Search NASA⌕ Search

SEARCH · Search NASA

Results for “machine learning operations”

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 1,567 records · Page 87

Spaceborne Lidar Retrievals of PM2.5 for Air Quality Studies and Applications

Fine particulate matter (PM2.5) substantially contributes to air pollution and negatively affects human health. While many studies have investigated the use of passive column-integrated aerosol optical depth to infer surface PM2.5, the use of lidar observations for air quality characterization is not nearly as extensive. Lidar measurements are critical, however, due to the vertical aerosol information they provide, including near the surface. In this presentation, we first provide an overview of various lidar-based approaches for estimating PM2.5 concentrations and then discuss how lidar measurements can assist other air quality applications. For example, estimates of PM2.5 have been obtained in a physics-based approach through CALIOP near-surface aerosol extinction retrievals, assumptions on the mass extinction efficiency, and incorporating other parameters (an aerosol hygroscopic growth factor and PM2.5/PM10 ratio). Application of this algorithm over the contiguous United States (CONUS) from 2006 to 2018 yielded larger PM2.5 values over the eastern and western CONUS (~10-15 μg/m³) and lower PM2.5 levels in the central CONUS (~5 μg/m³). These spatial patterns were similar to those from gridded PM2.5 concentrations obtained through in situ measurements at ground stations operated by the US Environmental Protection Agency. In another approach, the Cloud Aerosol Transport System (CATS) lidar was used with the Goddard Earth Observing System (GEOS) model in a 1D ensemble-based variational technique to obtain PM2.5 over the US and Europe, and the spatial patterns of the CATS/GEOS based PM2.5 concentrations generally captured those from surface stations (with corresponding hourly EPA PM2.5 vs CATS PM2.5 statistics of R=0.4 and bias=1.5 μg/m³). In our recent work, as part of the Models, In situ, and Remote sensing of Aerosols (MIRA) Working Group, we have applied both the CALIOP and CATS/GEOS based approaches over the highly polluted country of India during the post-monsoon season (September-October 2016). We derived elevated levels of two-month mean PM2.5 (~100 μg/m³) in northern India, especially near New Delhi. These high PM2.5 concentrations in the Indo-Gangetic plain are driven in large part from the seasonal burning of crop residue and meteorological conditions typical at this time of the year, such as low wind speeds and a shallow boundary layer. While the satellite-derived PM2.5 moderately replicates (R = ~0.7-0.9) the spatial variability in the two-month mean of surface in situ PM2.5 from monitoring sites operated by the Central and State Pollution Control Boards, we show results from specific scenes for which there are large deviations between the satellite-derived PM2.5 and in situ measurements. Other current work on this topic focuses on developing PM2.5 estimates using airborne high spectral resolution lidar measurements through machine learning regression algorithms and involves several parameters (e.g., aerosol extinction, color ratio, lidar ratio). Application of this method over major metropolitan areas in the US and Asia have resulted in high correlations (R = 0.93) with surface measurements. This airborne lidar approach can be adapted to spaceborne lidar measurements, and all three of these approaches can be applied to ESA’s EarthCARE Atmospheric Lidar instrument, setting the stage for the future Cloud Aerosol Lidar for Global Scale Observations of the Ocean-Land Atmosphere System (CALIGOLA) mission. Ultimately, beyond estimates of PM2.5, the aerosol vertical distribution from lidars can benefit studies involving passive sensor approaches for PM2.5 proxies (including from geostationary satellites), wildfire smoke plume injection heights, volcanic emissions (e.g., ash height retrievals), and aerosol/air quality model assimilation, evaluation, and forecasts.

Travis D Toth↗

Pro2R: Procurement of Ramping Product and Regulation in CAISO Using Probabilistic Solar Power Forecasts

This presentation summarizes the objectives and findings from the project titled, "Pro2R: Coordinated Procurement of Ramping Product & Regulation in CAISO Using Probabilistic Solar Power Forecasts," funded by DOE SETO. The goal is to develop cutting edge probabilistic solar power forecasting methods, and integrate them into ISO market operations. The team has integrated in two ways: 1) to procure ramping product and regulation requirements (ancillary services) in the California ISO market, and 2) to develop an open source visitation and ramp alert related situational awareness using probabilistic forecasts. This deck also summarizes the results from the use of machine learning methods to relate probabilistic solar forecasts to the the needs for flexible ramp product in the California ISO system.

