Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “task tuning”

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 145 records · Page 8
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Sequential Image Recovery Using Joint Hierarchical Bayesian Learning

Abstract Recovering temporal image sequences (videos) based on indirect, noisy, or incomplete data is an essential yet challenging task. We specifically consider the case where each data set is missing vital information, which prevents the accurate recovery of the individual images. Although some recent (variational) methods have demonstrated high-resolution image recovery based on jointly recovering sequential images, there remain robustness issues due to parameter tuning and restrictions on the type of sequential images. Here, we present a method based on hierarchical Bayesian learning for the joint recovery of sequential images that incorporates prior intra- and inter-image information. Our method restores the missing information in each image by “borrowing” it from the other images. More precisely, we couple sequential images by penalizing their pixel-wise difference. The corresponding penalty terms (one for each pixel and pair of subsequent images) are treated as weakly-informative random variables that favor small pixel-wise differences but allow occasional outliers. As a result, all of the individual reconstructions yield improved accuracy. Our method can be used for various data acquisitions and allows for uncertainty quantification. Some preliminary results indicate its potential use for sequential deblurring and magnetic resonance imaging.

Xiao, Yao↗
DOE OSTI2024

SAM-I-Am: Semantic boosting for zero-shot atomic-scale electron micrograph segmentation

Image segmentation is a critical enabler for tasks ranging from medical diagnostics to autonomous driving. However, the correct segmentation semantics — where are boundaries located? what segments are logically similar? — change depending on the domain, such that state-of-the-art foundation models can generate meaningless and incorrect results. Moreover, in certain domains, fine-tuning and retraining techniques are infeasible: obtaining labels is costly and time-consuming; domain images (micrographs) can be exponentially diverse; and data sharing (for third-party retraining) is restricted. To enable rapid adaptation of the best segmentation technology, we propose the concept of semantic boosting: given a zero-shot foundation model, guide its segmentation and adjust results to match domain expectations. Here, we apply semantic boosting to the Segment Anything Model (SAM) to obtain microstructure segmentation for transmission electron microscopy. Our booster, SAM-I-Am, serves as a post-processing engine that extracts geometric and textural features of various intermediate masks to perform mask removal and mask merging operations. We demonstrate a zero-shot performance increase of (absolute) +21.35%, +12.6%, +5.27% in mean IoU, and a -9.91%, -18.42%, -4.06% drop in mean false positive masks across images of three difficulty classes over vanilla SAM (ViT-L).

36 MATERIALS SCIENCE↗
DOE OSTI2021

A generalized machine learning workflow to visualize mechanical discontinuity

Accurate detection and mapping of mechanical discontinuity in materials has widespread industrial and research applications. Herein, we developed a generalized machine-learning framework for visualizing single mechanical discontinuity embedded in material of any composition, velocity, density, porosity, and size with limited data. The proposed visualization of discontinuity requires accurate estimations of the length, location, and orientation of the embedded discontinuity by processing multipoint wave-transmission measurements. k-Wave simulator is used to create a large dataset of elastic waveforms recorded during multi-point wave-transmission measurements through materials containing single mechanical discontinuity. k-Wave simulator considers the wave attenuation, dispersion, and mode conversion in wave motion. Discrete wavelet transform (DWT) and statistical feature extraction are essential for data preprocessing prior to the data-driven model development. DWT also minimizes the effect of noise. Using hyper-parameter tuning and cross validation, gradient boosting regression can visualize the mechanical discontinuity with an accuracy of 0.85, in terms of coefficient of determination. A double-layered neural network-based regression has better performance with an accuracy of 0.95. Use of convolutional neural network converts the predictive task from a waveform processing to an image processing problem. Convolutional neural network achieved a generalization performance of 0.91. The proposed generalized workflow requires robust simulation of wave propagation, signal processing, feature engineering, and model evaluation. Sensors closest to the source and those located opposite the source are the most significant for the desired visualization. Notably, the sensors closest to the source capture the non-linear associations, whereas the sensor on the border opposite to the source capture the linear associations between the measured waveforms and the properties of the mechanical discontinuity.

