Results for “label transfer”

NASA NTRS1990

A predictive theoretical model for electron tunneling pathways in proteins

A practical method is presented for calculating the dependence of electron transfer rates on details of the protein medium intervening between donor and acceptor. The method takes proper account of the relative energetics and mutual interactions of the donor, acceptor, and peptide groups. It also provides a quantitative search scheme for determining the important tunneling pathways (specific sequences of localized bonding and antibonding orbitals of the protein which dominate the donor-acceptor electronic coupling) in native and tailored proteins, a tool for designing new proteins with prescribed electron transfer rates, and a consistent description of observed electron transfer rates in existing redox labeled metalloproteins and small molecule model compounds.

Onuchic, Jose Nelson↗

NASA NTRS2017

Interactive Visualization of High-Dimensional Petascale Ocean Data

We describe an application for interactive visualization of 5 petabytes of time-varying multivariate data from a high-resolution global ocean circulation model. The input data are 10311 hourly (ocean time) time steps of various 2D and 3D fields from a 22-billion point 1/48- degree “lat-lon cap” configuration of the MIT General Circulation Model (MITgcm). We map the global horizontal model domain onto our 128-screen (8x16) tiled display wall to produce a canonical tiling with approximately one MITgcm grid point per display pixel, and using this tiling we encode the entire time series for multiple native and computed scalar quantities at a collection of ocean depths. We reduce disk bandwidth requirements by converting the model’s floating point data to 16-bit fixed point values, and compressing those values with a lossless video encoder, which together allow synchronized playback at 24 time steps per second across all 128 displays. The application allows dynamic assignment of any two encoded tiles to any display, and has multiple interfaces for quickly specifying various orderly arrangements of tiles. All subsequent rendering is done on the fly, with run time control of colormaps, transfer functions, histogram equalization, and labeling. The two data streams on each screen can be rendered independently and combined in various ways, including blending, differencing, horizontal/ vertical wipes, and checkerboarding. The two data streams on any screen can optionally be displayed as a scatterplot in their joint attribute space. All scatterplots and map-view plots from the same x/y location and depth are linked so they all show the current brushable selection. Ocean scientists have used the system, and have found previously unidentified features in the data.

Interactive↗

NASA NTRS2018

GEONEX: Progressive Conditional Generative Adversarial Training Using Transfer learning

Obtaining accurate segmentation on large scale images is an open problem in deep learning. The main problem is the amount of labeled data that exists for large scale images. Traditionally, the common solution to this problem is to crop the large images into smaller images to increase the amount of available data and train a Conditional Generative Adversarial Network (CGAN). CGANs are currently the state of the art in image to image translation and provide better accuracy than the traditional method of training a encoder based conv-net architecture to minimize the loss at each pixel. This method can produce noisy and discontinuous images with inaccurate results. We seek to solve this problem by utilizing the concepts of transfer learning and progressive training to create a CGAN that can segment large scale images with a limited amount of labeled data. In transfer learning we recognize that many learned features are applicable to many classes from multiple domains. This introduces the concept of feature reusability, which is the basis for finetuning. Progressive training got its start in training models on the same images at different resolutions. In this work we instead train a GAN on increasing image scales by transferring the weights from the smaller scales to the larger scales. The learned features at the smaller scales are continually reused and applied to larger scales to create a CGAN that can perform accurate segmentation on large scale images. We apply this method to detect building footprints on very high-resolution overhead imagery (e.g Digital Globe and high resolution airborne platforms).

GEONEX↗

NASA NTRS1989

Facilitating information transfer in the EOS era

A simple interactive demonstration program has been written in C to allow a user to input data field descriptions as label format. This program generates a full RECFMT (record format) description and the complete transfer syntax description notation (TSDN) file. It is intended that this program be upgraded to operational quality and be made available to users to simplify the description and TSDN file construction task. The total set of capabilities, from the standard formatted data unit packaging of related files and consistent segment structures, through the type definition techniques and the call server, will constitute a unique tool for the systematic transfer of data. This software on each end may be independent, one end from the other. With it available, local software that will be needed to convert user files to and from the canonical interface will be appreciably simplified.

