Search NASA⌕ Search

SEARCH · Search NASA

Results for “AI Optimization”

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 199 records · Page 11
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Automated AI-driven Molecular Design for Therapeutic Discovery

In recent years, artificial intelligence and machine learning (AI/ML) approaches have revolutionized the process of designing new therapeutics, enabling scientists to rapidly respond to emerging threats from various pathogens. A prime example is the SARS-CoV-2 main protease, a key target for the development of antiviral inhibitors. In this study, we employed a novel, integrated approach that combines AI-driven iterative design of inhibitor candidates, screening based on physio-chemical properties and toxicity, physics-based computational modeling of protein-inhibitor interactions, and AI-assisted analysis of Native MS biophysical assay and characterization of designed candidates. Our deep learning 3D-scaffold model, which uses an input scaffold as a starting point, generated tens of thousands of compounds while preserving the key scaffold. To optimize these candidates, we calculated a comprehensive set of 136 descriptors, including both 2D and 3D molecular features, for compounds targeting the SARS-CoV-2 Main protease (Mpro) and a neurodegenerative disease-associated protein, cyclophilin (Cyp). The generated compounds were initially filtered based on their properties and then ranked according to their predicted binding affinity using our automated modeling and ML methods. Experimental validation of the Mpro candidates showing inhibitory activity demonstrates that our workflow can expedite the therapeutic discovery.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

SIDDA: SInkhorn Dynamic Domain Adaptation for image classification with equivariant neural networks

Modern neural networks (NNs) often do not generalize well in the presence of a ‘covariate shift’; that is, in situations where the training and test data distributions differ, but the conditional distribution of classification labels given the data remains unchanged. In such cases, NN generalization can be reduced to a problem of learning more robust, domain-invariant features. Domain adaptation (DA) methods include a broad range of techniques aimed at achieving this; however, these methods have struggled with the need for extensive hyperparameter tuning, which then incurs significant computational costs. In this work, we introduce SInkhorn Dynamic Domain Adaptation (SIDDA), an out-of-the-box DA training algorithm built upon the Sinkhorn divergence, that can achieve effective domain alignment with minimal hyperparameter tuning and computational overhead. We demonstrate the efficacy of our method on multiple simulated and real datasets of varying complexity, including simple shapes, handwritten digits, real astronomical observations, and remote sensing data. These datasets exhibit covariate shifts due to noise, blurring, differences between telescopes, and variations in imaging wavelengths. SIDDA is compatible with a variety of NN architectures, and it works particularly well in improving classification accuracy and model calibration when paired with symmetry-aware equivariant NNs (ENNs). We find that SIDDA consistently enhances the generalization capabilities of NNs, achieving up to a ${\approx}40\%$ improvement in classification accuracy on unlabeled target data, while also providing a more modest performance gain of $\lesssim 1\%$ on labeled source data. We also study the efficacy of DA on ENNs with respect to the varying group orders of the dihedral group DN, and find that the model performance improves as the degree of equivariance increases. Finally, if SIDDA achieves proper domain alignment, it also enhances model calibration on both source and target data, with the most significant gains in the unlabeled target domain—achieving over an order of magnitude improvement in the expected calibration error and Brier score. SIDDA’s versatility across various NN models and datasets, combined with its automated approach to domain alignment, has the potential to significantly advance multi-dataset studies by enabling the development of highly generalizable models.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2025

Microreactor Optimization Using Simulation And Economics (mouse)