41 EE - Solar Energy Technologies Office (EE-4S)↗

A two-stage optical fusion framework for wildfire severity mapping across the conterminous United States

Accurate wildfire severity mapping (WSM) is essential for post-fire recovery planning, erosion risk assessment, ecosystem monitoring, and disaster risk reduction. Although Landsat and Sentinel optical imagery have been widely used for burn severity assessment, the added value of fusing multiple optical sensors has not been sufficiently quantified across diverse fire events, particularly since the launch of Landsat-9. This study evaluates whether multisensor optical fusion improves wildfire severity mapping relative to single-sensor baselines using Sentinel-2, Landsat-8, and Landsat-9 imagery across 40 wildfire events in the conterminous United States. We tested a two-stage fusion framework that combines feature-level fusion with pixel-level dimensionality reduction. First, feature-level fused datasets were created through early fusion by combining standardized post-fire bands from each sensor into a single predictor stack. Both raw reflectance bands and pairwise spectral transforms were retained to capture within- and cross-sensor spectral interactions. Second, Linear Discriminant Analysis was applied to both single-sensor and fused datasets to produce comparable low-dimensional feature spaces. Six machine-learning classifiers were then used to benchmark model performance with repeated spatially buffered train–test splits. Results show that Landsat-9 was the strongest single-sensor baseline. Among the fusion strategies, Sentinel-2 + Landsat-9 produced the most consistent improvement and reduced performance variability. Landscape-condition analysis further showed that this fusion was most beneficial in shrubland-dominated and high-terrain fires, where it achieved the highest overall mean accuracy and the fewest failures. In contrast, its benefits were less reliable in evergreen forests, mixed vegetation, and low- to moderate-elevation terrain. In operational settings, the Sentinel-2 + Landsat-9 configuration offers a practical solution for post-fire recovery planning, erosion-risk assessment, watershed management, and ecological monitoring when field observations are available and timely satellite-based information is needed.

Landsat↗

Connectivity‐based parcellation of normal and anatomically distorted human cerebral cortex

Abstract For over a century, neuroscientists have been working toward parcellating the human cortex into distinct neurobiological regions. Modern technologies offer many parcellation methods for healthy cortices acquired through magnetic resonance imaging. However, these methods are suboptimal for personalized neurosurgical application given that pathology and resection distort the cerebrum. We sought to overcome this problem by developing a novel connectivity‐based parcellation approach that can be applied at the single‐subject level. Utilizing normative diffusion data, we first developed a machine‐learning (ML) classifier to learn the typical structural connectivity patterns of healthy subjects. Specifically, the Glasser HCP atlas was utilized as a prior to calculate the streamline connectivity between each voxel and each parcel of the atlas. Using the resultant feature vector, we determined the parcel identity of each voxel in neurosurgical patients ( n = 40) and thereby iteratively adjusted the prior. This approach enabled us to create patient‐specific maps independent of brain shape and pathological distortion. The supervised ML classifier re‐parcellated an average of 2.65% of cortical voxels across a healthy dataset ( n = 178) and an average of 5.5% in neurosurgical patients. Our patient dataset consisted of subjects with supratentorial infiltrating gliomas operated on by the senior author who then assessed the validity and practical utility of the re‐parcellated diffusion data. We demonstrate a rapid and effective ML parcellation approach to parcellation of the human cortex during anatomical distortion. Our approach overcomes limitations of indiscriminately applying atlas‐based registration from healthy subjects by employing a voxel‐wise connectivity approach based on individual data.