42 ENGINEERING↗
DOE OSTI2021

Microscopic characterization of Ising conformal field theory in Rydberg chains

Rydberg chains provide an appealing platform for probing conformal field theories (CFTs) that capture universal behavior in a myriad of physical settings. Focusing on a Rydberg chain at the Ising transition separating charge density wave and disordered phases, we establish a detailed link between microscopics and low-energy physics emerging at criticality. Here, we first construct lattice incarnations of primary fields in the underlying Ising CFT including chiral fermions, a nontrivial task given that the Rydberg chain Hamiltonian does not admit an exact fermionization. With this dictionary in hand, we compute correlations of microscopic Rydberg operators, paying special attention to finite, open chains of immediate experimental relevance. We further develop a method to quantify how second-neighbor Rydberg interactions tune the sign and strength of four-fermion couplings in the Ising CFT. Finally, we determine how the Ising fields evolve when four-fermion couplings drive an instability to Ising tricriticality. Our results pave the way to a thorough experimental characterization of Ising criticality in Rydberg arrays and can inform the design of novel higher-dimensional phases based on coupled critical chains.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2024

VLM4Bio: A Benchmark Dataset to Evaluate Pretrained Vision-Language Models for Trait Discovery from Biological Images

Images are increasingly becoming the currency for documenting biodiversity on the planet, providing novel opportunities for accelerating scientific discoveries in the field of organismal biology, especially with the advent of large vision-language models (VLMs). We ask if pre-trained VLMs can aid scientists in answering a range of biologically relevant questions without any additional fine-tuning. In this paper, we evaluate the effectiveness of 12 state-of-the-art (SOTA) VLMs in the field of organismal biology using a novel dataset, VLM4Bio, consisting of 469K question8 answer pairs involving 30K images from three groups of organisms: fishes, birds, and butterflies, covering five biologically relevant tasks. We also explore the effects of applying prompting techniques and tests for reasoning hallucination on the performance of VLMs, shedding new light on the capabilities of current SOTA VLMs in answering biologically relevant questions using images

Maruf, M [Virginia Tech, Blacksburg]↗
DOE OSTI2025

Understanding the optoelectronic properties of doped 2D organic-inorganic halide perovskite quantum wells: towards efficient ultrafast quantum well IR photodetectors

This project, titled “Understanding the optoelectronic properties of doped 2D organic-inorganic halide perovskite quantum wells: towards efficient quantum well IR photodetectors”, was funded by the U.S. Department of Energy to explore a new class of materials that could make future light-sensing technologies, such as infrared (IR) cameras and detectors more efficient, affordable, and widely available. The research focused on special layered materials called 2D halide perovskites, which are made up of alternating organic and inorganic layers only a few atoms thick. These materials can be tuned at the atomic level to absorb and emit light in precise ways, making them very attractive for use in optoelectronic devices. The main goal of the project was to understand how these perovskite materials absorb light and move electrical charges at very small scales. However, this is not an easy task. These materials often contain a mixture of different structures in the same film, and traditional tools like regular absorption or photoluminescence spectroscopy are not good at telling those structures apart. To solve this, the research team, led by Professor Luisa Whittaker-Brooks at the University of Utah developed a powerful method called electroabsorption spectroscopy. This technique uses electric fields to highlight the unique “fingerprints” of different excitons, which are tiny packets of energy formed when light hits the material. By using this method, the team could separate overlapping signals and learn exactly how the materials respond to light under different conditions, including changes in temperature, thickness, and chemical makeup.

