This conference provides a technical forum for members of both industry and academia to present their latest applications of machine learning. Machine learning has been applied to a broad domain of image/vision systems from medical imaging to consumer cameras. Learned tasks such as image recognition, noise reduction, or natural language processing, are currently being applied in many common devices such as mobile phones. Training datasets and training methods for machine learning are critical for the success of a system. Studies demonstrating the deployment and benchmarking of machine learning algorithms on specialized computer hardware is highly valuable to many groups in this field. Sensor hardware design or selection as it pertains to machine learning tasks; for example, an analysis of different camera designs and how each pertains to the performance of an image recognition task such as object detection, is of interest. Analysis of full systems that include the sensor technology, data processing hardware, and results are welcome as each area is critical for the successful application of machine learning.

Papers or tutorials reviewing the topics covered by this section are welcome. All abstracts will be reviewed by the program committee for originality and merit. Topics of interest include, but are not limited to, the following:

Algorithms Consumer Application Industrial Application Security Medicine Big Data Hardware Big Experimental Facilities ;
In progress – view active session
Conference 11843

Applications of Machine Learning 2021

In person: 4 August 2021 | Conv. Ctr. Room 7B
On demand now
View Session ∨
  • Poster Session
  • Signal, Image, and Data Processing Plenary Session
  • Signal, Image, and Data Processing Plenary Networking Event
  • 1: Machine Vision and Manufacturing
  • Remote Sensing
  • Algorithms and Imaging
  • 2: Big Data, Simulations, and Physics
  • 3: Consumer Applications and E-Commerce
  • Wednesday Surf Rock Chill and Beer Reception
Information
Co-located In-Person Program with:
Optics and Photonics for Information Processing XV
In person: 4 August 2021 • 9:30 AM - 10:10 AM PDT
Poster Session
In person: 3 August 2021 • 5:30 PM - 7:00 PM PDT | Conv. Ctr. Sails Pavilion, City Trellis Entrance
11843-34
Author(s): Salman Khan, Rochester Institute of Technology (United States); Anna Lang, Zylient, Inc. (United States); Carl Salvaggio, Rochester Institute of Technology (United States)
On demand | Presented Live 3 August 2021
Show Abstract + Hide Abstract
Every year, 60,000 lives are lost worldwide from disasters. Building collapse during earthquakes account for the majority of these deaths. Unreinforced masonry (URM) buildings are particularly vulnerable during seismic events due to the brittle nature of the construction material. Many communities have undertaken costly and timely mitigation programs to locate and retrofit or replace them before disaster strikes. An automated approach for identifying seismically vulnerable buildings using street level imagery has been met with limited success to this point with no promising results presented in literature. We achieved the best overall accuracy reported to date, at 83.6%, in identifying unfinished URM, finished URM, and non-URM buildings. Moreover, an accuracy of 98.8% was achieved for identifying both suspected URMs (finished or unfinished URM). We perform extensive empirical analysis to establish synergistic parameters on our deep neural network, namely ResNeXt-101- FixRes.
11843-35
Author(s): Wenli Huang, Kyle King, U.S. Military Academy (United States)
On demand | Presented Live 3 August 2021
Show Abstract + Hide Abstract
This paper presents the algorithms to detect trace chemicals using a multi-wavelength camera. Multispectral images of the chemical and the background were collected using the Ocean Thin Films SpectroCam. The camera had an integrated motor with 8 filter color wheels and 8 interchangeable custom band pass filters in the spectral range of 200–900 nm. Since chemicals have their unique spectral reflectance, the stack of 8-dimensional image data was obtained and subsequently analyzed to develop an algorithm that can uniquely identify the area where a chemical is present. In this study, we primarily used RDX, 1,3,5-Trinitroperhydro-1,3,5-triazine, the explosive component in C4. The aim of this study was to investigate the potential of the multispectral imaging system and the accuracy of the model in determining C4 chemical.
11843-36
Author(s): Neha Konakalla, Sai Charan Parasharam, Sridhar Varadala, Vidya Jyothi Institute of Technology (India)
On demand
11843-37
Author(s): Xin Zhang, Yi-Yung Chen, Jong-Woei Whang, Chih-Hsien Tsai, Wei-Chieh Tseng, National Taiwan Univ. of Science and Technology (Taiwan)
On demand
Show Abstract + Hide Abstract
Pupillary light reflex (PLR) is essential in the diagnosis of eye diseases and nervous system research. PLR mainly measures the size (diameter or area) of the pupil, in clinical practice, the size of the pupil is predicted by the experience of medical staff. The behavior will not be unified, because it depends on the subjective consciousness or the degree of fatigue of the medical staff. Therefore, this paper believes that there must be a way to quantify pupil size, which is convenient and low cost. This paper proposes an algorithm based on Convolution Neural Network (CNN) deep learning that allows real-time calculations in a low-cost mobile embedded system.