59 BASIC BIOLOGICAL SCIENCES↗

Residential Demand Flexibility: Modeling Occupant Behavior using Sociodemographic Predictors

Demand flexibility (DF) has the potential to increase the saturation of renewables in the grid and reduce operating costs for both utilities and customers. However, less than 8% of U.S. residential electric customers are enrolled in DF programs. A major research gap on this topic is an uneven understanding of behavioral drivers of electricity use and DF program participation at the household level. In this study, we employ machine learning models to predict residential occupant behavior in activities relevant to DF. We model occupants' extensive decisions (i.e., choice of action) and intensive behaviors (i.e., amount of time spent) during peak and off-peak time periods using the publicly available American Time Use Survey, which includes activities data for approximately 200,000 respondents. In our machine learning models, predictions for both extensive and intensive behavior fell within a +/-20% error margin at the aggregate level. We identify 13 key sociodemographic predictors of DF-related intensive behavior using LASSO inference and beta coefficient ranking. However, these top predictors differ by activity, suggesting potential scope for differential user targeting for DF events and technologies during program design. This work also contributes to understanding when and who might adopt these DF technologies based on their daily routine activities.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Crowdsourcing the Frontier: Advancing Hybrid Physics‐ML Climate Simulation via a $\$$50,000 Kaggle Competition

Subgrid machine-learning (machine learning [ML]) parameterizations have the potential to introduce a new generation of climate models that incorporate the effects of higher-resolution physics without incurring the prohibitive computational cost associated with more explicit physics-based simulations. However, important issues, ranging from online instability to inconsistent online performance, have limited their operational use for long-term climate projections. To more rapidly drive progress in solving these issues, domain scientists and ML researchers opened up the offline aspect of this problem to the broader ML and data science community with the release of ClimSim, a NeurIPS Data sets and Benchmarks publication, and an associated Kaggle competition. This paper reports on the downstream results of the Kaggle competition by coupling emulators inspired by the winning teams' architectures to an interactive climate model (including full cloud microphysics, a regime historically prone to online instability) and systematically evaluating their online performance. Our results demonstrate that online stability in the low-resolution real-geography setting is reproducible across multiple diverse architectures, which we consider a key milestone. All tested architectures exhibit strikingly similar offline and online biases, though their responses to architecture-agnostic design choices (e.g., expanding the list of input variables) can differ significantly. Multiple Kaggle-inspired architectures achieve state-of-the-art results on certain metrics such as zonal mean bias patterns and global Root Mean Squared Error, indicating that crowdsourcing the essence of the offline problem is one path to improving online performance in hybrid physics-AI climate simulation.

Environmental sciences↗

Ferroelectric capacitors and field-effect transistors as in-memory computing elements for machine learning workloads

Abstract This study discusses the feasibility of Ferroelectric Capacitors (FeCaps) and Ferroelectric Field-Effect Transistors (FeFETs) as In-Memory Computing (IMC) elements to accelerate machine learning (ML) workloads. We conducted an exploration of device fabrication and proposed system-algorithm co-design to boost performance. A novel FeCap device, incorporating an interfacial layer (IL) and$$\text {Hf}_{0.5}\text {Zr}_{0.5}\text {O}_2$$ Hf 0.5 Zr 0.5 O 2 (HZO), ensures a reduction in operating voltage and enhances HZO scaling while being compatible with CMOS circuits. The IL also enriches ferroelectricity and retention properties. When integrated into crossbar arrays, FeCaps and FeFETs demonstrate their effectiveness as IMC components, eliminating sneak paths and enabling selector-less operation, leading to notable improvements in energy efficiency and area utilization. However, it is worth noting that limited capacitance ratios in FeCaps introduced errors in multiply-and-accumulate (MAC) computations. The proposed co-design approach helps in mitigating these errors and achieves high accuracy in classifying the CIFAR-10 dataset, elevating it from a baseline of 10% to 81.7%. FeFETs in crossbars, with a higher on-off ratio, outperform FeCaps, and our proposed charge-based sensing scheme achieved at least an order of magnitude reduction in power consumption, compared to prevalent current-based methods.