36 MATERIALS SCIENCE↗
DOE OSTI2026

Computer vision-based rock bolt detection in orthomosaic imagery obtained in the Waste Isolation Pilot Plant underground facility

Assessing structural integrity of large underground tunnel facilities is often a time consuming and human-labor intensive task. Thus, research using various modes of sensing and automated detection of key structural components in mines is posed to aid in safety assessments and establishing overall structural health. We propose an approach utilizing off-the-shelf camera and lidar technology fixed to a custom sensing platform, image stitching techniques, fine-tuned object detection models, and specialized model-inference methods to automatically detect, count, and map roof bolts for assessment of structural safety in man-made underground tunnels. Results show a novel workflow for effective object counting in orthomosaic tunnel ceiling images generated from collections in GPS-denied mining environments. Additionally, we demonstrate effective fine-tuning of EfficientDet object detectors utilizing state-of-the-art image augmentation techniques known as the mosaic and mixup transformations. Our work is demonstrated on sensed data and imagery collected from the Department of Energy (DOE) Waste Isolation Pilot Plant (WIPP) where miles of tunnel ceiling must be assessed for structural integrity.

42 ENGINEERING↗
DOE OSTI2026

When does global attention help: a unified empirical study on atomistic graph learning

Graph neural networks (GNNs) are widely used as surrogates for costly experiments and first-principles simulations to study the behavior of compounds at atomistic scale, and their architectural complexity is constantly increasing to enable the modeling of complex physics. While most recent GNNs combine more traditional message passing neural networks (MPNNs) layers to model short-range interactions with more advanced graph transformers (GTs) with global attention mechanisms to model long-range interactions, it is still unclear when global attention mechanisms provide real benefits over well-tuned MPNN layers due to inconsistent implementations, features, or hyperparameter tuning. We introduce the first unified, reproducible benchmarking framework–built on HydraGNN–that enables seamless switching among four controlled model classes: MPNN, MPNN with chemistry/topology encoders, GPS-style hybrids of MPNN with global attention, and fully fused localglobal models with encoders. Using seven diverse open-source datasets for benchmarking across regression and classification tasks, we systematically isolate the contributions of message passing, global attention, and encoder-based feature augmentation. Our study shows that encoder-augmented MPNNs form a robust baseline, while fused localglobal models yield the clearest benefits for properties governed by long-range interaction effects. We further quantify the accuracycompute trade-offs of attention, reporting its overhead in memory. Together, these results establish the first controlled evaluation of global attention in atomistic graph learning and provide a reproducible testbed for future model development.

Equivariant graph neural networks↗
DOE OSTI2022

Giant g-factors and fully spin-polarized states in metamorphic short-period InAsSb/InSb superlattices

Abstract Realizing a large Landé g -factor of electrons in solid-state materials has long been thought of as a rewarding task as it can trigger abundant immediate applications in spintronics and quantum computing. Here, by using metamorphic InAsSb/InSb superlattices (SLs), we demonstrate an unprecedented high value of g ≈ 104, twice larger than that in bulk InSb, and fully spin-polarized states at low magnetic fields. In addition, we show that the g -factor can be tuned on demand from 20 to 110 via varying the SL period. The key ingredients of such a wide tunability are the wavefunction mixing and overlap between the electron and hole states, which have drawn little attention in prior studies. Our work not only establishes metamorphic InAsSb/InSb as a promising and competitive material platform for future quantum devices but also provides a new route toward g -factor engineering in semiconductor structures.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2024

HydraGNN_Predictive_GFM_2024 - Ensemble of predictive graph foundation models for ground state atomistic materials modeling