Microreactor Optimization Using Simulation and Economics (MOUSE) is a tool that integrates both nuclear microreactor design and reactor economics to provide comprehensive evaluations and optimizations. This tool enables stakeholders to explore the interplay between technical and economic variables, guiding them towards effective and competitive microreactor solutions. For the reactor core simulations, MOUSE leverages the OpenMC Monte Carlo Particle Transport Code to perform detailed core simulations for various microreactor designs. The included OpenMC models are 2D core designs of a Liquid Metal Thermal Microreactor (LMTR), a Gas-Cooled TRISO-Fueled Microreactor (GCMR), and a Heat Pipe Microreactor. Beyond core design, MOUSE includes simplified calculations for: - Calculating the masses of heat exchangers within the system. - Mechanical power of pumps. - Estimating the area occupied by various buildings within the nuclear plant. For the economic analysis, MOUSE provides detailed bottom-up cost estimates, encompassing a wide range of costs including preconstruction costs, direct costs, indirect costs, training costs, financial costs, operation & maintenance (O&M) costs, and fuel costs. These cost estimations are developed using data from the MARVEL project and additional literature sources, enabling the calculation of total capital costs and levelized cost of energy for both first-of-a-kind and nth-of-a-kind microreactors. MOUSE also enables analysis of the cost drivers and competitiveness in the electricity market. MOUSE allows users to modify a wide array of technical and economic parameters to evaluate different scenarios and their impacts. Examples of these parameters include: Fuels, coolants, or reflector materials Enrichment levels Control drum materials and geometry Fuel pin geometry and materials Moderator pin geometry and materials Reactor core and reflector dimensions Packing factor for the TRISO particles Nuclear reactor power and reactor burnup Number of sensors Shielding thickness Reactor vessel and guard vessel dimensions Operational staff requirements Number of emergency shutdowns Levelization period Interest rate Construction duration Since MOUSE is powered by the WATTS toolkit, it supports optimization studies, parametric analyses, and uncertainty calculations/propagation. The optimization techniques enable users to identify optimal design and economic configurations. The parametric analysis tools allow users to explore the sensitivity of various parameters, while uncertainty propagation helps quantify the impact of uncertainties on overall performance and cost. User Interface and Workflow: Currently, MOUSE is a command-line-based tool. Users can input various reactor design or economic parameters, modify the designs, run simulations, and visualize results through comprehensive data visualization and reporting capabilities. The typical workflow involves setting up the reactor model, defining economic parameters, running simulations, and analyzing the results to make informed decisions. By combining advanced design calculations with detailed economic modeling, MOUSE provides a robust framework for optimizing nuclear microreactor technologies, enhancing their competitiveness, and guiding stakeholders towards innovative and cost-effective solutions.

Hanna, Botros [Idaho National Laboratory (INL), Id↗
NASA NTRS1983

Airborne Imaging Spectrometer - An advanced concept instrument

The Airborne Imaging Spectrometer (AIS) is a multispectral infrared imaging instrument utilizing a 32- x 32-element mercury-cadmium-telluride detector array to achieve high spatial resolution. A novel optical design provides high spectral resolution, and, in conjunction with the 32-element spatial resolution, makes optimal use of the capabilities of the area array. The experiment requirements and the details of instrument design, implementation, and calibration are presented. The superior resolution, accuracy, and data acquisition rate of this sensor/instrument concept for reconnaissance and geophysical survey is detailed and plans for upgrading/improving the instrument and future applications of this design concept are discussed.

Labaw, C.↗
DOE OSTI2026

FAIR Data and Interpretable AI Framework for Architectured Metamaterials

Our interdisciplinary effort successfully generated FAIR (Findable, Accessible, Interoperable, and Reusable) benchmark datasets for mechanical metamaterials while introducing a novel Artificial Intelligence (AI) framework known as Learning Refined Compositional Rules (LRCR). This framework was specifically designed to bridge the gap across varying computational length scales and extract the underlying physical mechanisms that connect a material's structural geometry to its bulk acoustic properties. Historically, the discovery of such structured materials relied heavily on human intuition or opaque, black-box optimization algorithms that were difficult to generalize. By combining interpretable machine learning techniques with rigorous experimental validation, this project established clear, generalizable design guidelines for tuning wave dispersion and controlling vibrations. Ultimately, the public availability of these structured datasets and algorithms will significantly reduce computational costs and accelerate the design of advanced multi-functional acoustic devices, offering broad societal impacts across fields like aerospace engineering, telecommunications, and biomedical implant design.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Comprehensive AI-based System for Control, Sensor Estimation, and Fault Detection of Cascaded Multilevel Inverters