Science & Technology - Other Topics↗

Protecting Customer Privacy Through Distributed Energy Resource Anonymization

Due to their stochastic nature, the increase of Renewable Energy Resources (RERs) as a primary source of energy for power grids creates challenges regarding the reliability and resilience of the system. In order to combat these obstacles, expansion of Distributed Energy Resources (DERs) and their participation in Demand Response (DR) programs is necessary. Widespread participation requires prioritizing customer privacy and addressing concerns that may arise regarding communication between DERs and the Grid Service Provider (GSP). This paper discusses the use of flow reservation resources to split the operating cycles of DER load profiles into unique phases. The splitting of phases increases anonymization of the DERs by making it more difficult to determine the individual characteristics of the device. We discuss an example of this using simulated DER load profile data and examine the resulting effectiveness by using a machine learning algorithm for classification, called Support Vector Machine (SVM).

Distributed Energy Resource, Anonymization, Renewa↗

ML-Based Reconstruction in a Pixelated LArTPC

The Deep Underground Neutrino Experiment (DUNE) will address open issues in neutrino physics such as the measurement of the CP-violating phase in neutrino oscillations and the neutrino mass ordering. DUNE is expected to have the most energetic neutrino beam in the world. The 2x2 demonstrator is a single-phase liquid argon time projection chamber (LArTPC), with four modules, operated as a prototype for the DUNE Liquid Argon Near Detector (ND-LAr). Each module in the 2x2 demonstrator is 0.67m x 0.67m x 1.8m. Based on the ArgonCube design concept, the 2x2 features a novel pixelated charge readout and advanced high-coverage photon detection system. Machine learning can be used to form a complete reconstruction pipeline for physics interactions in the 2x2 detector. This poster will describe the workings of a package under development called SPINE and its current performance.

43 PARTICLE ACCELERATORS↗

SWIPE: Spectral Water Inversion Processor and Emulator

Degradation of Earth’s inland water resources due to anthropogenic perturbations and climate anomalies at both local and global scales continues to place human health at substantial risk. There is now a growing necessity to develop pragmatic approaches that allow timely and effective extrapolation of local processes, to spatially resolved global products, and to promote operational and sustainable resource policy management. This presentation will be discussing the progress made developing SWIPE: Spectral Water Inversion Processor and Emulator. SWIPE is a platform for advanced modeling of coastal and inland aquatic habitats. The goal is create a comprehensive and cohesive system to leverage recent advancements in computation and machine learning to develop a synthetic training ground for sensitivity studies and algorithm development. The four principal facets of SWIPE include: 1. Advanced two-layer coated sphere bio-optical modeling and GPU radiative transfer modeling, 2. Big Data involving massive synthetic spectral libraries of optical properties of various global aquatic particles, surface reflectance, and top-of-atmosphere reflectance, all at hyperspectral resolution leveraging high-end computing systems at NASA Ames Research Center, 3. Deep Learning for algorithm development for water quality inversion of concentrations of common biogeophysical variables as well as optics, full uncertainty characterization by water type, and forward emulation, and lastly, 4. Image Processing for application of developed retrieval algorithms for both hyperspectral and multispectral sensors with experimental corrections for global adjacency, noise, sunglint, and benthic reflectance. This presentation will demonstrate the Equivalent Algal Populations (EAP) two-layer coated sphere scattering model which has been used develop spectral libraries of hyperspectral inherent optical properties of roughly 80 species of phytoplankton, covering 15 different classes and nine taxonomic functional types. The EAP model was also used to derive spectral properties of 10 different non-algal particle functional types. Examples of how the SMART-G (Speed-up Monte-carlo Advanced Radiative Transfer using GPU) radiative transfer code is used to model optically complex aquatic signals will be presented and discussed in the context of creating a massive synthetic database which can leverage the full power of next generation machine learning techniques and high end computing for water quality inversion. We will discuss our active investigation in things like appropriate model architectures, dimensionality reduction techniques such as PCA and autoencoders, uncertainty quantification and abstaining, and which variables actually benefit most from hyperspectral information versus multispectral resolution. We are also curious about questions relating to cost/benefit analysis in terms of computation resources, neural network complexity, and data volumes. Answers to these questions will hopefully elaborate on cost efficiency for potential future sensor design considerations.

