Engineering topics

Choudhury, Sutanay

Publications and source records attributed to Choudhury, Sutanay.

At least 19 records

DOE OSTI2025

Uncertainty quantification for neural network potential foundation models

Abstract For neural network potentials (NNPs) to gain widespread use, researchers must be able to trust model outputs. However, the blackbox nature of neural networks and their inherent stochasticity are often deterrents, especially for foundation models trained over broad swaths of chemical space. Uncertainty information provided at the time of prediction can help reduce aversion to NNPs. In this work, we detail two uncertainty quantification (UQ) methods. Readout ensembling, by finetuning the readout layers of an ensemble of foundation models, provides information about model uncertainty, while quantile regression, by replacing point predictions with distributional predictions, provides information about uncertainty within the underlying training data. We demonstrate our approach with the MACE-MP-0 model, applying UQ to the foundation model and a series of finetuned models. The uncertainties produced by the readout ensemble and quantile methods are demonstrated to be distinct measures by which the quality of the NNP output can be judged.

36 MATERIALS SCIENCE↗

DOE OSTI2024

Electronic structure simulations in the cloud computing environment

The transformative impact of modern computational paradigms and technologies, such as high-performance computing, quantum computing, and cloud computing, has opened up profound new opportunities for scientific simulations. Scalable computational chemistry is one beneficiary of this technological progress. The main focus of this paper is on the performance of various quantum chemical formulations, ranging from low-order methods to high-accuracy approaches, implemented in different computational chemistry packages, such as NWChem, NWChemEx, SPEC, ExaChem, and FLOSIC codes on the Azure Quantum Element (AQE) Microsoft cloud services. We pay particular attention to the intricate workflows for performing composite chemistry simulations, associated data curation, and mechanisms for accuracy assessment, as defined by the enabling cloud Computational Chemistry as a Service (CCaaS). Our focus also extends to Arrows' automated workflow for high throughput simulations. Finally, we provide a perspective on the role of cloud computing in supporting the mission of leadership computational facilities (LCFs).

computational chemistry, electronic structure, Clo↗

DOE OSTI2024

Evaluating Emerging AI/ML Accelerators: IPU, RDU, and and NVIDIA/AMD GPUs

As the size and complexity of AI/ML workloads continue to increase, the demand for specialized hardware accelerators has grown rapidly. To better understand the landscape of commercial AI/ML accelerators, we evaluate and compare several popular platforms, including Graphcore IPU, Sambanova RDU, and GPUs, on a range of AI workloads. Our research aims to help researchers and developers understand the unique features of commercial AI/ML accelerators and provide reference design and performance numbers for research prototypes. By shedding light on the current landscape of commercial AI/ML accelerators, we hope to offer insights into the future development of specialized hardware accelerators for AI/ML.

Peng, Hongwu↗

DOE OSTI2024

Acceleration of Graph Neural Network-Based Prediction Models in Chemistry via Co-Design Optimization on Intelligence Processing Units

Atomic structure prediction and associated property calculations are the bedrock of chemical physics. Since high-fidelity ab initio modeling techniques for computing the structure and properties can be prohibitively expensive, this motivates the development of machine-learning (ML) models that make these predictions more efficiently. Training graph neural networks over large atomistic databases introduces unique computational challenges such as the need to process millions of small graphs with variable size and support communication patterns that are distinct from learning over large graphs such as social networks. We demonstrate a novel hardware-software co-design approach to scale up the training of atomistic graph neural networks (GNN) for structure and property prediction. First, to eliminate redundant computation and memory associated with alternative padding techniques and to improve throughput via minimizing communication, we formulate the effective coalescing of the batches of variable-size atomistic graphs as the bin packing problem and introduce a hardware-agnostic algorithm to pack these batches. In addition, we propose hardware-specific optimizations including a planner and vectorization for the gather-scatter operations targeted for Graphcore’s Intelligence Processing Unit (IPU), as well as model-specific optimizations such as merged communication collectives and optimized softplus. Putting these all together, we demonstrate the effectiveness of the proposed co-design approach by providing an implementation of a well-established atomistic GNN on the Graphcore IPUs. We evaluate the training performance on multiple atomistic graph databases with varying degrees of graph counts, sizes and sparsity. Here, we demonstrate that such a co-design approach can reduce the training time of atomistic GNNs and can improve the performance by up to 1.5× compared to the baseline implementation of the model on the IPUs. Additionally, we compare our IPU implementation with a Nvidia GPU-based implementation and show that our atomistic GNN implementation on the IPUs can run 1.8× faster on average compared to the execution time on the GPUs.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2024

pnnl/chemreasoner

ChemReasoner - Catalyst Discovery via Large Language Model-driven Reasoning

Choudhury, Sutanay↗

DOE OSTI2024

High dimensional predictions of suicide risk in 4.2 million US Veterans using ensemble transfer learning

