Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “TRAINING PLATFORMS”

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 379 records · Page 21

O3BNN-R: An Out-Of-Order Architecture for High-Performance and Regularized BNN Inference

Binarized Neural Networks (BNN) have drawn tremendous attention due to significantly reduced computational complexity and memory demand. They have especially shown great potential in cost- and power-restricted domains, such as IoT and smart edge-devices, where reaching a certain accuracy bar is often sufficient, and real-time is highly desired.In this work, we demonstrate that the highly-condensed BNN model can be shrunk significantly further by dynamically pruning irregular redundant edges. Based on two new observations on BNN-specific properties, an out-of-order (OoO) architecture – O3BNN-R, can curtail edge evaluation in cases where the binary output of a neuron can be determined early. Similar to Instruction-Level-Parallelism(ILP), these fine-grained, irregular, runtime pruning opportunities are traditionally presumed to be difficult to exploit. In order to increase the pruning opportunities, we also optimize the training process by adding 2 regularization items in the loss function (1) for pooling pruning and (2) for threshold pruning. We evaluate our design on an FPGA platform using three well-known networks, including VggNet-16, AlexNet for ImageNet, and a VGG-like network for Cifar-10.

Geng, Tong↗

Spy in the GPU-box: Covert and Side Channel Attacks on Multi-GPU System

The deep learning revolution has been enabled in large part by GPUs, and more recently accelerators, which make it possible to carry out computationally demanding training and inference in acceptable times. As the size of machine learning networks and workloads continues to increase, multi-GPU machines have emerged as an important platform offered on High Performance Computing and cloud data centers. Since these machines are shared among multiple users, it becomes increasingly important to protect applications against potential attacks. In this paper, we explore the vulnerability of Nvidia's DGX multi-GPU machines to covert and side channel attacks. These machines consist of a number of discrete GPUs that are interconnected through a combination of custom interconnect (NVLink) and PCIe connections. We reverse engineer the interconnected cache hierarchy and show that it is possible for an attacker on one GPU to cause contention on the L2 cache of another GPU. We use this observation to first develop a covert channel attack across two GPUs, achieving the best bandwidth of around 4 MB/s. We also develop a prime and probe attack on a remote GPU allowing an attacker to recover the cache access pattern of another workload. This access pattern can be used in any number of side channel attacks: we demonstrate a proof of concept attack that fingerprints the application running on the remote GPU, with high accuracy. We also develop a proof of concept attack to extract hyperparameters of a machine learning workload. Our work establishes for the first time the vulnerability of these machines to microarchitectural attacks and can guide future research to improve their security.

Dutta, Sankha↗

National Wind Distance Learning Collaborative

Executive Summary The energy development assumptions identified in the Department of Energy's position paper, 20% Wind Energy by 2030, projected an exploding demand for wind energy-related workforce development. These primary assumptions drove a secondary set of assumptions that early stage wind industry workforce development and training paradigms would need to undergo significant change if the workforce needs were to be met. The current training practice and culture within the wind industry is driven by a relatively small number of experts with deep field experience and knowledge. The current training methodology is dominated by face-to-face, classroom based, instructor present training. Given these assumptions and learning paradigms, the purpose of the National Wind Distance Learning Collaborative was to determine the feasibility of developing online learning strategies and products focused on training wind technicians. The initial project scope centered on (1) identifying resources that would be needed for development of subject matter and course design/delivery strategies for industry-based (non-academic) training, and (2) development of an appropriate Learning Management System (LMS). As the project unfolded, the initial scope was expanded to include development of learning products and the addition of an academic-based training partner. The core partners included two training entities, industry-based Airstreams Renewables and academic-based Lake Area Technical Institute. A third partner, Vision Video Interactive, Inc. provided technology-based learning platforms (hardware and software).

17 WIND ENERGY↗

Cascading a systolic array and a feedforward neural network for navigation and obstacle avoidance using potential fields

A technique is developed for vehicle navigation and control in the presence of obstacles. A potential function was devised that peaks at the surface of obstacles and has its minimum at the proper vehicle destination. This function is computed using a systolic array and is guaranteed not to have local minima. A feedfoward neural network is then used to control the steering of the vehicle using local potential field information. In this case, the vehicle is a trailer truck backing up. Previous work has demonstrated the capability of a neural network to control steering of such a trailer truck backing to a loading platform, but without obstacles. Now, the neural network was able to learn to navigate a trailer truck around obstacles while backing toward its destination. The network is trained in an obstacle free space to follow the negative gradient of the field, after which the network is able to control and navigate the truck to its target destination in a space of obstacles which may be stationary or movable.

