dynamic behavior; new techniques to dramatically improve the performance of checkpoint-restart; and new techniques to address membership service issues at scale.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2015

Final Technical Report: Distributed Controls for High Penetrations of Renewables

The goal of this effort was to apply four potential control analysis/design approaches to the design of distributed grid control systems to address the impact of latency and communications uncertainty with high penetrations of photovoltaic (PV) generation. The four techniques considered were: optimal fixed structure control; Nyquist stability criterion; vector Lyapunov analysis; and Hamiltonian design methods. A reduced order model of the Western Electricity Coordinating Council (WECC) developed for the Matlab Power Systems Toolbox (PST) was employed for the study, as well as representative smaller systems (e.g., a two-area, three-area, and four-area power system). Excellent results were obtained with the optimal fixed structure approach, and the methodology we developed was published in a journal article. This approach is promising because it offers a method for designing optimal control systems with the feedback signals available from Phasor Measurement Unit (PMU) data as opposed to full state feedback or the design of an observer. The Nyquist approach inherently handles time delay and incorporates performance guarantees (e.g., gain and phase margin). We developed a technique that works for moderate sized systems, but the approach does not scale well to extremely large system because of computational complexity. The vector Lyapunov approach was applied to a two area model to demonstrate the utility for modeling communications uncertainty. Application to large power systems requires a method to automatically expand/contract the state space and partition the system so that communications uncertainty can be considered. The Hamiltonian Surface Shaping and Power Flow Control (HSSPFC) design methodology was selected to investigate grid systems for energy storage requirements to support high penetration of variable or stochastic generation (such as wind and PV) and loads. This method was applied to several small system models.

14 SOLAR ENERGY↗
DOE OSTI2017

Analyst-to-Analyst Variability in Simulation-Based Prediction

This report describes findings from the culminating experiment of the LDRD project entitled, "Analyst-to-Analyst Variability in Simulation-Based Prediction". For this experiment, volunteer participants solving a given test problem in engineering and statistics were interviewed at different points in their solution process. These interviews are used to trace differing solutions to differing solution processes, and differing processes to differences in reasoning, assumptions, and judgments. The issue that the experiment was designed to illuminate -- our paucity of understanding of the ways in which humans themselves have an impact on predictions derived from complex computational simulations -- is a challenging and open one. Although solution of the test problem by analyst participants in this experiment has taken much more time than originally anticipated, and is continuing past the end of this LDRD, this project has provided a rare opportunity to explore analyst-to-analyst variability in significant depth, from which we derive evidence-based insights to guide further explorations in this important area.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2017

Spatio-Temporal Data Analysis at Scale Using Models Based on Gaussian Processes

Gaussian processes are the most commonly used statistical model for spatial and spatio-temporal processes that vary continuously. They are broadly applicable in the physical sciences and engineering and are also frequently used to approximate the output of complex computer models, deterministic or stochastic. We undertook research related to theory, computation, and applications of Gaussian processes as well as some work on estimating extremes of distributions for which a Gaussian process assumption might be inappropriate. Our theoretical contributions include the development of new classes of spatial-temporal covariance functions with desirable properties and new results showing that certain covariance models lead to predictions with undesirable properties. To understand how Gaussian process models behave when applied to deterministic computer models, we derived what we believe to be the first significant results on the large sample properties of estimators of parameters of Gaussian processes when the actual process is a simple deterministic function. Finally, we investigated some theoretical issues related to maxima of observations with varying upper bounds and found that, depending on the circumstances, standard large sample results for maxima may or may not hold. Our computational innovations include methods for analyzing large spatial datasets when observations fall on a partially observed grid and methods for estimating parameters of a Gaussian process model from observations taken by a polar-orbiting satellite. In our application of Gaussian process models to deterministic computer experiments, we carried out some matrix computations that would have been infeasible using even extended precision arithmetic by focusing on special cases in which all elements of the matrices under study are rational and using exact arithmetic. The applications we studied include total column ozone as measured from a polar-orbiting satellite, sea surface temperatures over the Pacific Ocean, and annual temperature extremes at a site in New York City. In each of these applications, our theoretical and computational innovations were directly motivated by the challenges posed by analyzing these and similar types of data.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2017