We present an ensemble transfer learning method to predict suicide from Veterans Affairs (VA) electronic medical records (EMR). A diverse set of base models was trained to predict a binary outcome constructed from reported suicide, suicide attempt, and overdose diagnoses with varying choices of study design and prediction methodology. Each model used twenty cross-sectional and 190 longitudinal variables observed in eight time intervals covering 7.5 years prior to the time of prediction. Ensembles of seven base models were created and fine-tuned with ten variables expected to change with study design and outcome definition in order to predict suicide and combined outcome in a prospective cohort. The ensemble models achieved c-statistics of 0.73 on 2-year suicide risk and 0.83 on the combined outcome when predicting on a prospective cohort of ~4.2 M veterans. The ensembles rely on nonlinear base models trained using a matched retrospective nested case-control (Rcc) study cohort and show good calibration across a diversity of subgroups, including risk strata, age, sex, race, and level of healthcare utilization. In addition, a linear Rcc base model provided a rich set of biological predictors, including indicators of suicide, substance use disorder, mental health diagnoses and treatments, hypoxia and vascular damage, and demographics. Similar content being viewed by others

60 APPLIED LIFE SCIENCES↗

DOE OSTI2023

Monte Carlo Thought Search: Large Language Model Querying for Complex Scientific Reasoning in Catalyst Design

Discovering novel catalysts requires complex reasoning involving multiple chemical properties and resultant trade-offs, leading to a combinatorial growth in the search space. While large language models (LLM) have demonstrated novel capabilities for chemistry through complex instruction following capabilities and high quality reasoning, a goal-driven combinatorial search using LLMs has not been explored in detail. In this work, we present a Monte Carlo Tree Search-based approach that improves beyond state-of-the-art chain-of-thought prompting variants to augment scientific reasoning. We introduce two new reasoning datasets: 1) a curation of computational chemistry simulations, and 2) diverse questions written by catalysis researchers for reasoning about novel chemical conversion processes. We improve over the best baseline by 25.8\% and find that our approach can augment scientist's reasoning and discovery process with novel insights.\footnote{All resources will be publicly available upon publication.

Sprueill, Henry W.↗

DOE OSTI2023

Report on Pure Mathematics of Machine Learning

We report on the findings conducted by the Pure Mathematics in Machine Learning (PMML) Project under the umbrella of the Mathematics in Artificial Reasoning in Sciences (MARS) Laboratory Initiative at the Pacific Northwest National Laboratory. The findings contained in this report spans three years of research from the project team. The objectives of the PMML project are two-fold: 1. To gain understanding of modern machine learners through empirical research via structured and sound experimental design, and 2. To prove mathematical theories based upon these empirical findings. We present both empirical results along with mathematical theories in this document.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2023

pnnl/projection_ntk

This is a repository for work conducted in our research paper we would like to make open source. The main features are that we forked another open source library and added functionality that enabled us to compute neural-kernel objects using random vector projection software for both computer vision and large language text models. The new features are specifically code that enables the calculation of objects we define and refer to as the projection pseudo neural tangent kernel and the projection trace neural tangent kernel; both of which are approximations of the empirical neural tangent kernel. The software that computes the projections themselves are not part of the invention, but are necessarily tied to our software and licensed via MIT license, and BSD 2 license.

Georgiev, Kristian↗

DOE OSTI2023

Software Defined Architectures for Portability and Performance

The Software Defined Architectures for Portability and Performance (SODAPOP) project developed a co-design framework to partition and map converged applications on specialized heterogeneous architectures. We started from key domain applications that combine scientific simulation with data analytics and machine learning as drivers to integrate our framework. The framework includes high-level compilers that interfaces with high-level programming frameworks, domain-specific optimization passes, and hardware-oriented optimizations. The framework leverages hardware generators to enable specialization and facilitate exploration of custom system designs.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2023