We provide the ensemble of fifteen pre-trained graph foundation models (GFMs) for atomistic materials modeling applications. Each one of the fifteen GFMs has been trained on five open-source datasets that (once aggregated) amount to over 154 million atomistic structures, which cover over two-thirds of the natural elements of the periodic table and that comprises a broad set of organic and inorganic compounds. This vast set of atomistic structures comprises ground state configurations that are dynamically stable (i.e., equilibrated structures with atomic forces approximately close to zero values) as well as dynamically unstable structures (i.e., non-equilibrium structures with non-negligible non-zero values of atomic forces). The ensemble of datasets aggregated does NOT include excited states. The datasets have been curated to remove atomistic structures with spectral norm of the force tensor above 100 eV/angstrom. Moreover, a linear term of the energy was computed for each dataset using a linear regression model that uses the chemical concentration of each natural element as regressor. The linear term predicted by the linear regression model has been subtracted from each original energy value to perform a re-alignment of the energy values across different electronic structures approximation theories performed to generate the diverse multi-source, multi-fidelity datasets. The folder "ADIOS_files" contains the set of pre-processed datasets in Adaptable I/O System (ADIOS) format (https://www.exascaleproject.org/research-project/adios/) that have been used for the development and training of GFMs in this work. The "ADIOS_files" directory contains 6 sub-directories named as follows: - ANI1x-v3.bp - MPTrj-v3.bp - OC2020-20M-v3.bp - OC2020-v3.bp - OC2022-v3.bp - qm7x-v3.bp Each sub-directory contains the pre-processed datasets converted in Adaptable I/O System (ADIOS) format (https://www.exascaleproject.org/research-project/adios/) that have been used to the development, training, and performance testing of the ensemble go predictive graph foundation models. Each GFM was developed using HydraGNN (https://github.com/ORNL/HydraGNN) as underlying graph neural network (GNN) architecture. The multi-task learning (MTL) capability of HydraGNN was used to simultaneously train the GFMs on labeled values for direct predictions of energy (a total system property of an atomistic structure that measures the chemical stability) and atomic forces (an atomic level property of an atomistic structure that measures the dynamical stability). The hyper parameters of the GFM have been tuned using scalable hyperparameter optimization (HPO) algorithms implemented in the software DeepHyper (https://github.com/deephyper/deephyper). The pre-training of each HPO trial was performed using distributed data parallelism (DDP) to scale the training across 128 compute nodes of the exascale OLCF supercomputer Frontier. Each HPO trial was trained only for 10 epochs and an early stopping was performed to avoid wasting significant computational resources on GNN architectures that were clearly underperforming. For each HPO trial, the 'omnistat' tool developed by (AMD Research - Advanced Micro Device) was used to measure the total energy consumption in kWh. The ensemble of GFMs was obtained by selecting the fifteen best performing HPO trials. Four models have been selected for their clear advantage in accuracy, and these are the GFMs with IDs 229, 156, 147, 260. Additional eleven models have been selected based on judicious balance between accuracy and energy consumption needed for training, and these are the GFMs with IDs 165, 78, 137, 1, 175, 171, 181, 67, 179, 167, 351. Each selected GFM of the ensemble was continued to cumulate a total of at most 30 epochs. In some cases, the total number of epochs actually performed was les than 30 due to two combined factors: (1) the size of the GFM (i.e., the number of model parameters to train) and (2) the total wall-clock time for which the computational resources could be allocated on OLCF-Frontier. The "Ensemble_of_models" directory contains 15 sub-directories named as follows: - gfm_0.229 - gfm_0.156 - gfm_0.147 - gfm_0.260 - gfm_0.165 - gfm_0.78 - gfm_0.137 - gfm_0.1 - gfm_0.175 - gfm_0.171 - gfm_0.181 - gfm_0.67 - gfm_0.179 - gfm_0.167 - gfm_0.351 Each one of these sub-directories refers to one of the fifteen HPO trials that have been selected to continue the pre-training with at most 30 epochs. With each sub-directory associated with a specific HPO trial, the following files can be found: - config.json: file for argument parsing to develop and train an HydraGNN architecture - gfm_0.ID_epoch_N.pk: file with model parameters for HPO ID trial after N epochs of training The ensemble of fifteen GFM architectures was used for (1) ensemble averaging to stabilize the predictions of energy and atomic forces after pre-training for post-processing analysis and (2) ensemble uncertainty quantification (UQ). The code used to develop, pre-train, and load the pre-trained models for post-processing analysis is available on the ORNL-GitHub at the following link: https://github.com/ORNL/HydraGNN/tree/Predictive_GFM_2024