Regression with Small Data Sets: A Case Study using Code Surrogates in Additive Manufacturing

There has been an increasing interest in recent years in the mining of massive data sets whose sizes are measured in terabytes. While it is easy to collect such large data sets in some application domains, there are others where collecting even a single data point can be very expensive, so the resulting data sets have only tens or hundreds of samples. For example, when complex computer simulations are used to understand a scientific phenomenon, we want to run the simulation for many different values of the input parameters and analyze the resulting output. The data set relating the simulation inputs and outputs is typically quite small, especially when each run of the simulation is expensive. However, regression techniques can still be used on such data sets to build an inexpensive \surrogate" that could provide an approximate output for a given set of inputs. A good surrogate can be very useful in sensitivity analysis, uncertainty analysis, and in designing experiments. In this paper, we compare different regression techniques to determine how well they predict melt-pool characteristics in the problem domain of additive manufacturing. Our analysis indicates that some of the commonly used regression methods do perform quite well even on small data sets.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2017

LAFD: TA-3 NISC & SCC Facility Familiarization Tour, OJT #53356

Los Alamos National Laboratory (LANL, the Laboratory, or the Lab) conducts familiarization tours for Los Alamos County Fire Department (LAFD) personnel at the Strategic Computing Complex (SCC), TA-03-2327, and the Nonproliferation & International Security Complex (NISC), TA-03-2322. These tours are official LANL business; their purpose is to orient the firefighters to the SCC and the NISC so that they can respond efficiently and quickly to a variety of emergency situations. The tour includes ingress and egress of the buildings, layout and organization of the buildings, evacuation procedures, and areas of concern to emergency responders within these buildings. LAFD firefighters have the training, skills, and abilities to perform these emergency responder tasks; other LANL personnel who have the required clearance level cannot perform these tasks. This handout provides details of the information, along with maps and diagrams, to be presented during the familiarization tours. The report will be distributed to the trainees at the time of the tour. A corresponding checklist will also be used as guidance during the familiarization tours to ensure that all required information is presented to the LAFD personnel.

99 GENERAL AND MISCELLANEOUS↗
DOE OSTI2017

The Nature of Scatter at the DARHT Facility and Suggestions for Improved Modeling of DARHT Facility

The U.S. Stockpile Stewardship Program [1] is designed to sustain and evaluate the nuclear weapons stockpile while foregoing underground nuclear tests. The maintenance of a smaller, aging U.S. nuclear weapons stockpile without underground testing requires complex computer calculations [14]. These calculations in turn need to be verified and benchmarked [14]. A wide range of research facilities have been used to test and evaluate nuclear weapons while respecting the Comprehensive Nuclear Test-Ban Treaty (CTBT) [2]. Some of these facilities include the National Ignition Facility (NIF) at Lawrence Livermore National Laboratory, the Z machine at Sandia National Laboratories, and the Dual Axis Radiographic Hydrodynamic Test (DARHT) facility at Los Alamos National Laboratory. This research will focus largely on DARHT (although some information from Cygnus and the Los Alamos Microtron may be used in this research) by modeling it and comparing to experimental data. DARHT is an electron accelerator that employs high-energy flash x-ray sources for imaging hydro-tests. This research proposes to address some of the issues crucial to understanding DARHT Axis II and the analysis of the radiographic images produced. Primarily, the nature of scatter at DARHT will be modeled and verified with experimental data. It will then be shown that certain design decisions can be made to optimize the scatter field for hydrotest experiments. Spectral effects will be briefly explored to determine if there is any considerable effect on the density reconstruction caused by changes in the energy spectrum caused by target changes. Finally, a generalized scatter model will be made using results from MCNP that can be convolved with the direct transmission of an object to simulate the scatter of that object at the detector plane. The region in which with this scatter model is appropriate will be explored.

