Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “hierarchy”

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 163 records · Page 9
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

On ParELAG's Parallel Element-based Algebraic Multigrid and its MFEM Miniapps for H(curl) and H(div) Problems: a report including lowest and next to the lowest order numerical results

This paper presents the utilization of element-based algebraic multigrid (AMGe) hierarchies, implemented in the ParELAG (Parallel Element Agglomeration Algebraic Multigrid Upscaling and Solvers) library, to produce multilevel preconditioners and solvers for H(curl) and H(div) formulations. This involves the construction of hierarchies of compatible nested spaces, forming an exact de Rham sequence on each level. This allows the application of hybrid smoothers on all levels and AMS (Auxiliary-space Maxwell Solver) or ADS (Auxiliary-space Divergence Solver) on the coarsest levels, obtaining complete multigrid cycles. Numerical results are presented, showing the parallel performance of the proposed methods. As a part of the exposition, this paper demonstrates some of the capabilities of ParELAG and outlines some of the components and procedures within the library.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Milestone M7 Report: Reducing Excess Data Movement Part 2

This milestone evaluates techniques to measure and, if possible, reduce data movement across all levels of the memory hierarchy, focusing on CPU/GPU page level data movement and on intra-GPU memory hierarchy. We quantitatively evaluate the efficacy of the techniques in reducing data movement and measure how performance tracks data movement reduction. We study a small collection of benchmarks and proxy mini-apps that run on advanced pre-exascale GPUs and on the Accelsim GPU simulator. Our approach has two thrusts: to measure advanced data movement reduction directives and techniques on the newest available GPUs, and to evaluate our benchmark set on simulated GPUs configured with architectural refinements to reduce data movement. We primarily evaluated NVidia-based architectures due to the unavailability of AMD GPU hardware and tools until very recently.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Universal Utility Data Exchange (UUDEX) – Security and Administration: Cybersecurity of Energy Delivery Systems (CEDS) Research and Development

A critical component of the Universal Utility Data Exchange (UUDEX) approach is the integrated security contained within its processing. This document describes how that security is designed and expected to be implemented by UUDEX Implementations (U-Implementations), including the UUDEX Server (U-Server) and UUDEX Clients (U-Clients). The UUDEX security hierarchy consists of three levels: 1. The UUDEX Instance (U-Instance) itself, which sits at the top of the hierarchy and contains the U-Server, the UUDEX Identity Authority (U-Identity Authority), and the UUDEX Administrator (U-Administrator) functions; 2. A group of one or more UUDEX Participants (U-Participants) that present “organizations” that participate in the U-Instance and contains the UUDEX Administrator Participant (U-U-Administrator Participant) function; 3. A group of one or more UUDEX Endpoints (U-Endpoints) that represent the individual UUDEX Publish Clients (U-Publish Client) responsible for supplying data to the U-Instance that is consumed by UUDEX Subscriber Clients (U-Subscriber Clients). U-Endpoints can be either autonomous devices that publish and subscribe data such as data exchange servers found in supervisory control and data acquisition and energy management systems, or they can be tied to users of applications that, for example, submit DOE OE-417 disturbance reports. U-Participants and U-Endpoints can be organized into UUDEX Groups (U-Groups). Any number of U-Participants or U-Endpoints can be members of a U-Group. A given U-Participant or U-Endpoint can be a member of multiple U-Groups, but a U-Group cannot contain other U-Groups. For example, a U-Group could be created to contain all U-Participant Transmission Operators within the purview of a Reliability Coordinator, and another U-Group could be created to contain all U-Participant Generator Operators within the purview of a Reliability Coordinator. U-Participants that are both Transmission Operators and Generator Operators would be members of both U-Groups. U-Participants, U-Endpoints, and U-Groups are used in the access control structures to provide access to individual UUDEX Subjects (U-Subjects). U-Groups are created by the U-Administrator and are managed by the U-Administrator or the designated U-Group Managers. U-Endpoints can be assigned UUDEX Roles (U-Roles) that can be used to further restrict access. U-Roles are assigned to individual U-Endpoints. For example, a U-Role of “Security Analyst” could be used to restrict which U-Endpoints can publish or subscribe security incident reports and vulnerability notifications, while a U-Role of “Transmission Planner” can be used to restrict which U-Endpoints can publish power system model updates. U-Role definitions are created by the U-Administrator, but the U-Roles are assigned to U-Endpoints by their respective UUDEX Participant Administrators (U-Participant Administrator). Because all information required to make security decisions is either included within the U-Endpoint’s X.509 digital certificate or stored in a datastore on the U-Server, all security decisions are performed and enforced within the U-Server. This reduces the complexity of the U-Client code and minimizes the chance for compromise of the integrity of the UUDEX security features.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Dynamical Aspects of the Quark Gluon Plasma