A Unification Framework for Euclidean and Hyperbolic Graph Neural Networks

Hyperbolic neural networks have recently gained significant attention due to their promising results on several graph problems including node classification and link prediction. The primary reason for this success is the effectiveness of hyperbolic space in capturing the inherent hierarchy of graph datasets. However, they are limited in terms of generalization, scalability, and have inferior performance when applied to non-hierarchical datasets. In this paper, we take a completely different perspective for modeling hyperbolic networks and answer the following question: is an Euclidean model able to approximate a function or behavior in the hyperbolic space? Extending the universal approximation theory developed for Euclidean models, We draw an analogy from the hyperbolic components to the Euclidean counterparts and conclude that, in order to capture hierarchical features, it is possible to generalize hyperbolic models to be a special case of Euclidean models with the proposed Pseudo-Poincaré technique. We applied our non-linear hyperbolic normalization to the current state-of-the-art homogeneous and multi-relational graph networks and demonstrate significant improvements in performance compared to both Euclidean and hyperbolic counterparts. The primary impact of this work lies in its ability to capture hierarchical features in the Euclidean space, and thus, can replace hyperbolic networks without any loss in performance metrics while simultaneously leveraging the power of Euclidean networks such as interpretability and efficient execution of various model components.

Khatir, Mehrdad↗

DOE OSTI2023

Cloud Services Enable Efficient AI-Guided Simulation Workflows across Heterogeneous Resources

Applications which fuse machine learning and simulation are rarely best served by a single computing resource. Highly parallel simulation codes are best deployed on super- computers, while AI tasks used to decide which simulations to perform may be best suited to specialized accelerators. Here we present a Function-as-a-Service (FaaS) system for executing complex, distributed computational campaigns that achieves performance parity with conventional workflow systems without the complexities of secure network connections between compute providers. One innovation enabling high performance is a subsystem that directly moves task data between sites, separate from the cloud-hosted FaaS system used to distribute task instructions. We also introduce a flexible scheduling system that allows us access factor of 2 trade offs between the amount of resources required to solve a problem at each compute site. We anticipate that this system will upgrade multi-site applications from demonstration projects to routine practice in computational science.

Ward, Logan↗

DOE OSTI2023

BitGNN: Unlocking the Performance Potential of Binary Graph Neural Networks on GPUs

Graph Neural Networks (GNNs) have shown compelling results in many graph-based learning tasks. They are, however, time-consuming. Recent work has shown a promising direction in improving GNN speed and shrinking the size — network binarization, which binarizes network values and operations. Prior work, however, mainly focused on algorithm designs, leaving it open on how to fully materialize the performance potential. This work fills the gap by proposing techniques to best map binary GNNs and their computations to fit the nature of bit manipulations, optimizations and algorithms to maximize BSpMM kernel efficiency, and solutions to other factors influencing the end-to-end time on GPUs. Results on real-world graphs show that the proposed techniques outperform state of-the-art binary GNN implementations by 21-67× with little accuracy loss.

Chen, Jou-An↗

DOE OSTI2022

Learning Distributed Geometric Koopman Operator for Sparse Networked Dynamical Systems

Koopman operator theory provides an alternative to study nonlinear networked dynamical systems by mapping the state space to an abstract higher dimensional space where the system evolution is linear. Recent works show the application of graph neural networks (GNNs) to learn state to object-centric embeddings and achieve centralized block-wise computation of Koopman operator (KO) under additional assumptions on the underlying agents properties and constraints on the KO structure. However, the computational complexity of learning the Koopman increases exponentially for networked systems where the number of possible system states grows in a combinatorial fashion with the number of nodes. The learning challenge is further amplified for sparse networks by two factors: 1) sample sparsity for learning the Koopman operator in the non-linear space, and 2) the divergence in the dynamics of individual nodes or from one subgraph to another. Our work aims to address these challenge by formulating the representation learning of networked dynamical systems into a multi-agent paradigm and learning the Koopman operator in a distributive manner. The computational as well as performance advantages of distributed Koopman is predominant for sparse networks whereas for fully connected networks, it is shown to coincide with the centralized one. The empirical study on rope system, network of oscillators and a synthetic power system show comparable and superior performance along with computational benefits with the state-of-the-art methods.

Mukherjee, Sayak↗

DOE OSTI2022

Identification of Novel, Replicable Genetic Risk Loci for Suicidal Thoughts and Behaviors Among US Military Veterans