SWIPE↗

Coaxial Rotor CFD Validation and ML Surrogate Model Generation

A fixed-pitch speed-controlled coaxial rotor system was tested in the NASA Langley Transonic Dynamics Tunnel in September 2022. The rotors have a diameter of 1.35 meters and an inter-rotor spacing of 25% of the diameter. Though this test was focused on the NASA New Frontiers Dragonfly mission, the resulting dataset is relevant to a wide array of applications of multirotor vehicles, especially those using fixed-pitch variable-speed rotors. Most notably beyond Dragonfly, perhaps, is the application of coaxial rotor pair configurations for eVTOL and Urban Air Mobility (UAM) aircraft. This work provides a thorough CFD validation study quantifying coaxial rotor performance estimation with accuracy on order 5-10% using an efficient hybrid BEMT-URANS flow solver over a wide range of operating conditions. This accuracy was achieved using novel approaches for the construction of both the C81 airfoil performance lookup tables and the BEMT rotor model. These novel approaches are combined with advanced scripting to further accelerate the commercial off-the-shelf CFD solver on GPU accelerated machines. Finally, the development of machine learning surrogate models is presented to produce highly efficient and accurate rotor performance predictions for fixed-pitch variable-speed multirotor aircraft over the complete flight regime including scout, cruise, climb, descent, as well as limiting cases of vortex-ring and windmill-brake states.

Coaxial↗

Utilizing Genetic Programming to Identify Failures in ICS Networks

Previously, researchers have attempted to apply machine learning techniques to network anomaly detection problems. Due to the staggering amount of variety that can occur in normal networks, the results have often been underwhelming. These challenges are far less pronounced when considering industrial control system (ICS) networks. The recurrent nature of these networks results in less noise and more consistent patterns for a machine learning algorithm to recognize. Here, we propose a method of evolving decision trees through genetic programming (GP) in order to detect network anomalies, such as device outages. Our approach extracts over a dozen features from network packet captures and netflows, normalizes them, and relates them in decision trees using fuzzy logic operators. Furthermore, the trees were used to detect three specific network events from three different points on the network across a statistically significant number of runs and achieved 100% accuracy on five of the nine experiments. When the trees attempted to detect more challenging events at points of presence further from the occurrence, the accuracy averaged to above 98%. Finally, using our method, all of the evolutionary cycles of the GP algorithm are computed a-priori, allowing the best resultant trees to be deployed as semi-real-time sensors with little overhead.

97 MATHEMATICS AND COMPUTING↗

Prediction of Weather Impacted Airport Capacity using Ensemble Learning

Ensemble learning with the Bagging Decision Tree (BDT) model was used to assess the impact of weather on airport capacities at selected high-demand airports in the United States. The ensemble bagging decision tree models were developed and validated using the Federal Aviation Administration (FAA) Aviation System Performance Metrics (ASPM) data and weather forecast at these airports. The study examines the performance of BDT, along with traditional single Support Vector Machines (SVM), for airport runway configuration selection and airport arrival rates (AAR) prediction during weather impacts. Testing of these models was accomplished using observed weather, weather forecast, and airport operation information at the chosen airports. The experimental results show that ensemble methods are more accurate than a single SVM classifier. The airport capacity ensemble method presented here can be used as a decision support model that supports air traffic flow management to meet the weather impacted airport capacity in order to reduce costs and increase safety.