Shortly after the Big Bang the entire universe was filled with a nearly perfect fluid known as the Quark Gluon Plasma. Relativistic heavy ion collisions can now reproduce this fluid in the laboratory where the Quark Gluon Plasma exhibits a rapid but smooth cross-over phase transition into hadrons at vanishing net-baryon densities. Recent experiments plan to explore finite baryon densities where a critical point is expected. If found, this would mark the first discovery of a critical point in a relativistic system described by a fundamental theory of nature, which would have far-reaching consequences for high-energy nuclear physics and nuclear astrophysics (such as in neutron star mergers). Characteristic temperatures of equilibrium (e.g. the inflection point of the entropy density) and transport coefficients (e.g. minimum of the shear viscosity over entropy density) vary widely at a cross-over phase transition but converge at a critical point, and extracting the behavior of these characteristic temperatures is a major focal point of this research. Specifically, the interplay between strange and light hadrons is exploited to study the flavor hierarchy in the cross-over region. To investigate this, a viscous relativistic hydrodynamics framework with two conserved charges is being developed into a new open-source code along with initial conditions that contain baryon number and strangeness. Flow observables sensitive to the equation of state and transport coefficients are calculated across beam energies. New techniques are being developed to study this flavor hierarchy from first principles and to extract the characteristic temperatures from experimental data. Through this new dynamical framework, this project provides essential guidance to the Beam Energy Scan II runs at Relativistic Heavy-Ion Collider and the future Facility for Antiproton and Ion Research facility in the search for the Quantum Chromodynamic critical point and subsequent investigation of the baryon-rich Quark Gluon Plasma.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2024

Probing New Physics with Tau Leptons using the CMS Detector

The Standard Model (SM) of particle physics is a thoroughly well-tested theory of all electroweak scale phenomena to date—the most recent test being the discovery of the long-predicted Higgs boson. However, there is much evidence to suggest that it is incomplete. On the formal side, the mass of the Higgs particle itself is perplexingly small, while the largest theoretical energy scale of the SM is 17 orders of magnitude larger, a puzzle known as the hierarchy problem. On the experimental side, among other things, the SM lacks a candidate for dark matter, which is now known to play an indispensable role in astrophysical large scale structure formation. Many theories that address the fundamental puzzles of the SM predict the existence of multiple Higgs bosons, including a very light boson that couples preferentially to third-generation fermions. Observation of this new light boson, either directly produced or in the decay of the recently discovered SM Higgs, would provide an unmistakable sign of new dynamics in nature. The research supported by this award used data collected by the Compact Muon Solenoid (CMS) detector at the Large Hadron Collider (LHC) to explore boosted di-tau signatures of a new light boson as a tool in the quest to understand the shortcomings of the SM. With a proton-proton collision energy of 13 trillion electron-volts (TeV) and peak instantaneous proton luminosity of over 10^34 collisions per square centimeter per second (cm –2 s –1 ), the LHC is the only facility in the world that could enable this line of research. The research sought to address the hierarchy problem and other shortcomings of the SM through a mixture of LHC data analysis, maintenance of the CMS detector to ensure high quality data, and contributions to the Phase 2 upgrade detector for the upcoming High Luminosity phase of the LHC (HL-LHC).