Billingsley, Frederic C.↗

NASA NTRSDate not supplied

MULTI-OBJECTIVE REINFORCEMENT LEARNING FOR LOW-THRUST TRANSFER DESIGN BETWEEN LIBRATION POINT ORBITS

Multi-Reward Proximal Policy Optimization (MRPPO) is a multi-objective reinforcement learning algorithm used to construct low-thrust transfers between periodic orbits in multi-body systems. Previous implementations of MRPPO have relied on a predefined reference transfer to successfully train each policy. In this paper, an algorithmic modification labeled the ‘moving reference’, is introduced to autonomously construct these reference trajectories during training. With this modification, MRPPO is used to recover various low-thrust transfers between two periodic orbits in the Earth-Moon circular restricted three-body problem to solve a multi-objective optimization problem. These results are then compared with the solutions recovered via a traditional optimization formulation.

algorithm↗

NASA NTRSDate not supplied

MULTI-OBJECTIVE REINFORCEMENT LEARNING FOR LOW-THRUST TRANSFER DESIGN BETWEEN LIBRATION POINT ORBITS

Multi-Reward Proximal Policy Optimization (MRPPO) is a multi-objective reinforcement learning algorithm used to construct low-thrust transfers between periodic orbits in multi-body systems. Previous implementations of MRPPO have relied on a predefined reference transfer to successfully train each policy. In this paper, an algorithmic modification labeled the ‘moving reference’, is introduced to autonomously construct these reference trajectories during training. With this modification, MRPPO is used to recover various low-thrust transfers between two periodic orbits in the Earth-Moon circular restricted three-body problem to solve a multi-objective optimization problem. These results are then compared with the solutions recovered via a gradient descent optimization scheme to validate the performance of MRPPO with the moving reference modification.

Christopher J. Sullivan↗

NASA NTRSDate not supplied

MULTI-OBJECTIVE REINFORCEMENT LEARNING FOR LOW-THRUST TRANSFER DESIGN BETWEEN LIBRATION POINT ORBITS

Multi-Reward Proximal Policy Optimization (MRPPO) is a multi-objective reinforcement learning algorithm used to construct low-thrust transfers between periodic orbits in multi-body systems. Previous implementations of MRPPO have relied on a predefined reference transfer to successfully train each policy. In this paper, an algorithmic modification labeled the ‘moving reference’, is introduced to autonomously construct these reference trajectories during training. With this modification, MRPPO is used to recover various low-thrust transfers between two periodic orbits in the Earth-Moon circular restricted three-body problem to solve a multi-objective optimization problem. These results are then compared with the solutions recovered via a gradient descent optimization scheme to validate the performance of MRPPO with the moving reference modification

Christopher John Sullivan↗

NASA NTRS2021

Multi-objective Reinforcement Learning for Low-thrust Transfer Design Between Libration Point Orbits

Multi-Reward Proximal Policy Optimization (MRPPO) is a multi-objective rein- forcement learning algorithm used to construct low-thrust transfers between pe- riodic orbits in multi-body systems. Previous implementations of MRPPO have relied on a predefined reference transfer to successfully train each policy. In this paper, an algorithmic modification labeled the ‘moving reference’, is introduced to autonomously construct these reference trajectories during training. With this modification, MRPPO is used to recover various low-thrust transfers between two periodic orbits in the Earth-Moon circular restricted three-body problem to solve a multi-objective optimization problem. These results are then compared with the solutions recovered via a gradient descent optimization scheme to validate the performance of MRPPO with the moving reference modification.