Plumer, Edward S.↗

Second-Generation Six-Limbed Experimental Robot

The figure shows the LEMUR II - the second generation of the Limbed Excursion Mechanical Utility Robot (LEMUR), which was described in "Six-Legged Experimental Robot" (NPO-20897), NASA Tech Briefs, Vol. 25, No. 12 (December 2001), page 58. The LEMUR II incorporates a number of improvements, including new features, that extend its capabilities beyond those of its predecessor, which is now denoted the LEMUR I. To recapitulate: the LEMUR I was a six-limbed robot for demonstrating robotic capabilities for assembly, maintenance, and inspection. The LEMUR I was designed to be capable of walking autonomously along a truss structure toward a mechanical assembly at a prescribed location and to perform other operations. The LEMUR I was equipped with stereoscopic video cameras and image-data-processing circuitry for navigation and mechanical operations. It was also equipped with a wireless modem, through which it could be commanded remotely. Upon arrival at a mechanical assembly, the LEMUR I would perform simple mechanical operations with one or both of its front limbs. It could also transmit images to a host computer. Each of the six limbs of the LEMUR I was operated independently. Each of the four rear limbs had three degrees of freedom (DOFs), while each of the front two limbs had four DOFs. The front two limbs were designed to hold, operate, and/or be integrated with tools. The LEMUR I included an onboard computer equipped with an assortment of digital control circuits, digital input/output circuits, analog-to-digital converters for input, and digital-to-analog (D/A) converters for output. Feedback from optical encoders in the limb actuators was utilized for closed-loop microcomputer control of the positions and velocities of the actuators. The LEMUR II incorporates the following improvements over the LEMUR I: a) The drive trains for the joints of the LEMUR II are more sophisticated, providing greater torque and accuracy. b) The six limbs are arranged symmetrically about a hexagonal body platform instead of in straight lines along the sides. This symmetrical arrangement is more conducive to omnidirectional movement in a plane. c) The number of degrees of freedom of each of the rear four limbs has been increased by one. Now, every limb has four degrees of freedom: three at the hip (or shoulder, depending on one s perspective) and one at the knee (or elbow, depending on one s perspective). d) Now every limb (instead of only the two front limbs) can perform operations. For this purpose, each limb is tipped with an improved quick-release mechanism for swapping of end-effector tools. e) New end-effector tools have been developed. These include an instrumented rotary driver that accepts all tool bits that have 0.125-in. (3.175-mm)-diameter shanks, a charge-coupled-device video camera, a super bright light-emitting diode for illuminating the work area of the robot, and a generic collet tool that can be quickly and inexpensively modified to accept any cylindrical object up to 0.5 in. (12.7 mm) in diameter. f) The stereoscopic cameras are mounted on a carriage that moves along a circular track, thereby providing for omnidirectional machine vision. g) The control software has been augmented with software that implements innovations reported in two prior NASA Tech Briefs articles: the HIPS algorithm ["Hybrid Image-Plane/Stereo Manipulation" (NPO-30492), Vol. 28, No. 7 (July 2004), page 55] and the CAMPOUT architecture ["An Architecture for Controlling Multiple Robots" (NPO-30345), Vol. 28, No. 10 (October 2004), page 65].

Kennedy, Brett↗

FInd Gas Source

FIGS is a neural network software that ingests real time synchronized field data on environmental flow fields and turbulence and gas concentration variations at high frequency and uses an error minimization algorithm to locate the gas source and quantify its strength. The software can be interfaced with atmospheric, oceanic and subsurface instruments in a variety of platforms stationary or mobile (e.g. cars, UAVs, submersible vehicles or boreholes) and used to find gas sources by smart use of data and phenomenology. FIGS can be trained by phenomenological model of the flow fields in the environment of interest and/or be calibrated by controlled release. After initial deployment the FIGS learning will grow with time as it accumulates data on source quantification. FIGS can be installed on any computer from small beagle-bones for field deployment/end-use to PC/MACs/main-frame for training/analysis. FIGS has been trained (using LANL's high resolution atmospheric simulations) and calibrated, tested and evaluated in the field and shown to perform well in finding and quantifying methane leaks at 10-100m scales at well pads by ingesting atmospheric measurements. The code is applicable to gas and particle source location at large scales.