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

To Interoperability And Beyond: Interoperable Types Through the Promises of C and C++ and ABI Abuse [Slides]

This presentation presents a technique that allows passing of Fortran nested-type hierarchies interoperably to C++. The technique is motivated by the Eulerian Application Project’s need to port code from Fortran to C++ to utilize the Kokkos performance portability library. The resulting method allows hierarchies of types to become interoperable while at the same time transforming Fortran array members of the original Fortran type into Kokkos::Views in the resulting C++ type.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Probing Naturalness with Searches for Supersymmetric Higgs Partners at the Large Hadron Collider

The discovery of the Higgs boson at the Large Hadron Collider completed the standard model of particle physics and led to the 2013 Nobel Prize. While it has been extremely successful in describing a wide variety of phenomena, the standard model cannot be the final theory of nature because it is unable to explain why the Higgs boson mass lies at the electroweak scale rather than 16 orders of magnitude larger at the Planck scale, a theoretical flaw known as the ``hierarchy problem.'' Supersymmetry is a standard model extension that could provide a natural solution to the hierarchy problem by introducing light supersymmetric Higgs partners. This proposal focused on searches that are sensitive to supersymmetric Higgs partners in data collected by the ATLAS experiment, which were carried out by the University of Illinois high-energy physics group led by PI Ben Hooberman as part of his Department of Energy Early Career Research Program award from 2017-2023. The searches were complemented by improvements to particle identification algorithms and the ATLAS trigger system that extend the sensitivity of future supersymmetry searches.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2026

Cooperative Agreement To Analyze variabiLity, change and predictabilitY in the earth SysTem (CATALYST)

CATALYST proposes to perform foundational coordinated research in a team-oriented collaborative effort aimed at advancing a robust understanding of modes of Earth system variability and change using models, observations and process studies. The proposed research will address the DOE/BER mission by exploring the limits to predictability, identifying fundamental underlying mechanisms, quantifying interactions among modes of variability, and discovering tipping points in the Earth system to understand the current and future impacts of these phenomena on regional and global climate. Four fundamental gaps are identified in our knowledge of the Earth system: 1) What are the limits to predictability on various timescales? 2) What are the interactions among modes of Earth system variability? 3) How may modes of Earth system variability change in response to changes in external forcing, and what are the tipping points involved with those changes? 4) How are high impact events connected to modes of Earth system variability and how may they change in the future? Related to those gaps in our knowledge, we formulate four research objectives to address those gaps using a combination of Earth system models (ESMs) and machine learning (ML) methods. Research Objective 1 (RO1) addresses the first gap above and proposes to understand modes of variability and their limits of predictability on subseasonal to decadal timescales using ESMs and ML. Research Objective 2 (RO2) addresses the second gap and proposes to use a hierarchy of models to understand relevant processes and feedbacks related to how modes of variability interact with each other. Research Objective 3 (RO3) is designed to study the third gap and proposes to examine the role of external forcings in changes of modes of Earth system variability and their interactions, and the likelihood and predictability of tipping points and irreversible changes. Research Objective 4 (RO4) will address the fourth gap and proposes to use high resolution ESMs, regionally refined models (RRMs), and ML methods to investigate the relationships between high impact events (e.g. flash droughts and precipitation extremes, atmospheric rivers (ARs), tropical cyclones (TCs), storm surge/sea level rise), the synoptic systems that produce them, and their changes related to modes of Earth system variability. The research will involve the use of the Community Earth System Model (CESM), Energy Exascale Earth System Model (E3SM), CMIP multi-model data sets, a hierarchy of simpler models, and numerous observational data sets. In the course of the proposed research, CATALYST will contribute to metrics and diagnostics that will be integrated in Coordinated Model Evaluation Capabilities (CMEC), particularly with regards to the Quasi-biennial Oscillation (QBO) and its interactions with the Madden-Julian Oscillation (MJO), high atmospheric pressure blocking, and new precipitation metrics.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Multifidelity Ensemble Kalman Filtering Using Surrogate Models Defined by Theory-Guided Autoencoders

Data assimilation is a Bayesian inference process that obtains an enhanced understanding of a physical system of interest by fusing information from an inexact physics-based model, and from noisy sparse observations of reality. The multifidelity ensemble Kalman filter (MFEnKF) recently developed by the authors combines a full-order physical model and a hierarchy of reduced order surrogate models in order to increase the computational efficiency of data assimilation. The standard MFEnKF uses linear couplings between models, and is statistically optimal in case of Gaussian probability densities. This work extends the MFEnKF into to make use of a broader class of surrogate model such as those based on machine learning methods such as autoencoders non-linear couplings in between the model hierarchies. We identify the right-invertibility property for autoencoders as being a key predictor of success in the forecasting power of autoencoder-based reduced order models. We propose a methodology that allows us to construct reduced order surrogate models that are more accurate than the ones obtained via conventional linear methods. Numerical experiments with the canonical Lorenz'96 model illustrate that nonlinear surrogates perform better than linear projection-based ones in the context of multifidelity ensemble Kalman filtering. We additionality show a large-scale proof-of-concept result with the quasi-geostrophic equations, showing the competitiveness of the method with a traditional reduced order model-based MFEnKF.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Leveraging Prior Concept Learning Improves Generalization From Few Examples in Computational Models of Human Object Recognition