In this paper, an Artificial Intelligence-based (AI) system is proposed for an 11-level cascaded H-bridge multilevel inverter (MLI) with the aims of harmonic suppression and reliability enhancement. The system consists of three seamlessly integrated Neural Networks (NNs). First, a multilayer perceptron is used to generalize the optimal switching angles for selective harmonic elimination under non-equal DC voltages. Next, an autoencoder NN estimates the voltage sensor readings to address potential drifting. Finally, a perceptron NN detects inverter faults based solely on the output voltage of the MLI. Simulation scenarios were evaluated, and the results show that the proposed system provides a comprehensive solution for the robust operation of the MLI. The proposed solution is capable of minimizing the targeted harmonics orders with minimal impact on the fundamental voltage, even when the voltage sensor drifts. Furthermore, the inverter under fault conditions was successfully identified.

Rezende da Costa Reis Kimpara, Renata↗
DOE OSTI2026

TETA Autoresearch [SWR-26-089]

TETA Autoresearch is a template repository based on github.com/karpathy/autoresearch for AI-assisted research science in the TETA group in the Center for Integrated Mobility Sciences (CIMS) center at the National Laboratory of the Rockies. This software is a template for running autonomous research experiments that iteratively improve an ML model for a single optimization objective. Two execution modes share one harness: LLM mode - an agent (e.g. Claude Code) edits a scaffold train.py one change at a time, tagging each experiment, logging reasoning, and pushing results. Defined by program.md. Optimizer mode - an Optuna-backed driver (TPE / CMA-ES / Random) iterates over a domain-defined search space. Defined by optimizers/. RouteE (vehicle energy prediction) is the reference domain under domains/routee/. Adding a new domain is mechanical - see EXTENDING.md.

Reinicke, Nicholas [National Laboratory of the Roc↗
DOE OSTI2026

Bayesian Optimization of Catalysis with In-Context Learning

Large language models (LLMs) can perform accurate classification with zero or few examples through in-context learning (ICL), allowing the model to observe query-relevant examples at inference time and eliminating the need for additional weight updates to generalize beyond its original training data. We extend this capability to regression with uncertainty estimation using frozen LLMs (e.g., GPT-4o, Gemini), enabling Bayesian optimization (BO) in natural language without explicit model training or feature engineering. We apply this to materials discovery by representing materials as synthesis and testing procedures for use in natural language prompts. This Bayesian, design-first approach prioritizes optimization toward target material properties before detailed characterization, in contrast to conventional experimental workflows that often emphasize characterization of suboptimal materials. On benchmarks like aqueous solubility and oxidative coupling of methane (OCM), BO-ICL matches or outperforms Gaussian processes. In live experiments on the reverse water–gas shift (RWGS) reaction, BO-ICL identifies multimetallic catalysts that approach equilibrium CO yield within 6 and 10 iterations from a pool of 3,700 and 360,000 candidates, respectively. Our method redefines materials representation and accelerates discovery, with broad applications across catalysis, materials science, and AI.