Travis, Bryan↗

Enabling Efficient and Confident Annotation of LC-MS Metabolomics Data through MS1 Spectrum and Time Prediction

Liquid chromatography coupled to electrospray ionization-mass spectrometry (LC–ESI-MS) is a versatile and robust platform for metabolomic analysis. However, while ESI is a soft ionization technique, in-source phenomena including multimerization, nonproton cation adduction, and in-source fragmentation complicate interpretation of MS data. Here, we report chromatographic and mass spectrometric behavior of 904 authentic standards collected under conditions identical to a typical nontargeted profiling experiment. The data illustrate that the often high level of complexity in MS spectra is likely to result in misinterpretation during the annotation phase of the experiment and a large overestimation of the number of compounds detected. However, our analysis of this MS spectral library data indicates that in-source phenomena are not random but depend at least in part on chemical structure. These nonrandom patterns enabled predictions to be made as to which in-source signals are likely to be observed for a given compound. Using the authentic standard spectra as a training set, we modeled the in-source phenomena for all compounds in the Human Metabolome Database to generate a theoretical in-source spectrum and retention time library. A novel spectral similarity matching platform was developed to facilitate efficient spectral searching for nontargeted profiling applications. Taken together, this collection of experimental spectral data, predictive modeling, and informatic tools enables more efficient, reliable, and transparent metabolite annotation.

60 APPLIED LIFE SCIENCES↗

Impact of Domain Knowledge on the Property Prediction of Specialized Machine Learning Models

Developing transferable machine learning models is trending in data-driven materials research. However, how to apply such models to a specific research domain remains unclear. Here, in this work, we choose high-entropy materials as a platform with a specialized data set containing 145,323 DFT-relaxed materials. This data set is used to explore the role of domain-specific knowledge in training effective models. Our tests with three representative graph neural network architectures indicate the model complexity has much smaller influence on performance than the data itself. Specifically, the consideration of low-energy atomic ordering, structures with diverse elemental coverage, and high-order interactions significantly influences the model performance. We also find that domain knowledge-driven sampling can greatly enhance unsupervised learning techniques. This research highlights that developing specialized data sets is more beneficial than further complicating deep learning architectures. Additionally, physics-inspired sampling algorithms are crucially needed for better machine learning models for a specific materials research domain.

36 MATERIALS SCIENCE↗

Design of Superconducting Optoelectronic Networks for Neuromorphic Computing

We have previously proposed a novel hardware platform (SOEN) for neuromorphic computing based on superconducting optoelectronics that presents many of the features necessary for information processing in the brain. Here we discuss the design and training of networks of neurons and synapses based on this technology. We present circuit models for the simplest neurons and synapses that we can use to build networks. We discuss the further abstracted integrate and fire model that we use for evolutionary optimization of small networks of these neurons. We show that we can use the TENNLab evolutionary optimization programming framework to design small networks for logic, control and classification tasks. We plan to use the results as feedback to inform our neuron design.

Buckley, Sonia↗

A Supervised Machine Learning Approach to Control Energy Storage Devices

This paper introduces a supervised machine learning approach to predict and schedule the real-time operation mode of the next operation interval for residential PV/Battery systems controlled by mode-based controllers. The performance of the mode-based economic model-predictive control (EMPC) approach is used as the benchmark. The residential load and PV data used in the paper are 1-minute data downloaded from the the Pecan Street Project website. The optimal operation mode for each control interval is first derived from the historical data used as the training set. Then, four machine learning algorithms (i.e. neural network, support vector machine, logistic regression, and random forest algorithms) are applied. We compared the performance of the four algorithms when using different number of features and length of the training sets extracted from different months of the year. Simulation results show that using the machine learning approach can effectively improve the performance of the mode-based control system and reduce the computation effort of local controllers because the training can be completed on a cloud-based Machine Learning engine. Furthermore, the work presented in this paper paves the way for using a shared-learning platform to design controllers of residential PV/storage systems. This may significantly reduce the cost for implementing such systems.