Humans quickly and accurately learn new visual concepts from sparse data, sometimes just a single example. The impressive performance of artificial neural networks which hierarchically pool afferents across scales and positions suggests that the hierarchical organization of the human visual system is critical to its accuracy. These approaches, however, require magnitudes of order more examples than human learners. We used a benchmark deep learning model to show that the hierarchy can also be leveraged to vastly improve the speed of learning. We specifically show how previously learned but broadly tuned conceptual representations can be used to learn visual concepts from as few as two positive examples; reusing visual representations from earlier in the visual hierarchy, as in prior approaches, requires significantly more examples to perform comparably. These results suggest techniques for learning even more efficiently and provide a biologically plausible way to learn new visual concepts from few examples.

Rule, Joshua S.↗
DOE OSTI2021

Development of a Unified Taxonomy for HVAC System Faults

Detecting and diagnosing HVAC faults is critical for maintaining building operation performance, reducing energy waste, and ensuring indoor comfort. An increasing deployment of commercial fault detection and diagnostics (FDD) software tools in commercial buildings in the past decade has significantly increased buildings’ operational reliability and reduced energy consumption. A massive amount of data has been generated by the FDD software tools. However, efficiently utilizing FDD data for ‘big data’ analytics, algorithm improvement, and other data-driven applications is challenging because the format and naming conventions of those data are very customized, unstructured, and hard to interpret. This paper presents the development of a unified taxonomy for HVAC faults. A taxonomy is an orderly classification of HVAC faults according to their characteristics and causal relations. The taxonomy includes fault categorization, physical hierarchy, fault library, relation model, and naming/tagging scheme. The taxonomy employs both a physical hierarchy of HVAC equipment and a cause-effect relationship model to reveal the root causes of faults in HVAC systems. A structured and standardized vocabulary library is developed to increase data representability and interpretability. The developed fault taxonomy can be used for HVAC system ‘big data’ analytics such as HVAC system fault prevalence analysis or the development of an HVAC FDD software standard. A common type of HVAC equipment-packaged rooftop unit (RTU) is used as an example to demonstrate the application of the developed fault taxonomy. Two RTU FDD software tools are used to show that after mapping FDD data according to the taxonomy, the meta-analysis of the multiple FDD reports is possible and efficient.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗
DOE OSTI2025

Systematic improvement of trial states in phaseless auxiliary-field quantum Monte Carlo

We extend the use of coupled cluster (CC) trial states in the phaseless auxiliary-field quantum Monte Carlo (AFQMC) method beyond single and double excitations to include both triple and quadruple excitations. With this AFQMC/CC hierarchy, we are able to systematically benchmark the method's performance on molecular systems as the quality of the trial is improved. Our results show that the phaseless AFQMC energy improves systematically and is typically significantly more accurate than the energy of the underlying trial state. However, the relative improvement compared to the trial CC energy decreases as we ascend the CC hierarchy. As the CC wavefunction is usually further approximated when used as an AFQMC trial, we also explore the relationship between the components of the CC wavefunction and the resulting AFQMC/CC error. Our results suggest that improving the representation of the CC wave function in the AFQMC trial does not always lower the bias even when it increases the fidelity of the trial with the exact ground state.

Chemical Physics (physics.chem-ph)↗
DOE OSTI2024

A Generalized approach to the operationalization of Software Quality Models

Comprehensive measures of quality are a research imperative, yet the development of software quality models is a wicked problem. Definitive solutions do not exist and quality is subjective at its most abstract. Definitional measures of quality are contingent on a domain, and even within a domain, the choice of representative characteristics to decompose quality is subjective. Thus, the operationalization of quality models brings even more challenges. A promising approach to quality modeling is the use of hierarchies to represent characteristics, where lower levels of the hierarchy represent concepts closer to real-world observations. Building upon prior hierarchical modeling approaches, we developed the Platform for Investigative software Quality Understanding and Evaluation (PIQUE). PIQUE surmounts several quality modeling challenges because it allows modelers to instantiate abstract hierarchical models in any domain by leveraging organizational tools tailored to their specific contexts. Here, we introduce PIQUE; exemplify its utility with two practical use cases; address challenges associated with parameterizing a PIQUE model; and describe algorithmic techniques that tackle normalization, aggregation, and interpolation of measurements.