Anderson, Rodney L.↗

NASA NTRS2021

Multi-objective Reinforcement Learning for Low-thrust Transfer Design Between Libration Point Orbits

Multi-Reward Proximal Policy Optimization (MRPPO) is a multi-objective rein- forcement learning algorithm used to construct low-thrust transfers between pe- riodic orbits in multi-body systems. Previous implementations of MRPPO have relied on a predefined reference transfer to successfully train each policy. In this paper, an algorithmic modification labeled the ‘moving reference’, is introduced to autonomously construct these reference trajectories during training. With this modification, MRPPO is used to recover various low-thrust transfers between two periodic orbits in the Earth-Moon circular restricted three-body problem to solve a multi-objective optimization problem. These results are then compared with the solutions recovered via a gradient descent optimization scheme to validate the performance of MRPPO with the moving reference modification.

Anderson, Rodney L↗

NASA NTRS2022

Mars Terrain Segmentation with Less Labels

Planetary rover systems need to perform terrain segmentation to identify drivable areas as well as identify specific types of soil for sample collection. The latest Martian terrain segmentation methods rely on supervised learning which is very data hungry and difficult to train where only a small number of labeled samples are available. Moreover, the semantic classes are defined differently for different applications (e.g., rover traversal vs. geological) and as a result the network has to be trained from scratch each time, which is an inefficient use of resources. This research proposes a semi-supervised learning framework for Mars terrain segmentation where a deep segmentation network trained in an unsupervised manner on unlabeled images is transferred to the task of terrain segmentation trained on few labeled images. The network incorporates a backbone module which is trained using a contrastive loss function and an output atrous convolution module which is trained using a pixel-wise cross-entropy loss function. Evaluation results using the metric of segmentation accuracy show that the proposed method with contrastive pre-training outperforms plain supervised learning by 2%-10%. Moreover, the proposed model is able to achieve a segmentation accuracy of 91.1% using only 161 training images (1% of the original dataset) compared to 81.9% with plain supervised learning.

Wilson, Brian D↗

NASA NTRSDate not supplied

Improving Sim-to-Real Transfer in Vision-Based Robot Navigation Via Instance-Level GAN-Based Data Augmentation

Achieving robust vision-based robotic tasks requires large amounts of data, which are often difficult to obtain in real-world scenarios. Simulators and synthetic data offer a cost-effective alternative, but the visual gap between simulation and reality hinders the performance of models when deployed in real-world environments. In this paper, we present a data augmentation pipeline that integrates a foundation model (Segment Anything Model) with an unsupervised image-to-image translation model (CycleGAN) for instance-level domain transfer from simulation to reality. This pipeline enables the generation of realistic labeled data from synthetic images for training supervised machine learning models in vision-based navigation tasks. We evaluate our approach on real-world data for ego-vehicle pose estimation, a critical autonomous navigation task involving the prediction of cross-track position and heading angle relative to road center line markings. The results of our tests show that our GAN-based data augmentation pipeline significantly outperforms models trained solely on simulation data or on data processed with standard image augmentation methods for sim-to-real transfer, enhancing model robustness and generalizability in real-world scenarios. Our method provides a scalable and flexible data augmentation tool for leveraging large synthetic datasets to enhance vision-based robotic navigation tasks.

artificial intelligence↗

NASA NTRS2001

Expression of the Acyl-Coenzyme A: Cholesterol Acyltransferase GFP Fusion Protein in Sf21 Insect Cells