25 ENERGY STORAGE↗

Robonaut 2 - Building a Robot on the International Space Station

In 2010, the Robonaut Project embarked on a multi‐phase mission to perform technology demonstrations on‐board the International Space Station (ISS), showcasing state of the art robotics technologies through the use of Robonaut 2 (R2). This phased approach implements a strategy that allows for the use of ISS as a test bed during early development to both demonstrate capability and test technology while still making advancements in the earth based laboratories for future testing and operations in space. While R2 was performing experimental trials onboard the ISS during the first phase, engineers were actively designing for Phase 2, Intra‐Vehicular Activity (IVA) Mobility, that utilizes a set of zero‐g climbing legs outfitted with grippers to grasp handrails and seat tracks. In addition to affixing the new climbing legs to the existing R2 torso, it became clear that upgrades to the torso to both physically accommodate the climbing legs and to expand processing power and capabilities of the robot were required. In addition to these upgrades, a new safety architecture was also implemented in order to account for the expanded capabilities of the robot. The IVA climbing legs not only needed to attach structurally to the R2 torso on ISS, but also required power and data connections that did not exist in the upper body. The climbing legs were outfitted with a blind mate adapter and coarse alignment guides for easy installation, but the upper body required extensive rewiring to accommodate the power and data connections. This was achieved by mounting a custom adapter plate to the torso and routing the additional wiring through the waist joint to connect to the new set of processors. In addition to the power and data channels, the integrated unit also required updated electronics boards, additional sensors and updated processors to accommodate a new operating system, software platform, and custom control system. In order to perform the unprecedented task of building a robot in space, extensive practice sessions and meticulous procedures were required. Since crew training time is at a premium, the R2 team took a skills‐based training approach to ensure the astronauts were proficient with a basic skill set while refining the detailed procedures over several practice sessions and simulations. In addition to the crew activities, meticulous ground procedures were required in order to upgrade firmware on the upper body motor drivers. The new firmware for the IVA mobility unit needed to be deployed using the old software system. This also provided an opportunity to upgrade the upper body joints with new software and allowed for limited insight into the success of the updates. Complete verification that the updated firmware was successfully loaded was not confirmed until the rewiring of the upper body torso was complete.

Diftler, Myron↗

Learning Scene Categories from High Resolution Satellite Image for Aerial Video Analysis

Automatic scene categorization can benefit various aerial video processing applications. This paper addresses the problem of predicting the scene category from aerial video frames using a prior model learned from satellite imagery. We show that local and global features in the form of line statistics and 2-D power spectrum parameters respectively can characterize the aerial scene well. The line feature statistics and spatial frequency parameters are useful cues to distinguish between different urban scene categories. We learn the scene prediction model from highresolution satellite imagery to test the model on the Columbus Surrogate Unmanned Aerial Vehicle (CSUAV) dataset ollected by high-altitude wide area UAV sensor platform. e compare the proposed features with the popular Scale nvariant Feature Transform (SIFT) features. Our experimental results show that proposed approach outperforms te SIFT model when the training and testing are conducted n disparate data sources.

99 GENERAL AND MISCELLANEOUS↗

Peregrine

The purpose of “Peregrine” is to perform in-situ quality control of the additive manufacturing process in powder bed systems. The software is an AI-based method to automatically detect and map multiple classes of defects present in the top layer of a powder bed system, either after material deposition, i.e. raking, or after printing the layer. Relying on image data or sensing modalities that can be converted into image data, the code allows for: (1) various steps of pre-processing of the data (registration, distortion correction, etc), (2) annotation of training datasets, (3) processing of data inline or offline, (4) creating 3D rendering of the defect map, and (5) implementation of corrective actions if allowed by the manufacturing platform.

Paquit, Vincent C↗

Low Cost Rapid Response Spacecraft, (LCRRS): A Research Project in Low Cost Spacecraft Design and Fabrication in a Rapid Prototyping Environment