36 MATERIALS SCIENCE↗
DOE OSTI2025

Machine learning-accelerated path integral molecular dynamics simulations of reactive organic electrolytes

Hydrogen bonded electrolytes that exhibit accelerated proton transport via sequential reactive hops have drawn interest for their promise in clean energy applications. Molecular dynamics simulations of these electrolytes offer the opportunity to uncover microscopic mechanistic details that could be used to design and tune the properties of candidate electrolyte technologies. However, accurately modeling the proton transfer reactions and transport properties that give rise to high charge conductivites in these electrolytes proves computationally challenging because of the need to perform lengthy condensed phase simulations, treating both the electronic and nuclear degrees of freedom quantum mechanically. In this paper, we demonstrate that such a modeling task can be efficiently achieved with the use of density functional theory (DFT)-trained machine learning potentials (MLP) to accelerate path integral molecular dynamics (PIMD) simulations. We highlight the practical utility of this approach by using it to benchmark how closely PIMD simulations employing different DFT exchange–correlation functionals reproduce the composition-dependent densities, diffusion coefficients, and electrical conductivities of mixtures consisting of imidazole and levulinic acid. Even with the speedup afforded by our MLPs, PIMD simulations remain quite expensive. Furthermore, in order to render PIMD more computationally tractable, we introduce and benchmark the accuracy of a ring polymer contraction approach that leverages a computationally efficient short-range MLP to accelerate our PIMD simulations by an additional factor of four.

Chemical bonding↗
DOE OSTI2025

Deep-learning-based domain adaptation for cavity fault prediction at Jefferson Laboratory

Superconducting radio-frequency (SRF) cavities are the core components of the Continuous Electron Beam Accelerator Facility (CEBAF) at Jefferson Lab, providing high-power electron beams for nuclear physics experiments. The facility comprises 418 SRF cavities, and any fault in these cavities can lead to interruptions in the electron beam supply. Cavity faults are the leading cause of beam trips in CEBAF. Predicting and mitigating those faults before onset can help maintain normal operation. Existing models face challenges in distinguishing between normal and fault signals when changes occur in the underlying time-series data, from changes in control software, operational parameters, or the environment. This work proposes a deep learning domain adaptation model that leverages transfer learning to address fault prediction challenges by improving accuracy. The model is trained and fine-tuned using a dataset collected for faulty and normal operation using a data acquisition system in CEBAF. Our deep learning-based domain adaptation model achieves a prediction accuracy of 89.61% of the fault and normal signals. The developed model effectively predicts normal running signals compared to the baseline approach without domain adaptation. This capacity is essential for the fault prediction task in the CEBAF because of heavily imbalanced data containing vast amounts of normal signals. The model performs well for predicting faults several hundred milliseconds before the fault onset compared to other models where no adaptation is applied. Incorporating deep learning-based domain adaptation techniques will significantly improve the fault prediction performance.

Rahman, Md Monibor [Old Dominion Univ., Norfolk, V↗
DOE OSTI2026

Scalable Generation of High-fidelity Synthetic Population Ensembles

Used within social simulations, synthetic population ensembles enable uncertainty quantification (UQ) methods for obtaining more robust model inference and prediction. A synthetic population ensemble is a series of plausible virtual reconstructions of an area’s population at the granularity of people and residences, generated stochastically to preserve privacy of the source population survey’s respondents. In this paper, we demonstrate the production of large synthetic population ensembles for the U.S. via Oak Ridge National Laboratory’s UrbanPop framework to support modeling of high spatial resolution energy affordability metrics from nationwide social surveys in collaboration with the fusionACS project. The study involves two scenarios: creating ensembles for (1) 17 U.S. metropolitan areas in 2019 and (2) full U.S. Census Divisions in 2023, with each scenario consisting of 41 population instances (a base realization and 40 replicates). To accomplish this task at scale, we configured an integrated system within a research cloud, comprised of virtual containerizations, GPU-enhanced functionality, and orchestrated deployments of UrbanPop’s maturing Likeness Python ecosystem. Results demonstrate we maintained high-fidelity approximations of residential totals by areas of interest and the demographic characteristics of neighborhoods while reducing manual workflow burdens. Finally, we discuss plans to fine-tune and further develop our automated workflows for truly distributed job orchestration to increase computational efficiency, as well as provide an outlook for broadening applications of the ensembles.