11843-38
Author(s): Jose Luis Haddad, Pontificia Univ. Católica de Chile (Chile), Jet Propulsion Lab. (United States); Eduardo Bendek, Jet Propulsion Lab. (United States); Catalina Flores, Univ. Andrés Bello (Chile), Millennium Institute of Astrophysics (Chile)
On demand
11843-39
Author(s): Venkata Sai Krithik Pothula, Adarsh Raghavan Madabhushi Tirumala , Sai Pratyusha Kundeti, Vasanth K., Mahesh Rajendran, Anand Pandarinath Madanwad, Vidya Jyothi Institute of Technology (India)
On demand
Show Abstract + Hide Abstract
Research from WHO shows in every 1.44 minutes a teenager is killed in road traffic crashes around the world. Our paper suggests using Deep Learning Algorithms to prevent such casualties through hardware and software implementation of the device in motor vehicles and overcome the potential limitation of Face Recognition to go that extra mile. We suggest using the integration of 4 models namely, Face Detection, Passive Liveness Detection(PLD), Face Recognition, Eye Detection. PLD model is used for Presentation Attack Detection. The face recognition model is trained using the shape predictor 68 landmarks which are unique for each face. Using these landmarks, monitoring of sleepiness is carried out. Along with that different current attack scenarios/ limitations of Facial Recognition that will be faced with these devices are described. Based on these scenarios, some of the preventive methods are elaborated to make the purpose of the device to its fullest performance.
11843-40
Author(s): Zahra Sobhaninia, Isfahan Univ. of Technology (Iran, Islamic Republic of); Hajar Danesh, Islamic Azad Univ. (Iran, Islamic Republic of); Rahele Kafieh, Isfahan Univ. of Medical Sciences (Iran, Islamic Republic of); Jayaram Jothi Balaji, Sankara Nethralaya (India), Medical Research Foundation (India); Vasudevan Lakshminarayanan, Univ. of Waterloo (Canada)
On demand
Show Abstract + Hide Abstract
We have developed a novel location-aware deep learning method for segmentation of the foveal avascular zone (FAZ) in en face optical coherence tomography angiography (OCTA) images. Using this we compare the FAZ related parameters determined using 4 methods, including ground truth (clinician marked image), and two other automated methods.. Our preliminary results suggest that the proposed method is a good substitute for device in-built software in and can replace the manual segmentation.
11843-41
Author(s): Aditya Chandra Mandal, Indian Institute of Technology (BHU), Varanasi (India); Abhijeet Phatak, Consultant (United States); Jayaram Jothi balaji, Sankara Nethralaya (India), Medical Research Foundation (India); Vasudevan Lakshminarayanan, Univ. of Waterloo (Canada)
On demand
Show Abstract + Hide Abstract
. Here we present a novel approach to pupillometry that uses deep-learning (DL) methodologies applied to visible light images. We report the accuracy of our models and compare it with real-world experimental data. This work is the first step toward a visible light smartphone-based pupillometer that is fast, accurate, and relies on on-device computing
11843-42
Author(s): Enzo Casamassima, Andrew Herbert, Cory Merkel, Rochester Institute of Technology (United States)
On demand
Show Abstract + Hide Abstract
Recent studies in the field of adversarial robustness have shown that Convolutional Neural Networks (CNNs) are not as resilient as we might have thought. While traditional attacks focus on small imperceptible image perturbations, they are also vulnerable to simple transformations, such as rotations and translations. In this paper, we explore visually recognizable transformations to humans, measuring the impact on the network classification accuracy. Furthermore, we visualize the learned feature representations by CNNs and analyze how robust these learned representations are and how they compare to the human visual system, hoping to serve as a basis for advancing this interdisciplinary area.
11843-45
Author(s): Abu Taufique, Andreas Savakis, Rochester Institute of Technology (United States); Michael Braun, Daniel Kubacki, Ethan Dell, Systems & Technology Research (United States); Lei Qian, Sean O'Rourke, Air Force Research Lab. (United States)
On demand
Show Abstract + Hide Abstract
Siamese deep-network trackers have received significant attention in recent years due to their real-time speed and state-of-the-art performance. However, Siamese trackers suffer from similar looking confusers, that are prevalent in aerial imagery and create challenging conditions due to prolonged occlusions where the tracker object re-appears under different pose and illumination. Our work proposes SiamReID, a novel re-identification framework for Siamese trackers, that incorporates confuser rejection during prolonged occlusions and is well-suited for aerial tracking. The re-identification feature is trained using both triplet loss and a class balanced loss. Our approach achieves state-of-the-art performance in the UAVDT single object tracking benchmark.