Calibration↗
DOE OSTI2024

Analyzing inference workloads for spatiotemporal modeling

Ensuring power grid resiliency, forecasting climate conditions, and optimization of transportation infrastructure are some of the many application areas where data is collected in both space and time. Spatiotemporal modeling is about modeling those patterns for forecasting future trends and carrying out critical decision-making by leveraging machine learning/deep learning. Once trained offline, field deployment of trained models for near real-time inference could be challenging because performance can vary significantly depending on the environment, available compute resources and tolerance to ambiguity in results. Users deploying spatiotemporal models for solving complex problems can benefit from analytical studies considering a plethora of system adaptations to understand the associated performance-quality trade-offs. To facilitate the co-design of next-generation hardware architectures for field deployment of trained models, it is critical to characterize the workloads of these deep learning (DL) applications during inference and assess their computational patterns at different levels of the execution stack. In this paper, we develop several variants of deep learning applications that use spatiotemporal data from dynamical systems. We study the associated computational patterns for inference workloads at different levels, considering relevant models (Long short-term Memory, Convolutional Neural Network and Spatio-Temporal Graph Convolution Network), DL frameworks (Tensorflow and PyTorch), precision (FP16, FP32, AMP, INT16 and INT8), inference runtime (ONNX and AI Template), post-training quantization (TensorRT) and platforms (Nvidia DGX A100 and Sambanova SN10 RDU). Overall, our findings indicate that although there is potential in mixed-precision models and post-training quantization for spatiotemporal modeling, extracting efficiency from contemporary GPU systems might be challenging. Instead, co-designing custom accelerators by leveraging optimized High Level Synthesis frameworks (such as SODA High-Level Synthesizer for customized FPGA/ASIC targets) can make workload-specific adjustments to enhance the efficiency.

97 MATHEMATICS AND COMPUTING↗
NASA NTRSDate not supplied

Predicting Team Functioning in Long Term Space Missions Using Acoustic and Linguistic Measures

Maintaining optimal team functioning is critical for long-duration space exploration missions, yet traditional monitoring methods, such as self-reports and wearable sensors, often impose operational burdens or suffer from bias. This paper investigates a non-intrusive speech-based artificial intelligence (AI) framework to predict degradations in team functioning using data from the Human Exploration Research Analog (HERA) of the U.S. National Aeronautics and Space Administration (NASA). Using acoustic features, linguistic descriptors, and semantic embeddings, we evaluate static non-linear and temporal machine learning models to predict both objective (task accuracy) and subjective (self-reported efficacy and cohesion) team functioning outcomes. Results indicate that temporal models outperform static approaches, with prediction of objective task accuracy in Team Interaction Battery (TIB) improving from near chance to 71%. Self-reported outcomes, including team efficacy and cohesion, are predicted more reliably than task performance, achieving balanced accuracies of up to 85.56% and 78.12%, respectively, and are found to be most strongly associated with acoustic features. In a second interdependent task, the MMSEV–EVA, accuracies of up to 78% are achieved using temporal models with acoustic features. Furthermore, incorporating just 1–2 days of team-specific historical data systematically improved performance, and acoustic markers from informal pre-task interactions provided modest predictive gains. Finally, while automated preprocessing yielded viable accuracy, humancorrected data provided moderate performance gains, though transcription error rates did not significantly correlate with model performance. These findings highlight the potential of speech as a passive, high-fidelity monitoring tool for autonomous habitats.

Temporal modeling↗
DOE OSTI2025

Artificial Intelligence for Data Center Operations (AIOps): Cooperative Research and Development (Final Report)

High performance computing data centers will increasingly need to rely on automation to keep pace with exascale growth in compute capability and to manage and optimize the data center environment and facility resources. Artificial intelligence and machine learning approaches provide the means to improve HPC data center operational efficiency, by learning historical trends and training models to operate on real-time data collected from both IT and facilities sources. NREL has developed methods of real-time collection, aggregation and streaming of these data in the ESIF HPC Data Center and has collected a significant dataset of relevant metrics across computer systems, racks, environmental, building and utility sources for research into various predictive analytics problems. HPE's Advanced Technology Group (ATG) is doing comprehensive research into exascale monitoring and management for High Performance Computing (HPC) systems (hereinafter HPE's Data Monitoring/ Management Technology). NREL and HPE will collaborate to add Artificial Intelligence (AI) to NREL's real-time data collection/ aggregation/ streaming system and HPE's Data Monitoring/ Management System, with the goal of improving the operational efficiency of NREL's Energy Systems Integration Facility (ESIF) HPC Data Center through data analytics on both historical and real-time data from IT systems and facilities operations. This collaboration will consist of efforts in Data Management, Data Analytics, and AI/ML Optimization for both manual and autonomous intervention in data center operations. This will be a multi-year, multi-staged effort with a goal towards building capabilities for an Advanced Smart Facility, and demonstration of these techniques in the NREL ESIF HPC Data Center.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Discovery of hydrogen storage molecules using large language models and machine learning