The enzyme acyl-coenzyme A:cholesterol acyltransferase (ACAT) is an important contributor to the pathological expression of plaque leading to artherosclerosis n a major health problem. Adequate knowledge of the structure of this protein will enable pharmaceutical companies to design drugs specific to the enzyme. ACAT is a membrane protein located in the endoplasmic reticulum.t The protein has never been purified to homogeneity.T.Y. Chang's laboratory at Dartmouth College provided a 4-kb cDNA clone (K1) coding for a structural gene of the protein. We have modified the gene sequence and inserted the cDNA into the BioGreen His Baculovirus transfer vector. This was successfully expressed in Sf2l insect cells as a GFP-labeled ACAT protein. The advantage to this ACAT-GFP fusion protein (abbreviated GCAT) is that one can easily monitor its expression as a function of GFP excitation at 395 nm and emission at 509 nm. Moreover, the fusion protein GCAT can be detected on Western blots with the use of commercially available GFP antibodies. Antibodies against ACAT are not readily available. The presence of the 6xHis tag in the transfer vector facilitates purification of the recombinant protein since 6xHis fusion proteins bind with high affinity to Ni-NTA agarose. Obtaining highly pure protein in large quantities is essential for subsequent crystallization. The purified GCAT fusion protein can readily be cleaved into distinct GFP and ACAT proteins in the presence of thrombin. Thrombin digests the 6xHis tag linking the two protein sequences. Preliminary experiments have indicated that both GCAT and ACAT are expressed as functional proteins. The ultimate aim is to obtain large quantities of the ACAT protein in pure and functional form appropriate for protein crystal growth. Determining protein structure is the key to the design and development of effective drugs. X-ray analysis requires large homogeneous crystals that are difficult to obtain in the gravity environment of earth. Protein crystals grown in microgravity are often larger and have fewer defects than those grown on earth. The analysis of higher quality space-grown crystals will assist in structure-based drug design. We have successfully grown GCAT-infected Sf21 cells in both adhesion and suspension cultures. Expression levels of GCAT in cell lines such as Sf9 and High Five appear to be reduced. We intend to replicate GCAT expression in all three cell lines using the NASA rotating wall bioreactor which effectively duplicates a microgravity environment. The bioreactor itself could be launched to study the expression of the GFP and GCAT proteins in the actual microgravity environment achieved in orbit.

Mahtani, H. K.↗

NASA NTRS2021

Deep Domain Adaptation based Cloud Type Detection using Active and Passive Satellite Data

Domain adaptation techniques have been developed to handle data from multiple sources or domains. Most existing domain adaptation models assume that source and target domains are homogeneous, i.e., they have the same feature space. Nevertheless, many real world applications often deal with data from heterogeneous domains that come from completely different feature spaces. In our remote sensing application, data in source domain (from an active spaceborne Lidar sensor CALIOP onboard CALIPSO satellite) contain 25 attributes, while data in target domain (from a passive spectroradiometer sensor VIIRS onboard Suomi-NPP satellite) contain 20 different attributes. CALIOP has better representation capability and sensitivity to aerosol types and cloud phase, while VIIRS has wide swaths and better spatial coverage but has inherent weakness in differentiating atmospheric objects on different vertical levels. To address this mismatch of features across the domains/sensors, we propose a novel end-to-end deep domain adaptation with domain mapping and correlation alignment (DAMA) to align the heterogeneous source and target domains in active and passive satellite remote sensing data. It can learn domain invariant representation from source and target domains by transferring knowledge across these domains, and achieve additional performance improvement by incorporating weak label information into the model (DAMA-WL). Our experiments on a collocated CALIOP and VIIRS dataset show that DAMA and DAMA-WL can achieve higher classification accuracy in predicting cloud types.