Importance: Suicide is a leading cause of death; however, the molecular genetic basis of suicidal thoughts and behaviors (SITB) remains unknown. Objective: To identify novel, replicable genomic risk loci for SITB. Design, Setting, and Participants: This genome-wide association study included 633 778 US military veterans with and without SITB, as identified through electronic health records. GWAS was performed separately by ancestry, controlling for sex, age, and genetic substructure. Cross-ancestry risk loci were identified through meta-analysis. Study enrollment began in 2011 and is ongoing. Data were analyzed from November 2021 to August 2022. Main Outcome and Measures: SITB. Results: A total of 633 778 US military veterans were included in the analysis (57 152 [9%] female; 121 118 [19.1%] African ancestry, 8285 [1.3%] Asian ancestry, 452 767 [71.4%] European ancestry, and 51 608 [8.1%] Hispanic ancestry), including 121 211 individuals with SITB (19.1%). Meta-analysis identified more than 200 GWS (P < 5 × 10 -8 ) cross-ancestry risk single-nucleotide variants for SITB concentrated in 7 regions on chromosomes 2, 6, 9, 11, 14, 16, and 18. Top single-nucleotide variants were largely intronic in nature; 5 were independently replicated in ISGC, including rs6557168 in ESR1, rs12808482 in DRD2, rs77641763 in EXD3, rs10671545 in DCC, and rs36006172 in TRAF3. Associations for FBXL19 and AC018880.2 were not replicated. Gene-based analyses implicated 24 additional GWS cross-ancestry risk genes, including FURIN, TSNARE1, and the NCAM1-TTC12-ANKK1-DRD2 gene cluster. Cross-ancestry enrichment analyses revealed significant enrichment for expression in brain and pituitary tissue, synapse and ubiquitination processes, amphetamine addiction, parathyroid hormone synthesis, axon guidance, and dopaminergic pathways. Seven other unique European ancestry–specific GWS loci were identified, 2 of which (POM121L2 and METTL15/LINC02758) were replicated. Two additional GWS ancestry-specific loci were identified within the African ancestry (PET112/GATB) and Hispanic ancestry (intergenic locus on chromosome 4) subsets, both of which were replicated. Further, no GWS loci were identified within the Asian ancestry subset; however, significant enrichment was observed for axon guidance, cyclic adenosine monophosphate signaling, focal adhesion, glutamatergic synapse, and oxytocin signaling pathways across all ancestries. Within the European ancestry subset, genetic correlations (r > 0.75) were observed between the SITB phenotype and a suicide attempt-only phenotype, depression, and posttraumatic stress disorder. Additionally, polygenic risk score analyses revealed that the Million Veteran Program polygenic risk score had nominally significant main effects in 2 independent samples of veterans of European and African ancestry.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2022

pnnl/TransMED

Transfer Learning From Existing Diseases Via Hierarchical Multi-Modal BERT Models

Ham, Colby↗

DOE OSTI2022

Attention-based Aspect Reasoning for Knowledge Base Question Answering on Clinical Notes

Question Answering (QA) in clinical notes has gained a lot of attention in the past few years. Existing machine reading comprehension approaches in clinical domain can only handle questions about a single block of clinical texts and fail to retrieve information about different patients and clinical notes. To handle more complex questions, we aim at creating knowledge base from clinical notes to link different patients and clinical notes, and performing knowledge base question answering (KBQA). Based on the expert annotations in n2c2, we first created the ClinicalKBQA dataset that includes 8,952 QA pairs and covers questions about seven medical topics through 322 question templates. Then, we proposed an attention-based aspect reasoning (AAR) method for KBQA and investigated the impact of different aspects of answers (e.g., entity, type, path, and context) for prediction. The AAR method achieves better performance due to the well-designed encoder and attention mechanism. In the experiments, we find that both aspects, type and path, enable the model to identify answers satisfying the general conditions and produce lower precision and higher recall. On the other hand, the aspects, entity and context, limit the answers by node-specific information and lead to higher precision and lower recall.

Wang, Ping↗

DOE OSTI2022

Preparing for the next pandemic via transfer learning from existing diseases with hierarchical multi-modal BERT: a study on COVID-19 outcome prediction

Abstract Developing prediction models for emerging infectious diseases from relatively small numbers of cases is a critical need for improving pandemic preparedness. Using COVID-19 as an exemplar, we propose a transfer learning methodology for developing predictive models from multi-modal electronic healthcare records by leveraging information from more prevalent diseases with shared clinical characteristics. Our novel hierarchical, multi-modal model ( $${\textsc {TransMED}}$$ T R A N S MED ) integrates baseline risk factors from the natural language processing of clinical notes at admission, time-series measurements of biomarkers obtained from laboratory tests, and discrete diagnostic, procedure and drug codes. We demonstrate the alignment of $${\textsc {TransMED}}$$ T R A N S MED ’s predictions with well-established clinical knowledge about COVID-19 through univariate and multivariate risk factor driven sub-cohort analysis. $${\textsc {TransMED}}$$ T R A N S MED ’s superior performance over state-of-the-art methods shows that leveraging patient data across modalities and transferring prior knowledge from similar disorders is critical for accurate prediction of patient outcomes, and this approach may serve as an important tool in the early response to future pandemics.

59 BASIC BIOLOGICAL SCIENCES↗