Accelerating the discovery of new molecules with targeted properties is a central challenge in molecular design. In this contribution, we present an AI-driven molecular discovery framework that integrates Large Language Models (LLMs) for generative molecular design with Machine Learning (ML)-based screening to identify novel Liquid Organic Hydrogen Carrier (LOHC) candidates. Using the developed framework, LOHC molecules were systematically generated, evaluated, and refined iteratively, combining LLM-guided molecular generation and ML-predicted hydrogenation enthalpies (Δ H ), under physicochemical property constraints such as optimal melting points (MP), desired hydrogen storage capacity (wt% H 2 ), and synthetic accessibility (SA) scores. This approach enabled the discovery of 42 new LOHC candidates in two distinct campaigns, one seeded with experimentally known and another with previously computationally identified LOHCs, respectively. Although we began with different numbers of starting molecules (31 vs . 7 seed molecules), both runs yielded a comparable number of viable candidates, suggesting an influence of chemically intuitive seed molecule selection for success. Selected LOHC molecules, such as 3-methyl pyridine, 1-ethylnapthalene, 1,1-diphenylethane, and benzofuran, were experimentally tested and compared with benchmark LOHCs (toluene and 9-ethylcarbazole) for hydrogenation using a series of commercial supported metal catalysts. The order of conversion into fully hydrogenated products at 200 °C was 3-methyl pyridine (100%) > 9-ethyl carbazole (86.4%) > 2,3-benzofuran (74%) > 1,1-diphenylethane (66.9%) > 1-ethylnapthalene (66.7%) > toluene (57%), further validating the AI-guided molecular design. This study demonstrates promise of LLM-driven molecular design in conjunction with ML-based screening for accelerated discovery and design of molecules.

Harb, Hassan [Argonne National Laboratory (ANL), A↗
DOE OSTI2025

Scalable Data Center Capacity for DOE's AI Prototype: A Rapidly Available Gigawatt Data Center for DOE

The multilaboratory Gigawatt Data Center working group was commissioned to identify approaches to rapidly establish federal data centers with scalable capacities up to 1,000 MW. These state-of-the-art facilities will serve as hubs for interdisciplinary collaboration, industry partnerships, and transformative applications of artificial intelligence. The proposed strategic shift includes facilitating multilaboratory collaboration, prioritizing operational efficiency, expanding public–private partnerships, optimizing investments, ensuring long-term contractual flexibility, supporting open science and secure data enclaves, and exploiting high-speed national networks. Owing to their extensive experience and best practices, the US Department of Energy national laboratories are uniquely positioned to lead this initiative. We recommend conducting a feasibility analysis to rapidly identify the optimal sites for this initiative, and the effort will likely involve private industry for design, construction, financing, and operational integration. We also propose establishing multiple geographically diverse sites to ensure energy resilience, high operational reliability, and a diverse user base, thereby effectively addressing the nation’s critical needs.