45 MILITARY TECHNOLOGY, WEAPONRY, AND NATIONAL DEF↗
DOE OSTI2018

Charge of the massive leptons

A simple picture of the massive leptons emerges by invoking a πλ 2 cross section for vacuum virtual photons to stimulate an isolated lepton to emit an additional virtual photon. This reaction mechanism leads to the storage of a lepton’s energy and momentum in the surrounding cloud of virtual photons. The stimulated-emission process allows for the exchange of virtual photons between a pair of leptons, and thus an estimate the fine-structure constant. A simple sharp low-energy cutoff for the exchanged photons, associated with near-field effects, results in a calculated charge of 1.65 x 10 –19 C. The discrepancy with experiment might be reduced by including additional computational complexity.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2019

Coordinated PEV Charging for Distribution System Management

Plug-in electric vehicles (PEVs) can help to reduce worldwide dependence on petroleum and carbon emissions. In the past a few years, PEVs have received considerable attention as an eco-friendly and cost-effective alternative to conventional gasoline vehicles. PEVs consume higher power from the grid during charging compared to conventional residential loads. Therefore, the emerging fleet of PEVs will introduce a considerable amount of additional load on power system. On the other hand, majority of PEVs are parked for more than 90% of the time, making them ideal for providing various services through Electric Vehicle Supply Equipment (EVSE), a.k.a. charging stations which are PEVs’ connection points to the power system. In this project, coordinated PEV charging are developed to provide distribution system services, considering PEV characteristics, realistic travel pattern, charging behaviors, and EVSE power rating and availability. The main efforts and contributions are summarized as follows. • Mobility models and EVSE charging capability/availability are indispensable for evaluating PEV charging coordination strategies and developing PEV charging control. Many existing studies utilize simplified mobility models assuming that the entire vehicle fleet returns home in the evening and is parked at home until the next morning. Some other PEV models better represent diversified home arrival and departure time, but cannot capture varying charging flexibility and capability at different locations, and therefore are not appropriate to study the impacts of public EVSE on PEV load and utilization. In this work, a mobility and charging flexibility model is proposed to better represent the temporal availability and varying charging capability from PEV onboard charger and EVSE. • Based on the mobility model, a generalized optimization method is proposed to evaluate different PEV charging coordination strategies. In existing literature, different algorithms and methods need to be designed to evaluate each charging strategy from both vehicle owners’ and system’s perspective. With the optimization method proposed in this work, different charging control strategies can be studied and compared by only updating objective functions. Moreover, optimization tricks are provided to convert the optimization problems to equivalent linear programming problems, which can be efficiently solved with existing solvers. • An innovative scheduling and control framework is proposed to enable smart PEV charging for grid services while meeting PEV owners’ travel needs. In the proposed framework, each set of EVSE is equipped with a controller that estimates charging power and energy flexibility based on vehicle characteristics, EVSE power rating, battery energy state, and upcoming trip information. With the simplified charging flexibility model received from each EVSE controller, the central coordinator determines the optimal power allocation for a look-ahead time window for the given grid services. The proposed charging control can help reduce the computational complexity and communication requirement compared with existing methods. It is also scalable to the expanding PEV fleet and robust to uncertainties in upcoming vehicle trips and future system condition. • The proposed charging strategy evaluation and smart charging control methods are applied to one of the prototypical feeders developed by the Pacific Northwest National Laboratory (PNNL) for case studies. Detailed trip information extracted from the National Household Travel Survey (NHTS) is used to represent travel and parking patterns. Different scenarios are simulated to evaluate the performance of the proposed methods and understand how PEV mobility and public EVSE availability affect PEV load and potential grid services.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2019

A Themodynamically Consistent Fractional Visco-Elasto-Plastic Model with Memory-Dependent Damage for Anomalous Materials

We develop a thermodynamically consistent, fractional visco-elastoplastic model coupled with damage for anomalous materials. The model utilizes Scott-Blair rheological elements for both visco- elastic/plastic parts. The constitutive equations are obtained through Helmholtz free-energy potentials for Scott-Blair elements, together with a memory-dependent fractional yield function and dissipation inequalities. A memory-dependent Lemaitre-type damage is introduced through fractional damage energy release rates. For time-fractional integration of the resulting nonlinear system of equations, we develop a first-order semi-implicit fractional return-mapping algorithm. We also develop a finite-difference discretization for the fractional damage energy release rate, which results into Hankel-type matrix-vector operations for each time-step, allowing us to reduce the computational complexity from O(N 3 ) to O(N 2 ) through the use of Fast Fourier Transforms. Our numerical results demonstrate that the fractional orders for visco-elasto-plasticity play a crucial role in damage evolution, due to the competition between the anomalous plastic slip and bulk damage energy release rates.