Weather impact↗

Machine-learning-assisted deciphering of microstructural effects on ionic transport in composite materials: A case study of Li 7 La 3 Zr 2 O 12 -LiCoO 2

The effective diffusivity of ionic species in multiphase materials is critical for the design and function of composite materials for electrochemical energy storage. In practice, effective diffusivity depends sensitively not only on the intrinsic diffusivities of constituting materials but also on their topological arrangement; nevertheless, these coupled contributions are oversimplified in most analytical models. Here, we combine atomistically informed mesoscale modeling and machine learning (ML) analysis to unravel how such features affect effective diffusivity in two-phase composites. Using the Li 7 La 3 Zr 2 O 12 -LiCoO 2 composite solid-state battery cathode as a model system, we compute effective diffusivity for 600 distinct dense polycrystalline microstructures with different topological configurations of grains, grain boundaries, and heterointerfaces. We verify that in addition to atomic-scale variabilities, microstructural feature diversity can significantly impact effective transport properties. Across the ensemble of test microstructures, this often results in bimodal distributions of effective diffusivity that encompass two qualitatively distinct operating mechanisms, which we identify via flux analysis. An ML approach reveals that the most critical determining factors for effective diffusivity are the connectivity of bulk phases and their heterointerfaces. The role of ionic mobility at the heterointerfaces is also discussed. These insights highlight the combined importance of microstructure and interface engineering in tuning the transport properties of ionic species in composite materials. In conclusion, our framework can also be extended for understanding generic microstructure-property relationships in other complex multiphase materials.

25 ENERGY STORAGE↗

Dynamic Role-Based Access Control Policy for Smart Grid Applications: An Offline Deep Reinforcement Learning Approach

Role-based access control (RBAC) is adopted in the information and communication technology domain for authentication purposes. However, due to a very large number of entities within organizational access control (AC) systems, static RBAC management can be inefficient, costly, and can lead to cybersecurity threats. In this paper, a novel hybrid RBAC model is proposed, based on the principles of offline deep reinforcement learning (RL) and Bayesian belief networks. The considered framework utilizes a fully offline RL agent, which models the behavioral history of users as a Bayesian belief-based trust indicator. Thus, the initial static RBAC policy is improved in a dynamic manner through off-policy learning while guaranteeing compliance of the internal users with the security rules of the system. By deploying our implementation within the smart grid domain and specifically within a Distributed Energy Resources (DER) ecosystem, we provide an end-to-end proof of concept of our model. Finally, detailed analysis and evaluation regarding the offline training phase of the RL agent are provided, while the online deployment of the hybrid RL-based RBAC model into the DER ecosystem highlights its key operation features and salient benefits over traditional RBAC models.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

An Unobtrusive System to Measure, Assess, and Predict Cognitive Workload in Real-World Environments

Across many careers, individuals face alternating periods of high and low attention and cognitive workload, which can result in impaired cognitive functioning and can be detrimental to job performance. For example, some professions (e.g., fire fighters, emergency medical personnel, doctors and nurses working in an emergency room, pilots) require long periods of low workload (boredom), followed by sudden, high-tempo operations during which they may be required to respond to an emergency and perform at peak cognitive levels. Conversely, other professions (e.g., air traffic controllers, market investors in financial industries, analysts) require long periods of high workload and multitasking during which the addition of just one more task results in cognitive overload resulting in mistakes. An unobtrusive system to measure, assess, and predict cognitive workload could warn individuals, their teammates, or their supervisors when steps should be taken to augment cognitive readiness. In this talk I will describe an approach to this problem that we have found to be successful across work domains including: (1) a suite of unobtrusive, field-ready neurophysiological, physiological, and behavioral sensors that are chosen to best suit the target environment; (2) custom algorithms and statistical techniques to process and time-align raw data originating from the sensor suite; (3) probabilistic and statistical models designed to interpret the data into the human state of interest (e.g., cognitive workload, attention, fatigue); (4) and machine-learning techniques to predict upcoming performance based on the current pattern of events, and (5) display of each piece of information depending on the needs of the target user who may or may not want to drill down into the functioning of the system to determine how conclusions about human state and performance are determined. I will then focus in on our experimental results from our custom functional near-infrared spectroscopy sensor, designed to operate in real-world environments to be worn comfortably (e.g., positioned into a baseball cap or a surgeons cap) to measure changes in brain blood oxygenation without adding burden to the individual being assessed.