Data aggregation↗
DOE OSTI2020

Understanding the Impact of Memory Access Patterns in Intel Processors

Because of increasing complexity in the memory hierarchy, predicting the performance of a given application in a given processor is becoming more difficult. The problem is worsened by the fact that the hardware needed to deal with more complex memory traffic also affects energy consumption. Moreover, in a heterogeneous system with shared main memory, the memory traffic between the last level cache (LLC) and the memory creates contention between other processors and accelerator devices. For these reasons, it is important to investigate and understand the impact of different memory access patterns on the memory system. This study investigates the interplay between Intel processors' memory hierarchy and different memory access patterns in applications. The authors explore sequential streaming and strided memory access patterns with the objective of predicting LLC-dynamic random access memory (DRAM) traffic for a given application in given Intel architectures. Moreover, the impact of prefetching is also investigated in this study. Experiments with different Intel micro-architectures uncover mechanisms to predict LLC-DRAM traffic that can yield up to 99% accuracy for sequential streaming access patterns and up to 95% accuracy for strided access patterns.

Alaul haque monil, Mohammad↗
DOE OSTI2023

Adaptive cache management based on programming model information

A processing system selectively allocates space to store a group of one or more cache lines at a cache level of a cache hierarchy having a plurality of cache levels based on memory access patterns of a software application executing at the processing system. The processing system generates bit vectors indicating which cache levels are to allocate space to store groups of one or more cache lines based on the memory access patterns, which are derived from data granularity and movement information. Based on the bit vectors, the processing system provides hints to the cache hierarchy indicating the lowest cache level that can exploit the reuse potential for a particular data.

Na, Weon Taek↗
DOE OSTI2023

Systems and methods for hierarchical multi-objective optimization

A method for optimizing objective functions can include selecting an objective function based at least on a hierarchy, applying parameters to the objective function to generate an output, responsive to the output not satisfying a tolerance condition, assigning a penalty to the set of parameters and evaluating a convergence condition using the set of parameters and the penalty, responsive to the output satisfying the tolerance condition, evaluating an additional objective function using the parameters in an order corresponding to the hierarchy or evaluating the convergence condition responsive to the selected objective function being a final objective function, modifying the set of parameters using a genetic algorithm responsive to the set of parameters not satisfying the convergence condition, and outputting the set of parameters responsive to the set of parameters satisfying the convergence condition.