11843-47
Author(s): Zhaowei Chen, Mikey R. Holtz, Samuel J. Vidourek, Hossein Alisafaee, Rose-Hulman Institute of Technology (United States)
On demand | Presented Live 3 August 2021
Show Abstract + Hide Abstract
We have designed and built a machine vision system for the inspection and monitoring at the production line of fiber interlock armor cables. Our economical solution for the design approach has been to utilize a vision system based on a single camera and a tunnel source of uniform illumination in conjunction with two flat mirrors to obtain a 360-degree view of the product. The resolution of imaging system allows the detection of features on the order of tens of microns. The measurement and imperfection method utilizes a deep learning algorithm to detect manufacturing defects in the cable in-line with the production. Our vision system is able to inspect a variety of interlock armor cables with different sizes and shapes, making it uniquely versatile.
11843-48
Author(s): Viviane Oliveira Das Merces, Univ. Federal da Bahia (Brazil); Anderson Dourado Sisnando, Univ. Federal do Recôncavo of Bahia (Brazil); Vitaly Felix Rodriguez-Esquerre, Escola Politécnica da Univ. Federal da Bahia (Brazil)
On demand
Show Abstract + Hide Abstract
The present work deals with the implementation of machine learning algorithms for the analysis of the coupling efficiency of tapers for silicon photonics applications operating in the C band. The analyzed tapers are used for coupling a continuous waveguide with a periodical subwavelength waveguide and they are composed by several segments with variable lengths. The training, testing and validating data sets have been numerically obtained by an efficient frequency domain finite element method which solves the wave equation and determines the spatial distribution of the electromagnetic fields and the coupling efficiency for each taper configuration. An excellent agreement has been observed for the coupling efficiency calculation using the machine learning algorithms when compared with the one obtained by using the finite element method.
Signal, Image, and Data Processing Plenary Session
In person / Livestream: 4 August 2021 • 11:00 AM - 11:45 AM PDT | Conv. Ctr. Room 6A
Session Chair: Andrew Brown, SPIE (United States)
11841-501
Author(s): Nibir K. Dhar, U.S. Army CCDC C5ISR Center Night Vision and Electronics Sensors Directorate (United States)
On demand | Presented Live 4 August 2021
Show Abstract + Hide Abstract
The warfighter’s effectiveness in current and future combat missions can be severely limited by a lack of adequate situational awareness (SA). Better target discrimination, better view of the operational scene with larger fields of regard and longer standoff distances are some of the important criteria. SA also strongly depends on the information, signal and data processing that can provide visual and analytics at the edge. This presentation will highlight current and future challenges as well as discuss a path forward to leverage the AI/ML and other imaging technologies. In addition, highlights of the new innovation platform will be presented.
Signal, Image, and Data Processing Plenary Networking Event
In person: 4 August 2021 • 11:45 AM - 12:15 PM PDT | Conv. Ctr. Room 6A
Join your colleagues for 30 minutes of networking and discussion after the Signal, Image, and Data Processing plenary talk.
Session 1: Machine Vision and Manufacturing
In person / Livestream: 4 August 2021 • 2:00 PM - 2:20 PM PDT | Conv. Ctr. Room 7B
Session Chair: Bob Hainsey, SPIE (United States)
11843-1
Author(s): Nathan Mundhenk, Ian Palmer, Brian J. Gallagher, T. Yong Han, Lawrence Livermore National Lab. (United States)
On demand
Show Abstract + Hide Abstract
We recently developed a deep learning method that can determine the critical peak stress of a material by looking at scanning electron microscope (SEM) images of the material’s crystals. However, it has been somewhat unclear what kind of image features the network is keying off of when it makes its prediction. It is common in computer vision to employ an explainable AI saliency map to tell one what parts of an image are important to the network’s decision. One can usually deduce the important features by looking at these salient locations. However, SEM images of crystals are more abstract to the human observer than natural image photographs. As a result, it is not easy to tell what features are important at the locations which are most salient. To solve this, we developed a method that helps us map features from important locations in SEM images to non-abstract textures that are easier to interpret.
11843-3
Author(s): Eric Bianchi, Virginia Polytechnic Institute and State Univ. (United States); Matthew Hebdon, The University of Texas at Austin (United States)
On demand | Presented Live 4 August 2021
Show Abstract + Hide Abstract
Synthetic image generation using deep learning techniques like generative adversarial networks (GANS), has drastically improved since its inception. There has been significant research using human face datasets like FFHQ, and city-semantic datasets for self-driving car applications. Utilizing latent space distributions, researchers have been able to generate or edit images to exhibit specific traits in the resultant images, like face-aging. However, there has been little GAN research and datasets in the structural infrastructure domain. We propose a GAN-inversion application to embed real structural bridge detail images and incrementally age them using learned semantic boundaries. The trained model offers the ability to forecast deterioration incrementally, which is valuable to inspectors, engineers, and owners because it gives a window into the future on how and where damage may progress.