workload collection↗

Human Factors Directions for Civil Aviation

Despite considerable progress in understanding human capabilities and limitations, incorporating human factors into aircraft design, operation, and certification, and the emergence of new technologies designed to reduce workload and enhance human performance in the system, most aviation accidents still involve human errors. Such errors occur as a direct or indirect result of untimely, inappropriate, or erroneous actions (or inactions) by apparently well-trained and experienced pilots, controllers, and maintainers. The field of human factors has solved many of the more tractable problems related to simple ergonomics, cockpit layout, symbology, and so on. We have learned much about the relationships between people and machines, but know less about how to form successful partnerships between humans and the information technologies that are beginning to play a central role in aviation. Significant changes envisioned in the structure of the airspace, pilots and controllers' roles and responsibilities, and air/ground technologies will require a similarly significant investment in human factors during the next few decades to ensure the effective integration of pilots, controllers, dispatchers, and maintainers into the new system. Many of the topics that will be addressed are not new because progress in crucial areas, such as eliminating human error, has been slow. A multidisciplinary approach that capitalizes upon human studies and new classes of information, computational models, intelligent analytical tools, and close collaborations with organizations that build, operate, and regulate aviation technology will ensure that the field of human factors meets the challenge.

Hart, Sandra G.↗

Even Higher-Level Synthesis: An Exploration of AI Hardware Accelerators using HLS4ML

With the rise of artificial intelligence, the popularization of deep learning, and a constantly evolving industry, the demand for flexible and efficient tools has never been greater. As algorithms grow more complex, their runtime and energy consumption increase exponentially. Customized hardware accelerators, long used for specific mathematical operations, remain essential for managing modern applications' computational and power demands. Hardware accelerators can speed up complex computations by orders of magnitude, but their manual design and verification processes are often challenging and time-consuming. High-Level Synthesis (HLS) provides a solution by transforming high-level algorithm descriptions, typically written in C++ or SystemC, into synthesizable RTL suitable for hardware implementation. This approach reduces development time for RTL engineers while offering flexibility beyond what traditional handwritten RTL can provide. We extended this capability to the machine-learning domain with the open-source framework hls4ml, which allows neural networks trained in Python frameworks like Tensorflow or PyTorch to be synthesized into efficient hardware representations for the traditional FPGA and ASIC flows. This breakthrough addresses the growing need for reduced design turnaround and easy verification of ML hardware accelerators with low latency and power efficiency constraints. During this tutorial, we will demonstrate how Python complements HLS by simplifying the ML design process, bridging the gap between software and hardware development. Attendees will explore how we translate neural networks modeled in Python into fixed-point C++ models suitable for HLS workflows. We will dive into strategies like Value-Range Analysis and Quantization-Aware Training, which optimize these designs for deployment and evaluate their accuracy, power consumption, and energy efficiency. To exemplify these concepts, experts from Fermilab will share their experiences applying this technology to high-energy physics experiments, where real-time, low-latency processing is critical. Over the years, Fermilab engineers have demonstrated how deep neural networks, optimized for hardware using hls4ml, can meet the stringent requirements of trigger systems at the CERN Large Hadron Collider. These systems rely on rapid decision-making to process immense data volumes while retaining only the most relevant events for further analysis. The application of hls4ml has also been extended to innovative technologies like smart pixel arrays. These smart pixels integrate ML inference capabilities directly into sensor devices, enabling localized data processing at the pixel level. This approach drastically reduces the need to transmit raw data to external processing units, significantly decreasing power consumption and latency. By embedding neural networks within the pixel architecture, the smart pixels can identify and prioritize relevant data in real time, providing a highly efficient solution for edge computing in scenarios such as particle detectors and imaging systems. Fermilab's work highlights the potential of hardware-accelerated ML in scenarios where both speed and power efficiency are mission-critical. Through this tutorial, attendees will gain valuable insights into the challenges and solutions of deploying ML in hardware. Understanding how HLS and hls4ml streamline the development of neural network-based hardware accelerators is fundamental for the industry's future. Participants will learn how these technologies are shaping the future of AI and scientific computing.

Di Guglielmo, Giuseppe [Fermilab]↗