36 MATERIALS SCIENCE↗
DOE OSTI2021

Understanding Transients: Needs from the Laboratory Astrophysics Community

A growing number of astrophysical transients are pushing astronomers to develop increasingly complex computational tools to model both radiation hydrodynamics and electron transport. Here we review the physics needs and simulation uncertainties associated with modeling these transients. Although this review will focus on theory and simulation aspects of this problem, we also review some experiments designed to study this physics.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2021

IMS Rapid Response FY21 Summary Report for: Integrating Patterned Probes with Four-Dimensional Scanning Transmission Electron Microscopy for Unrivaled Crystallographic Structure Determination in Nanomaterials

The initial goal of our 4-dimensional scanning transmission electron microscopy (4D-STEM)-based project was to develop strain resolution two orders of magnitude better than what is now currently possible with electron-based scattering techniques, all while collecting scattering information from 7 different tilt axes at one time [multi-beam electron diffraction (MBED)1 ] through the development of a new electron probe-forming aperture with non-circular features (patterned probes 2 ). We set out to accomplish this through a collaboration with Drs. Colin Ophus and Ben Savitsky at Lawrence Berkeley Laboratory (they are the world-leading experts in developing the complex computational codes required to perform orientation analysis and quantitative strain mapping on our 4D-STEM data sets. We are motivated to invest in this area as it will be the only technique sensitive enough to perform three- dimensional automated crystallographic orientation mapping (ACOM) and strain mapping for materials exposed to external stimulus (a focus of our larger efforts).

36 MATERIALS SCIENCE↗
DOE OSTI2021

AGGREGATE: dAta-driven modelinG preservinG contRollable dEr for outaGe mAnagemenT and rEsiliency (Final Report)