42 ENGINEERING↗
DOE OSTI2025

The ICEBERG Test Stand for DUNE Cold Electronics Development

ICEBERG is a liquid argon time projection chamber at Fermilab for the purpose of testing detector components and software for the Deep Underground Neutrino Experiment (DUNE). The detector features a 1.15m x 1m anode plane following the specifications of the DUNE horizontal drift far detector and a newly installed X-ARAPUCA photodetector. The status of ICEBERG is reported along with analysis of noise, pulser, and cosmic ray data from the ninth run beginning May 2024 with the goal of advising the DUNE collaboration on the optimal wire readout electronics configuration. In addition, development of an absolute energy scale calibration method is currently underway using known sources such as cosmic ray muon Michel electrons at the ~10 MeV scale and $^{39}$Ar decay electrons at the ~100keV scale. Research into AI-based identification of such events at the data acquisition level is introduced.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2024

Real-time High-resolution X-Ray Computed Tomography

Computed Tomography (CT) serves as a key imaging technology that relies on computationally intensive filtering and back-projection algorithms for 3D image reconstruction. While conventional high-resolution image reconstruction (> 2K3) solutions provide quick results, they typically treat reconstruction as an offline workload to be performed remotely on large-scale HPC systems. The growing demand for post-construction AI-driven analytics and the need for real-time adjustments call for high-resolution reconstruction solutions that are feasible on local computing resources, i.e. a multi-GPU server at most. In this paper, we propose a novel approach that utilizes Tensor Cores to optimize image reconstruction without sacrificing precision. We also introduce a framework designed to enable real-time execution of end-to-end distributed image reconstruction in a multi-GPU environment. Evaluations conducted on a single Nvidia A100 and H100 GPU show performance improvements of 1.91 × and 2.15 × compared to highly optimized production libraries. Furthermore, our framework, when deployed on 8-card Nvidia A100 GPU system, demonstrates the ability to reconstruct real-world datasets into 20483 volumes (32 GB) in slightly more than one minute and 40963 volumes (256 GB) in 7 minutes.

Wu, Du↗
DOE OSTI2026

Beyond Component Optimization: Systems Level Biodesign for Lanthanide Recovery

Global demand for lanthanides (Ln) is projected to rise sharply over the next decade, while geographically concentrated supply chains and the low concentrations and matrix complexity of secondary feedstocks limit the reach of conventional hydro- and pyrometallurgical separation. Engineered biological systems offer a selective, low-energy alternative, and component-level advances in Ln-binding proteins, AI-designed selective scaffolds, and cell-surface display platforms now rival synthetic chelators in affinity and selectivity. These components, however, remain functionally isolated. Currently, there are no engineered chassis coupling recognition, intracellular trafficking, accumulation, and controlled release into an end-to-end pipeline. Here, we outline how new biodesign strategies and chassis selection must move beyond bioleaching to encompass the full recovery pathway. Achieving this requires integrating AI/ML-guided design, genome-scale build tools, high-throughput phenotyping, and biophysical transport modeling within a Design–Build–Test–Learn cycle tuned to recognition, trafficking, accumulation, and release.

Biodesign↗
DOE OSTI2025

AI-powered topic modeling: comparing LDA and BERTopic in analyzing opioid-related cardiovascular risks in women

Topic modeling is a crucial technique in natural language processing (NLP), enabling the extraction of latent themes from large text corpora. Traditional topic modeling, such as Latent Dirichlet Allocation (LDA), faces limitations in capturing the semantic relationships in the text document although it has been widely applied in text mining. BERTopic, created in 2022, leveraged advances in deep learning and can capture the contextual relationships between words. In this work, we integrated Artificial Intelligence (AI) modules to LDA and BERTopic and provided a comprehensive comparison on the analysis of prescription opioid-related cardiovascular risks in women. Opioid use can increase the risk of cardiovascular problems in women such as arrhythmia, hypotension etc. 1,837 abstracts were retrieved and downloaded from PubMed as of April 2024 using three Medical Subject Headings (MeSH) words: “opioid,” “cardiovascular,” and “women.” Machine Learning of Language Toolkit (MALLET) was employed for the implementation of LDA. BioBERT was used for document embedding in BERTopic. Eighteen was selected as the optimal topic number for MALLET and 23 for BERTopic. ChatGPT-4-Turbo was integrated to interpret and compare the results. The short descriptions created by ChatGPT for each topic from LDA and BERTopic were highly correlated, and the performance accuracies of LDA and BERTopic were similar as determined by expert manual reviews of the abstracts grouped by their predominant topics. The results of the t-SNE (t-distributed Stochastic Neighbor Embedding) plots showed that the clusters created from BERTopic were more compact and well-separated, representing improved coherence and distinctiveness between the topics. Our findings indicated that AI algorithms could augment both traditional and contemporary topic modeling techniques. In addition, BERTopic has the connection port for ChatGPT-4-Turbo or other large language models in its algorithm for automatic interpretation, while with LDA interpretation must be manually, and needs special procedures for data pre-processing and stop words exclusion. Therefore, while LDA remains valuable for large-scale text analysis with resource constraints, AI-assisted BERTopic offers significant advantages in providing the enhanced interpretability and the improved semantic coherence for extracting valuable insights from textual data.