Chan, Henry↗
DOE OSTI2020

MH-1A Sturgis Decommissioning and Dismantlement - 20281

The U.S. Army Corps of Engineers (USACE) with its prime contractor, Aptim Federal Services, LLC (APTIM), recycled 5,260 metric tons (11.6 million pounds) of material from Sturgis Barge (Sturgis) during the Decommissioning and Dismantlement of the MH-1A nuclear power reactor. The overall objective of the project was to reduce residual radioactivity associated with MH-1A to levels that permitted release of Sturgis for dismantlement and termination of the Army Reactor Office permit. By effectively applying waste hierarchy's three Rs - reduce, reuse and recycle - the Sturgis project not only minimized the amount of waste that required disposal at landfills, but also reduced the potential for long-term environmental liability emanating from these landfills. The project team completed the physical decommissioning efforts in June 2018 in Galveston, TX. In September 2018, radiological surveys were completed to demonstrate the vessel could be released for shipbreaking. Sturgis was towed from Galveston, TX to Brownsville, TX in late September 2018. Shipbreaking, dismantlement and recycling efforts began in early October 2018 and were completed on 15 March 2019. As part of the decommissioning effort in Galveston, the team shipped 69 shipments (860 metric tons) of low-level radioactive waste and radioactive components to the Waste Control Specialist (WCS) facility in Andrews TX for disposal. Certain radioactive components had to be transferred to the Department of Energy prior to placing the materials into the Federal Waste Facility located within WCS. However, most of the radioactive waste was characterized, profiled, approved and managed under the WCS permitted radioactive waste exemption process authorized and implemented by the Texas Commission on Environmental Quality (TCEQ) and the Radioactive Materials Division. This allows LLRW and LLMW to be shipped as regulated waste and then upon receipt at WCS through satisfying the relevant waste acceptance criteria the waste is exempted and placed into the WCS RCRA permitted cell. An additional 35 shipments (544 metric ton) of contaminated hazardous waste water were transported to U.S. Ecology in Robstown, TX for treatment/disposal. An additional 36 shipments (500 metric tons) of non-hazardous wastewater was sent to Republic Waste Services' facility in Fresno, TX. The disposal of these materials required close coordination with State of Texas regulators. During decommissioning, the project team recycled approximately 270 metric tons (600,000 pounds) of lead and steel. As part of the dismantlement in Brownsville, TX the team recycled approximately 5,000 metric tons (11 million pounds) of ferrous and non-ferrous material, limiting our disposal requirements to about 180 metric tons (400,000 pounds) of material (<4% from entire shipbreaking activity). Although the primary hazard being mitigated by this project was radiological, recycling was always a priority for the project. The team strived to achieve sustainability goals as we implemented this one of a kind project. Scrap metal recycling has a large positive impact on the environment and can also favorably impact project disposal costs. Steel is among the most recycled material in the world. Nearly 40% of the world's steel production is made from scrap. Recycling steel also requires 75% less energy than producing it from raw materials. By using recycled steel rather than virgin materials, 2.33 kg of carbon emissions are eliminated per kg of steel [1]. The project recycled more than 4,500 metric tons (10 million pounds) of steel, which eliminated about 10,400 metric tons (23 million pounds) of CO{sub 2}. By implementing a recycling initiative for the Sturgis project, the team was able to realize cost avoidance for disposal of scrap, cost savings from the metals recycled, plus the project provided benefits to the environment through our recycling efforts. Once the dismantlement was complete, the team prepared a detailed decommissioning closure report, which allowed for the termination of the Army Reactor Decommissioning Permit. While it not only reduced any potential long-term environmental liability, this project to decommission and dismantle a floating nuclear power plant is truly unprecedented - it is a prime example of the USACE mission which is: 'Engineering solutions for the Nation's toughest challenges'. This unique, one of a kind, historical power plant was never designed to be taken apart, and the available information about its construction was lacking in many details. The hazards that required mitigation dictated a painstaking and deliberate process in order to avoid any release to the environment and the community, and to protect the health and safety of the workers involved while keeping the waste hierarchy's three R's - reduce, reuse and recycle at the forefront. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗
DOE OSTI2025

SHF: Symmetrical Hierarchical Forest with Pretrained Vision Transformer Encoder for High-Resolution Medical Segmentation

This paper presents a novel approach to addressing the long-sequence problem in high-resolution medical images for Vision Transformers (ViTs). Using smaller patches as tokens can enhance ViT performance, but quadratically increases computation and memory requirements. Therefore, the common practice for applying ViTs to high-resolution images is either to: (a) employ complex sub-quadratic attention schemes or (b) use large to medium-sized patches and rely on additional mechanisms within the model to capture the spatial hierarchy of details. We propose Symmetrical Hierarchical Forest (SHF), a lightweight approach that adaptively patches the input image to increase token information density and encode hierarchical spatial structures into the input embedding. We then apply a reverse depatching scheme to the output embeddings of the transformer encoder, eliminating the need for convolution-based decoders. Unlike previous methods that modify attention mechanisms or use a complex hierarchy of interacting models, SHF can be retrofitted to any ViT model to allow it to learn the hierarchical structure of details in high-resolution images without requiring architectural changes. Experimental results demonstrate significant gains in computational efficiency and performance: on the PAIP WSI dataset, we achieved a 3∼32×speedup or a 2.95%∼7.03% increase in accuracy (measured by Dice score) at a 64K2 resolution with the same computational budget, compared to state-of-the-art production models. On the 3D medical datasets BTCV and KiTS, training was 6×faster, with accuracy gains of 6.93% and 5.9%, respectively, compared to models without SHF.

Zhang, Enzhi [Hokkaido University, Japan]↗