The AGGREGATE project team successfully developed and validated various modules for outage management. Brief summaries of each module are provided to showcase their strength for outage management and restoration for a distribution system with a high penetration of connected distribution energy resources (DERs). In recent years, inverter-based DERs have been widely deployed in distribution system. A most of behind-the-meter (BTM) solar power generation is not visible to the utility. The data-driven DER and load estimation modules are using machine learning (ML) and artificial intelligence (AI) to manage this issue, which provides an opportunity for distribution system operators (DSOs) to operate systems and make decisions in real-time for a distribution system with a high penetration of DERs deployed. Also, the estimated DER and true load can be further leveraged in network aggregation and cold-load pick up estimation for reducing the computing complexity and providing for fast restoration. After load demand and DER power generations have been estimated, the information will support topology and state estimation (SE). The topology estimation module demonstrated the viability of mixed integer linear programming (MILP) formulation to estimate the most likely operational radial topology and outage sections using power flow measurements, historical/estimated load and DERs data and smart meter ping measurements. Formulation includes continuous (power flow, load and DERs data) and binary measurements (smart meter ping measurements) in a single formulation. Errors in continuous data and binary data are modeled as normal distribution and Bernoulli distribution, respectively. In the future distribution grid, the power injection from controllable DERs will be essential for efficient and resilient grid operation. However, determining the optimal DER injections and restoration actions is dependent on knowledge of the system states. State estimation (SE), already the cornerstone of transmission energy management systems, will become commonplace in distribution management systems as more measurements become available from deployment of automated metering infrastructure (AMI). Observability analysis is the first step in SE, as it determines the sufficiency of the available measurements for accurately estimating the current system states. A new type of pseudo-measurement called a Correlational Measurement (CM) is introduced in this module, to enhance the observability of the system to enable more accurate SE. CMs encapsulate knowledge of correlation between demand patterns for similar classes of loads as well as injection patterns for same-technology renewable DERs. During grid contingency scenarios, DERs have been traditionally disconnected, without any fault ride-through capabilities. However, with new regulations and better technology, it is feasible for these resources to contribute to the grid’s restoration after an adverse event and hence enhance resilience. The controllability module proposes a two-step restoration scheme for the power system restoration process by leveraging additional degrees of freedom in power electronics interfaced DERs for mitigating voltage problems. In a resilience mode without the utility system, the distribution grid relies on DERs to serve critical load. In such a severe event with multiple faults on the distribution feeders, actuation of various protective devices (PDs) divides the distribution system into electrical islands. The undetected actuated PDs due to fault current contributions from DERs can delay the restoration process, thereby reducing the system resilience. The Advanced Outage Management (AOM) and the Advanced Feeder Restoration (AFR) modules developed in this project provide improved system resilience with multiple DERs. AOM identifies the faulted sections and actuated PDs in a distribution system with DERs by incorporating smart meter data. The most credible outage scenario including fault locations, PD actuations, and fault indicator (FI) failures is identified by a set of binary integer linear programming incorporating hypotheses. The AFR module serves to restore a distribution system with available energy resources taking into consideration the availability of utility sources and DERs. By partitioning the system into islands, critical load will be served with the available generation resources within islands based on the solution of a MILP. When the utility systems become available, the optimal path will be determined by a spanning tree search algorithm that reconnects these islands back to substations and restores the remaining load. The transmission and distribution (T&D) co-simulation module was used to validate the effect of a control action performed on the distribution side assets as it propagates to the transmission side. This ensures that the control action performed results in a feasible operating point on both the transmission and the distribution system. In addition to validation, the team used the T&D co-simulation module to demonstrate how distribution system assets can be used to mitigate issues on the transmission system. Specifically, the team demonstrated that appropriate switching operations on the distribution side can alleviate the line overload condition on the transmission side without causing new operational constraint violations.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2022

Influence of Hybridization on the Capacity Value of PV and Battery Resources

Utility-scale systems that combine solar photovoltaic and battery (PV+battery) technologies are growing in popularity on the U.S. bulk power system. The business case for PV+battery systems depends on both their ability to reduce costs and their ability to generate value synergies associated with the provision of energy, capacity, and ancillary services. Capacity value can constitute a significant portion of the value PV+battery hybrids provide to the grid (e.g., through avoided or deferred capacity) and receive through revenues. Throughout this report, we define capacity value as the monetary value of a plant's contribution towards the planning reserve margin, which ultimately depends on market rules and structures. PV+battery hybrids do not always fit into current market structures because of the interactions between the PV and battery components. Unique considerations for the capacity value of PV+battery hybrids include the disparate nature of participation models for PV and battery technologies in existing market rules and the potential influence of a shared interconnection capacity; limitations imposed by a shared inverter; limited ability to charge the battery in advance of capacity events if charging must be sourced from the coupled PV; and challenges or uncertainties associated with co-optimizing the operations of the PV and battery components. Grid operators are currently considering how market structures can be modified to optimally determine the capacity value provided by PV+battery systems, and the rules of how they are integrated into markets are still being written. As with any resource, poorly designed rules could increase the cost of energy and reduce system reliability, while well-designed rules could allow markets to receive the full benefits hybrid systems can offer without overcompensating them for the services they provide. Well-designed rules for PV+battery systems must consider the unique aspects listed above, while leveraging the commonalities with existing resource types. In this report, we summarize the technical capability and market rules that influence the capacity value of PV+battery systems. We further discuss the potential tradeoffs between computational complexity and accuracy for the various ways in which grid operators can credit PV+battery systems for capacity. Finally, we describe markets for capacity, survey current wholesale market rules applying to PV+battery systems, and provide a snapshot of the current regulatory landscape for PV+battery systems.