Research & Experimental Medicine↗
NASA NTRS2004

The Role of Ontologies in Schema-based Program Synthesis

Program synthesis is the process of automatically deriving executable code from (non-executable) high-level specifications. It is more flexible and powerful than conventional code generation techniques that simply translate algorithmic specifications into lower-level code or only create code skeletons from structural specifications (such as UML class diagrams). Key to building a successful synthesis system is specializing to an appropriate application domain. The AUTOBAYES and AUTOFILTER systems, under development at NASA Ames, operate in the two domains of data analysis and state estimation, respectively. The central concept of both systems is the schema, a representation of reusable computational knowledge. This can take various forms, including high-level algorithm templates, code optimizations, datatype refinements, or architectural information. A schema also contains applicability conditions that are used to determine when it can be applied safely. These conditions can refer to the initial specification, to intermediate results, or to elements of the partially-instantiated code. Schema-based synthesis uses AI technology to recursively apply schemas to gradually refine a specification into executable code. This process proceeds in two main phases. A front-end gradually transforms the problem specification into a program represented in an abstract intermediate code. A backend then compiles this further down into a concrete target programming language of choice. A core engine applies schemas on the initial problem specification, then uses the output of those schemas as the input for other schemas, until the full implementation is generated. Since there might be different schemas that implement different solutions to the same problem this process can generate an entire solution tree. AUTOBAYES and AUTOFILTER have reached the level of maturity where they enable users to solve interesting application problems, e.g., the analysis of Hubble Space Telescope images. They are large (in total around 100kLoC Prolog), knowledge intensive systems that employ complex symbolic reasoning to generate a wide range of non-trivial programs for complex application do- mains. Their schemas can have complex interactions, which make it hard to change them in isolation or even understand what an existing schema actually does. Adding more capabilities by increasing the number of schemas will only worsen this situation, ultimately leading to the entropy death of the synthesis system. The root came of this problem is that the domain knowledge is scattered throughout the entire system and only represented implicitly in the schema implementations. In our current work, we are addressing this problem by making explicit the knowledge from Merent parts of the synthesis system. Here; we discuss how Gruber's definition of an ontology as an explicit specification of a conceptualization matches our efforts in identifying and explicating the domain-specific concepts. We outline the dual role ontologies play in schema-based synthesis and argue that they address different audiences and serve different purposes. Their first role is descriptive: they serve as explicit documentation, and help to understand the internal structure of the system. Their second role is prescriptive: they provide the formal basis against which the other parts of the system (e.g., schemas) can be checked. Their final role is referential: ontologies also provide semantically meaningful "hooks" which allow schemas and tools to access the internal state of the program derivation process (e.g., fragments of the generated code) in domain-specific rather than language-specific terms, and thus to modify it in a controlled fashion. For discussion purposes we use AUTOLINEAR, a small synthesis system we are currently experimenting with, which can generate code for solving a system of linear equations, Az = b.

Bures, Tomas↗