The Low Cost Rapid Response Spacecraft (LCRRS) is an ongoing research development project at NASA Ames Research Center (ARC), Moffett Field, California. The prototype spacecraft, called Cost Optimized Test for Spacecraft Avionics and Technologies (COTSAT) is the first of what could potentially be a series of rapidly produced low-cost satellites. COTSAT has a target launch date of March 2009 on a SpaceX Falcon 9 launch vehicle. The LCRRS research system design incorporates use of COTS (Commercial Off The Shelf), MOTS (Modified Off The Shelf), and GOTS (Government Off The Shelf) hardware for a remote sensing satellite. The design concept was baselined to support a 0.5 meter Ritchey-Chretien telescope payload. This telescope and camera system is expected to achieve 1.5 meter/pixel resolution. The COTSAT team is investigating the possibility of building a fully functional spacecraft for $500,000 parts and $2,000,000 labor. Cost is dramatically reduced by using a sealed container, housing the bus and payload subsystems. Some electrical and RF designs were improved/upgraded from GeneSat-1 heritage systems. The project began in January 2007 and has yielded two functional test platforms. It is expected that a flight-qualified unit will be finished in December 2008. Flight quality controls are in place on the parts and materials used in this development with the aim of using them to finish a proto-flight satellite. For LEO missions the team is targeting a mission class requiring a minimum of six months lifetime or more. The system architecture incorporates several design features required by high reliability missions. This allows for a true skunk works environment to rapidly progress toward a flight design. Engineering and fabrication is primarily done in-house at NASA Ames with flight certifications on materials. The team currently employs seven Full Time Equivalent employees. The success of COTSATs small team in this effort can be attributed to highly cross trained engineering team. The engineers on the team are capable of functioning in two to three engineering disciplines which allows highly efficient interdisciplinary engineering collaboration. NASA Ames is actively proposing mission concepts to use the COTSAT platform to accomplish science. If the COTSAT team validates this approach, it will allow the possibility for remote sensing missions to produce a high science yield for minimal cost and reduced schedule. Another aim of this approach is to yield an accelerated pathway from a Phase A study to mission launch. Leaders in the aerospace industry have shown interest in this methodology. Several visits and tours have been given in the lab. Although the concept of low-cost development is initially met with skepticism from some within the prohibitive aerospace industry, the project's efforts have been highly praised for the accomplishments met within a limited time and budget. Overall the development has progressed tremendously well and the team is answering critical questions for current and future low-cost small satellite developments. COTSAT subsystems are not limited to a specific weight class and could be adapted to produce smaller platforms and to fit various launch vehicles.

Spremo, Stevan↗

BRAINSTACK – A Platform for Artificial Intelligence & Machine Learning Collaborative Experiments on a Nano-Satellite

As the space economy continues to expand through increasingly easy access to advanced and inexpensive technology, space missions themselves have become more ambitious with exploration targets growing ever distant while simultaneously requiring larger guidance and communication budgets. These conflicting desires of distance and control drive the need for advanced on-board intelligent decision making to reduce communication and control limitations by automating as many mission functions as possible in-situ. While the amount of research on such Artificial Intelligence and Machine Learning (AI/ML) software modules has grown exponentially, the capacity to experimentally validate such software modules in space in a rapid and inexpensive format has not. To this end, the Nano Orbital Workshop (NOW) group at NASA Ames Research Center has been at the forefront of performing initial flight evaluation tests of ‘commercially’ available bleeding-edge computational platforms via what is programmatically referred to as the BrainStack on the TechEdSat (TES-n) flight series. This on-orbit computational platform provides an evaluation laboratory where advanced software experiments are pre-loaded into memory prior to launch, then executed as payloads during mission operations with results reported back and program tweaks or new training sets uploaded as needed. Processors selected as part of the BrainStack are of ideal size, packaging, and power consumption for easy integration into a cube satellite structure. These experiments have included the evaluation of small, high-performance GPUs and, more recently, neuromorphic processors, in LEO operations. Neuromorphic processors are of particular interest due to their superior power efficiency over GPUs in intelligent automation applications. The first TES-n flight test of an Intel first-generation Loihi neuromorphic processor launched on TES-13, January 13, 2022, and continues to operate in orbit despite no significant modifications to harden the processor against the space environment. The Intel Loihi Gen-1 on TES-13 is characterized by a 14nm 128-core Spiking Neural Network (SNN) able to support on-chip training. The processor is packaged in the Kapoho Bay USB module, providing a relatively straight-forward interface to the bus avionics system. The Kapoho Bay was in turn managed by an Intel Pentium single-board computer to handle scheduling of the software application payloads and communications with the satellite’s primary computer. The recently released Intel Loihi Gen-2, able to support integer-valued spike payloads and produced using 7nm process, will form part of the continually evolving BrainStack in the upcoming three TES-n/NOW flights. The Kapoho Point unit will incorporate eight Loihi-2 processors, enabling neural networks of up to one million neurons and one billion synapsis. Additionally, it is planned to measure the radiation environment these processors experience to understand any degradation or computational artifacts caused by long term space radiation exposure on these novel architectures. This evolving flexible and collaborative environment involving various research teams across NASA and other organizations is intended to be a convenient orbital test platform from which many anticipated future space automation applications may be initially tested.