14 SOLAR ENERGY↗
DOE OSTI2020

Multi-Stage and Multi-Timescale Robust Co-Optimization Planning for Reliable and Sustainable Power Systems. Final Report

In this project, Clarkson University, in collaboration with Southern Methodist University and University of Pittsburgh, conducted the research to model, design, and implement a sophisticated generation and transmission co-optimization planning decision tool, called Multi-stage and Multi-timescale robust Co-Optimization Planning (MMCOP). The MMCOP decision tool intends to facilitate generation and transmission co-optimization planning of emerging power systems, while mitigating risks and uncertainties in both short-term operation dynamics and long-term policy and technology changes. Long-term power system planning aims at optimizing asset utilization by investing in a proper mix of various generation technologies and transmission lines to supply the future load growth. In particular, the Clean Power Plan (CPP), which is designed to combat climate change and reduce carbon emissions by setting a national limit on carbon pollution from power plants, may dramatically change the landscape of the power industry by further promoting clean energy and phasing out emissions-intensive generation technologies. In addition, novel non-wire alternatives (e.g., demand response (DR), distributed generation (DG), energy efficiency (EE), and smart grid technologies) and the computational complexity for large-scale systems significantly complicate the system planning procedure even further. However, existing conventional planning approaches neglect short-term variability and uncertainty of renewable energy, hourly chronological operation details, and physical nonlinear characteristics of the alternating current transmission network. As a result, existing conventional planning approaches may not work properly, and power systems reliability could be in jeopardy. In observing the limitations of existing conventional planning approaches and addressing new challenges of emerging power systems, the main scope of this project to develop co-optimization planning models within a multi-stage and multi-timescale framework. In particular, random contingencies, key uncertainty factors, and AC power flows are included to derive expansion plans while considering both long-term reliability and short-term flexibility.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2022

Resilience Enhancements through Deep Learning Yields

This report documents the Resilience Enhancements through Deep Learning Yields (REDLY) project, a three-year effort to improve electrical grid resilience by developing scalable methods for system operators to protect the grid against threats leading to interrupted service or physical damage. The computational complexity and uncertain nature of current real-world contingency analysis presents significant barriers to automated, real-time monitoring. While there has been a significant push to explore the use of accurate, high-performance machine learning (ML) model surrogates to address this gap, their reliability is unclear when deployed in high-consequence applications such as power grid systems. Contemporary optimization techniques used to validate surrogate performance can exploit ML model prediction errors, which necessitates the verification of worst-case performance for the models.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2023

Extending Parsimonious Bayesian Inference

Parsimonious Bayesian inference is a theoretical framework for efficient data assimilation that seeks to balance increased consistency between predictions and training data against corresponding increases in model complexity. Within this framework, over-training is understood as optimization that encodes excessive information within model parameters while only achieving small improvements between predictions and training data. This project aims to develop practical methods of limiting excess model information during optimization. One key observation is that practical heuristics for parsimonious learning in high-dimensions must balance expressivity, i.e. the ability of the model to capture diverse predictions with only a few non-zero parameters, against discoverability, i.e. the ability to train the model with gradient-based optimization and drive parameters to low information states. As such, we developed logical activation functions that are able to adaptively approximate arbitrary truth tables that define Boolean logic operations within a probabilistic framework. These functions have demonstrated the ability to learn exclusive disjunction (XOR) and conditioned disjunction (if [condition] then [result_if_true] else [result_if_false]) within a single layer of a neural network. To efficiently exploit these activation functions to drive parsimonious learning required several other advances within the domain of variational inference. The most efficient form of complexity suppression is structured sparsification, driving most model parameters to zero while achieving the structural coherence among nonzeros needed for bandwidth reduction. Such models are not only far more efficient at suppressing information-theoretic complexity, they also reduce the other forms of complexity (computations, communication, storage, and the number of dependencies needed to evaluate predictions). Aiming to support enhanced sparsification, this project examined new approaches to high-dimensional variational inference that allow us to calibrate and control parameter uncertainty during optimization. By identifying which parameters can sustain sparsifying perturbations with little impact on prediction quality, we can develop better pruning strategies by framing them as approximate Bayesian inference. These advances also open paths to mitigate concerns with deploying advanced learning methods in resource-constrained environments, such as running models on power-limited or communication-limited devices.

97 MATHEMATICS AND COMPUTING↗