domain adaptation↗

NASA NTRS2007

Increasing Linear Dynamic Range of a CMOS Image Sensor

A generic design and a corresponding operating sequence have been developed for increasing the linear-response dynamic range of a complementary metal oxide/semiconductor (CMOS) image sensor. The design provides for linear calibrated dual-gain pixels that operate at high gain at a low signal level and at low gain at a signal level above a preset threshold. Unlike most prior designs for increasing dynamic range of an image sensor, this design does not entail any increase in noise (including fixed-pattern noise), decrease in responsivity or linearity, or degradation of photometric calibration. The figure is a simplified schematic diagram showing the circuit of one pixel and pertinent parts of its column readout circuitry. The conventional part of the pixel circuit includes a photodiode having a small capacitance, CD. The unconventional part includes an additional larger capacitance, CL, that can be connected to the photodiode via a transfer gate controlled in part by a latch. In the high-gain mode, the signal labeled TSR in the figure is held low through the latch, which also helps to adapt the gain on a pixel-by-pixel basis. Light must be coupled to the pixel through a microlens or by back illumination in order to obtain a high effective fill factor; this is necessary to ensure high quantum efficiency, a loss of which would minimize the efficacy of the dynamic- range-enhancement scheme. Once the level of illumination of the pixel exceeds the threshold, TSR is turned on, causing the transfer gate to conduct, thereby adding CL to the pixel capacitance. The added capacitance reduces the conversion gain, and increases the pixel electron-handling capacity, thereby providing an extension of the dynamic range. By use of an array of comparators also at the bottom of the column, photocharge voltages on sampling capacitors in each column are compared with a reference voltage to determine whether it is necessary to switch from the high-gain to the low-gain mode. Depending upon the built-in offset in each pixel and in each comparator, the point at which the gain change occurs will be different, adding gain-dependent fixed pattern noise in each pixel. The offset, and hence the fixed pattern noise, is eliminated by sampling the pixel readout charge four times by use of four capacitors (instead of two such capacitors as in conventional design) connected to the bottom of the column via electronic switches SHS1, SHR1, SHS2, and SHR2, respectively, corresponding to high and low values of the signals TSR and RST. The samples are combined in an appropriate fashion to cancel offset-induced errors, and provide spurious-free imaging with extended dynamic range.

Pain, Bedabrata↗

NASA NTRS2019

Carbon Nanostructures in Biological Sensors for Space and Terrestrial Applications

Biosensing devices comprised of carbon nanotubes and nanofibers have been developed for astronaut crew point-of-care. Their inherent nanometer scale, high conductivity, wide potential window, good biocompatibility and well-defined surface chemistry make them ideal candidates as biosensor electrodes. Here, we report two studies using carbon nanotube and carbon nanofiber electrodes for biomedical applications. First, a 3x3 electrode device, with each electrode containing 40,000 carbon nanofiber nanoelectrodes was fabricated on silicon using traditional microfabrication processing. The device was demonstrated as a multiplexed immunosensor for simultaneous, label-free detection of cardiac troponin-I, C-reactive protein and myoglobin. Antibodies specific to cardiac troponin-I, C-reactive protein and myoglobin were covalently bound to the CNF surface and were characterized using electrochemical impedance spectroscopy and differential pulse voltammetry. Each step of the modification process resulted in changes in resistance to charge transfer due to the changes at the electrode surface upon antibody immobilization and binding to the specific cardiac protein. The real-time label free detection of the three cardiac markers from pure components and mixtures was demonstrated with high sensitivity, down to 0.2 ng/mL, and good selectivity. Detection in human blood serum did not present false positives from non-specific protein adsorption. More recently, this detection scheme has been applied to inkjet printed carbon nanotube electrodes on Kapton and paper. Printed devices have several unique advantages including simple and inexpensive fabrication. The results demonstrate that these sensors can serve a miniaturized, low cost device for detection of proteins in complex mixtures making this platform a good candidate for early stage diagnosis of myocardial infarction. Future inkjet printed devices can be fabricated have the added advantage in their suitability to be manufactured in an in-space, microgravity environment.