Cluster computing↗
DOE OSTI2021

Evaluating the Performance of Integer Sum Reduction in SYCL on GPUs

SYCL is a promising programming model for heterogeneous computing—allowing a single-source code to target devices from multiple vendors. One significant task performed on these accelerators is a primitive operation for integer sum reduction. This paper presents several SYCL implementations of integer sum reduction—using atomic functions, shared local memory, vectorized memory accesses and parameterized workload sizes—to compare the performance and maturity of SYCL against open-source vendor-specific implementations of the same reduction. For a sufficiently large number of integers, tuning the parameters of our SYCL implementations achieves 1.4X speedup over the open-source implementations on an Intel UHD630 integrated GPU. The SYCL reduction is 3% faster than the templated reduction in Thrust, and 0.3% faster than the device reduction in CUB on an Nvidia P100 GPU. The SYCL reduction is 1.9% faster than the templated reduction in Thrust, and 0.4% faster than the device reduction in CUB on an Nvidia V100 GPU.

Jin, Zheming↗
DOE OSTI2020

Super resolution for root imaging

Premise High‐resolution cameras are very helpful for plant phenotyping as their images enable tasks such as target vs. background discrimination and the measurement and analysis of fine above‐ground plant attributes. However, the acquisition of high‐resolution images of plant roots is more challenging than above‐ground data collection. An effective super‐resolution (SR) algorithm is therefore needed for overcoming the resolution limitations of sensors, reducing storage space requirements, and boosting the performance of subsequent analyses. Methods We propose an SR framework for enhancing images of plant roots using convolutional neural networks. We compare three alternatives for training the SR model: (i) training with non‐plant‐root images, (ii) training with plant‐root images, and (iii) pretraining the model with non‐plant‐root images and fine‐tuning with plant‐root images. The architectures of the SR models were based on two state‐of‐the‐art deep learning approaches: a fast SR convolutional neural network and an SR generative adversarial network. Results In our experiments, we observed that the SR models improved the quality of low‐resolution images of plant roots in an unseen data set in terms of the signal‐to‐noise ratio. We used a collection of publicly available data sets to demonstrate that the SR models outperform the basic bicubic interpolation, even when trained with non‐root data sets. Discussion The incorporation of a deep learning–based SR model in the imaging process enhances the quality of low‐resolution images of plant roots. We demonstrate that SR preprocessing boosts the performance of a machine learning system trained to separate plant roots from their background. Our segmentation experiments also show that high performance on this task can be achieved independently of the signal‐to‐noise ratio. We therefore conclude that the quality of the image enhancement depends on the desired application.

Ruiz‐Munoz, Jose F.↗
DOE OSTI2022

Mitigate: An Adaptive Network Data Anonymization Tool Using Condensation-Based Differential Privacy