Artificial Intelligence↗

Connecting Space to Village: Tethys Adoption Within SERVIR

The SERVIR program, a joint development initiative of National Aeronautics and Space Administration (NASA) and United States Agency for International Development (USAID), works in partnership with leading regional organizations world-wide to help developing countries use information provided by Earth observing satellites and geospatial technologies for managing climate risks and land use. SERVIR empowers decision-makers with tools, products, and services to act locally on critical issues related to four key areas: Food Security and Agriculture, Water and Water-related Disasters, Land Use/Land Cover and Ecosystems, Weather and Climate. SERVIR is improving awareness, increasing access to information, and supporting analysis to help people in West Africa, Eastern and Southern Africa, Hindu Kush-Himalaya (HKH), the Lower Mekong, South America and Mesoamerica. NASA funds an Applied Sciences Team (AST), composed of experts from universities and research institutions around the United States. Through AST, SERVIR adopted the Tethys platform introduced by Brigham Young University (BYU) for the implementation of tools to deliver scientific data to stakeholders in the Hindu Kush-Himalaya region. Several Tethys applications developed within SERVIR have raised interest amongst decision makers in the countries served by our regional hubs. They see these new apps as key vehicles to disseminate complex information related to weather models, hydrological models, and air quality monitoring, and as such the scope of our Tethys efforts continue to expand to other regions. As of July 2019, each one of the SERVIR hubs (with the exception of Amazonia, which has only recently started), have stood up Tethys portals and have at least one individual trained in the implementation and maintenance of a Tethys portal and in the creation of basic applications. HKH, with BYU's valuable support, has implemented very sophisticated applications that showcase the possibilities and the benefits that the Tethys platform brings.

Tethys↗

Predictive Model for Starlink Maritime Performance Using Multi-Horizon RandomForest

Low Earth orbit (LEO) satellite systems have become a crucial enabler of broadband access for maritime industries, where traditional networks are unavailable. However, the high mobility of LEO constellations and constantly changing weather conditions result in unpredictable link fluctuations, limiting the ability of maritime platforms to plan bandwidth usage proactively. To the best of our knowledge, no prior work has developed a short-term predictive model for maritime LEO connectivity using real experimental field measurements. This paper proposes a data-driven forecasting model that predicts future downlink throughput using multi-horizon RandomForest regression. The model is trained using real experimental coastal measurement data incorporating recent throughput history, network-layer indicators, and environmental variables. The proposed approach reduces mean absolute error by approximately 31% compared to a persistence baseline for 15-minute horizons. It maintains a measurable improvement at 30 minutes, despite increased stochasticity. These findings confirm that proactive bandwidth awareness is feasible on maritime platforms and can effectively support operational decisions such as adaptive streaming, routing, and resource scheduling. The performance gap between forecasting horizons also highlights the need for expanded offshore datasets to improve prediction robustness under harsher maritime environments.

97 MATHEMATICS AND COMPUTING↗

32 examples of LLM applications in materials science and chemistry: towards automation, assistants, agents, and accelerated scientific discovery

Abstract Large language models (LLMs) are reshaping many aspects of materials science and chemistry research, enabling advances in molecular property prediction, materials design, scientific automation, knowledge extraction, and more. Recent developments demonstrate that the latest class of models are able to integrate structured and unstructured data, assist in hypothesis generation, and streamline research workflows. To explore the frontier of LLM capabilities across the research lifecycle, we review applications of LLMs through 32 total projects developed during the second annual LLM hackathon for applications in materials science and chemistry, a global hybrid event. These projects spanned seven key research areas: (1) molecular and material property prediction, (2) molecular and material design, (3) automation and novel interfaces, (4) scientific communication and education, (5) research data management and automation, (6) hypothesis generation and evaluation, and (7) knowledge extraction and reasoning from the scientific literature. Collectively, these applications illustrate how LLMs serve as versatile predictive models, platforms for rapid prototyping of domain-specific tools, and much more. In particular, improvements in both open source and proprietary LLM performance through the addition of reasoning, additional training data, and new techniques have expanded effectiveness, particularly in low-data environments and interdisciplinary research. As LLMs continue to improve, their integration into scientific workflows presents both new opportunities and new challenges, requiring ongoing exploration, continued refinement, and further research to address reliability, interpretability, and reproducibility.

Computer Science↗