carbon nanotubes↗

NASA NTRS2019

Carbon Nanomaterials for Biosensing Applications

Biosensing devices comprised of carbon nanotubes and nanofibers have been developed for astronaut crew point-of-care. Their inherent nanometer scale, high conductivity, wide potential window, good biocompatibility and well-defined surface chemistry make them ideal candidates as biosensor electrodes. Here, we report two studies using carbon nanotube and carbon nanofiber electrodes for biomedical applications. First, a 3x3 electrode device, with each electrode containing 40,000 carbon nanofiber nanoelectrodes was fabricated on silicon using traditional microfabrication processing. The device was demonstrated as a multiplexed immunosensor for simultaneous, label-free detection of cardiac troponin-I, C-reactive protein and myoglobin. Antibodies specific to cardiac troponin-I, C-reactive protein and myoglobin were covalently bound to the CNF surface and were characterized using electrochemical impedance spectroscopy and differential pulse voltammetry. Each step of the modification process resulted in changes in resistance to charge transfer due to the changes at the electrode surface upon antibody immobilization and binding to the specific cardiac protein. The real-time label free detection of the three cardiac markers from pure components and mixtures was demonstrated with high sensitivity, down to 0.2 ng/mL, and good selectivity. Detection in human blood serum did not present false positives from non-specific protein adsorption. More recently, this detection scheme has been applied to inkjet printed carbon nanotube electrodes on Kapton and paper. Printed devices have several unique advantages including simple and inexpensive fabrication. The results demonstrate that these sensors can serve a miniaturized, low cost device for detection of proteins in complex mixtures making this platform a good candidate for early stage diagnosis of myocardial infarction. Future inkjet printed devices can be fabricated have the added advantage in their suitability to be manufactured in an in-space, microgravity environment.

carbon nanotubes↗

NASA NTRS2022

Microstructure Segmentation With Deep Learning Encoders Pre-Trained on a Large Microscopy Dataset

This study examined the improvement of microscopy segmentation intersection over union accuracy by transfer learning from a large dataset of microscopy images called MicroNet. Many neural network encoder architectures were trained on over 100,000 labeled microscopy images from 54 material classes. These pre-trained encoders were then embedded into multiple segmentation architectures including UNet and DeepLabV3+ to evaluate segmentation performance on created benchmark microscopy datasets. Compared to ImageNet pre-training, models pre-trained on MicroNet generalized better to out-of-distribution micrographs taken under different imaging and sample conditions and were more accurate with less training data. When training with only a single Ni-superalloy image, pre-training on MicroNet produced a 72.2% reduction in relative intersection over union error. These results suggest that transfer learning from large in-domain datasets generate models with learned feature representations that are more useful for downstream tasks and will likely improve any microscopy image analysis technique that can leverage pre-trained encoders.

machine learning↗

NASA NTRS1998

Fluorescence Studies of Lysozyme Nucleation

Fluorescence is one of the most powerful tools available for the study of macromolecules. For example, fluorescence can be used to study self association through methods such as anisotropy (the rotational rate of the molecule in solution), quenching (the accessibility of a bound probe to the bulk solution), and resonance energy transfer (measurement of the distance between two species). Fluorescence can also be used to study the local environment of the probe molecules, and the changes in that environment which accompany crystal nucleation and growth. However fluorescent techniques have been very much underutilized in macromolecular growth studies. One major advantage is that the fluorescent species generally must be at low concentration, typically ca 10-5 to 10-6 M. Thus one can study a very wide range of solution conditions, ranging from very high to very low protein concentration, he latter of which are not readily accessible to scattering techniques. We have prepared a number of fluorescent derivatives of chicken egg white lysozyme (CEWL). Fluorescent probes have been attached to two different sites, ASP 101 and the N-terrninal amine, with a sought for use in different lines of study. Preliminary resonance energy transfer studies have been -carried out using pyrene acetic acid (Ex 340 mn, Em 376 nm) lysozyme as a donor and cascade blue (Ex 377 run, Em 423 nm) labeled lysozyme as an acceptor. The emission of both the pyrene and cascade blue probes was followed as a function of the salt protein concentrations. The data show an increase in cascade blue and a concomitant decrease in the pyrene fluorescence as either the salt or protein concentrations are increased, suggesting that the two species are approaching each other close enough for resonance energy transfer to occur. This data can be analyzed to measure the distance between the probe molecules and, knowing their locations on the protein molecule their distances from and orientations with respect to each other. The results of these and other studies will be discussed.

Pusey, Marc L.↗