Modern network devices collect a large amount of data that can be analyzed to identify bottlenecks, anomalies, cyber-attacks, etc. Therefore, there is often a need to analyze such collections of network data quite often by an external expert or by the research community. However, these collections of data contain sensitive, proprietary information. In order for the network data to be shared, it must first be anonymized. The overall objective of this project is to develop an innovative privacy management tool to anonymize network data and achieve sufficient privacy, acceptable data utility, and efficient data analysis at the same time. No existing anonymization methods can achieve all of these at the same time. The core of this technology is a differential private clustering algorithm that provides strong privacy protection, preserves data properties important for subsequent analysis, and allows the party receiving the anonymized data to conduct analysis directly on anonymized data without the need of decryption or any extra processing. The research carried out was to design, implement and verify a solution to this problem by completing the following tasks: 1) developing the core technology; 2) developing a context based method that automatically recommends fields that must be anonymized; 3) conducted experiments showing superior results using our approach compared to existing tools, and 4) developed an intuitive but basic user interface. The research that was conducted generated novel algorithmic techniques that utilize state-of-the-art methods such as condensation, differential privacy preservation, clustering, automated tuning based on contextual awareness, and recommendation techniques to specify columns to users for anonymization leading to optimal privacy that allows research analysis on the dataset. Experiments were conducted to evaluate the efficacy of these novel algorithmic techniques by performing analysis on original non-anonymized datasets, then conducting analysis on the same yet anonymized datasets and comparing the results of the analyses. Overall, the anonymized analysis results were within 1% of the original results, verifying that the generated technology not only guarantees a high level of privacy but also enables research analysis as if it were conducted on the original dataset. Potential applications of this technology include anonymization of any type of structured network datasets that contain sensitive identifiers, such as IP addresses, that can be used in multiple applications. For example, to create an AI or machine learning model for cyber security, e.g., to detect attacks, or for performance analysis, e.g., identify bottlenecks or predict performance. In addition, a market analysis that was conducted for potential applications of this technology identified a broader range of applications of our anonymization technology beyond the network sector that includes healthcare, banking, insurance, securities, finance (FISB), data brokering, cloud services, ad sales, and government.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Tiling Framework for Heterogeneous Computing of Matrix based Tiled Algorithms

Tiling matrix operations can improve the load balancing and performance of applications on heterogeneous computing resources. Writing a tile-based algorithm for each operation with a traditional, hand-tuned tiling approach that uses for loops in C/C++ is cumbersome and error prone. Moreover, it must enable and support the heterogeneous memory management of data objects and also explore architecture-supported, native, tiled-data transfer APIs instead of copying the tiled data to continuous memory before the data transfer. The tiling framework provides a tiled data structure for heterogeneous memory mapping and parameterization to a heterogeneous task specification API. We have integrated our tiled framework into MatRIS (Math kernels library using IRIS). IRIS is a heterogeneous run-time framework with a heterogeneous programming model, memory model, and task execution model. Experiments reveal that the tiled framework for BLAS operations has improved the programmability of tiled BLAS and improved performance by ~20% when compared against the traditional method that copies the data to continuous memory locations for heterogeneous computing.

Miniskar, Narasinga Rao↗
DOE OSTI2026

Selective Amnesia using Contrastive Subnet Erasure for Class Level Unlearning in Vision Models

We study concept-level forgetting in pretrained vision models: removing an entire semantic category so the system no longer recognizes that object in unseen images and contexts, rather than merely forgetting specific training examples. Prior work either applies blunt global projections or fine-tunes parameters, which can introduce collateral damage to unrelated features, add compute, and become unstable as forgetting strength increases. We introduce Contrastive Subnet Erasure (CSE), a training-free, encoder-centric edit that targets a compact set of channels most responsible for the class and attenuates them in a calibrated manner. The modification is algebraically folded into the subsequent layer, yielding no inference-time overhead and leaving task heads unchanged. To evaluate whether forgetting generalizes beyond the data used to specify the class, we introduce a cross dataset protocol in which the class is defined on a source dataset and performance is measured on a disjoint target dataset drawn from a different distribution with no shared images. This setup tests whether the model still fails to recognize the object when it looks different or appears in new scenes, and it helps avoid overfitting to patterns in the source dataset. Across CIFAR 10, CIFAR 100, and ImageNet under this protocol, CSE achieves stronger forgetting of the target class while better preserving non target utility than existing baselines in both single class and multi class settings. Overall, CSE provides a simple, stable, and deployment-ready mechanism for class-level unlearning in vision.

Kotevska, Olivera [ORNL] (ORCID:0000000316772243)↗