11843-2
Author(s): Kyle S. Hickmann, Deborah Shutt, Los Alamos National Lab. (United States); Andrew Robinson, Jonathan Lind, Lawrence Livermore National Laboratory (United States)
On demand
Show Abstract + Hide Abstract
In this work we demonstrate a method for leveraging high-fidelity, multi-physics simulations of high-speed impacts in a particular manufactured material to encode prior information regarding the impactor material's strength properties. Our simulations involve a material composed of stacked cylindrical ligaments impacted by a high-velocity aluminum plate. We show that deep neural networks of relatively simple architecture can be trained on the simulations to make highly-accurate inferences of the strength properties of the impactor material. We detail our neural architectures and the considerations that went into their design. In addition, we discuss the simplicity of our network architecture which lends itself to interpretability of learned features in radiographic observations.
11843-4
Author(s): Aneek E. James, Alexander Wang, Songli Wang, Keren Bergman, Columbia Univ. (United States)
On demand
Remote Sensing
11843-6
Author(s): Alice M. Durieux, Descartes Labs, Inc. (United States); Rose Rustowicz, Nikhil Sharma, Descartes Labs (United States); Jason Schatz, Matthew T. Calef, Descartes Labs, Inc. (United States); Christopher X. Ren, Los Alamos National Laboratory (United States)
On demand
Show Abstract + Hide Abstract
In this work, we leverage deep learning to reproduce and expand Synthetic Aperture Radar (SAR) based deforestation detections generated using a probabilistic Bayesian model. Our Bayesian detections leverage SAR to provide near-real time alerting for deforestation in cloudy areas. These detections are timely but have imperfect recall. Here, we utilize deforestation detections generated using the Bayesian model as labels and optical Sentinel-2 composites as input features to train a deep learning model to detect deforested patches at various stages of regrowth in a test area of interest (AOI) in Sumatra (Indonesia). Results suggest that a deep learning model trained on Bayesian deforestation detections can identify deforestation at various stages of regrowth despite highly imperfect labels. This provides a promising avenue to develop deforestation models that can provide more accurate forest loss acreage estimates than the Bayesian model.
11843-8
Author(s): Ryan Connal, Wade Pines, Meg Borek, Timothy Bauch, Nina Raqueno, Rochester Institute of Technology (United States); Brian d'Entremont, Alfred Garrett, Savannah River National Lab. (United States); Carl Salvaggio, Rochester Institute of Technology (United States)
On demand
Show Abstract + Hide Abstract
The process of automatically segmenting objects from complex backgrounds is extremely useful when trying to utilize those objects for computer vision research, such as object detection, autonomous driving, pedestrian tracking, etc. Instance segmentation is imperative towards ongoing research between the Digital Imaging and Remote Sensing Laboratory at the Rochester Institute of Technology and the Savannah River National Laboratory. This research involves careful experimental design towards the capture and development of a custom dataset to utilize instance segmentation on this custom dataset with the Matterport Mask R-CNN implementation [1], where condensed water vapor plumes were identified and masked from various complex backgrounds for the purpose of 3D reconstruction. [1] W. Abdulla. ”Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow.” Matterport, 2017. https://github.com/matterport/Mask RCNN
11843-12
Author(s): Branndon Jones, Ali Ahmadibeni, Tennessee State Univ (United States); Amir Shirkhodaie, Tennessee State Univ. (United States)
On demand
Show Abstract + Hide Abstract
In this study, we used IRIS electromagnetic modeling and simulation (IRIS-EM) virtual environment system to systematically generate a large-scale synthetic SAR imagery dataset of the test marine scenarios. A typical marine scenario includes CAD models of test marine vehicle(s) and their associated wake(s) as well as the ocean layer. Prior to generating the synthetic images of each test scenario, we augment to each CAD model in the virtual environment appropriate physics properties. In this paper, we present our systematic approach for generating synthetic SAR Imagery of marine test scenarios and detail our methodology for annotating them properly. To evaluate and verify the effectiveness of this approach, we bench-marked our generated simulated marine SAR imagery with similar context images taken by the physical SAR imaging systems.
Algorithms and Imaging
11843-14
Author(s): Diane Oyen, Michal Kucer, Brendt Wohlberg, Los Alamos National Lab. (United States)
On demand
Show Abstract + Hide Abstract
We present VisHash for retrieving copies of images, particularly drawings and diagrams used in technical documents. Recent advances in computer vision using deep learning methods have significantly advanced our ability to analyze and retrieve natural images, yet most of these advances do not directly apply to drawings and diagrams due to the very different nature of low-level features of the images. We take advantage of the effectiveness of the relative-brightness signature and extend the approach to develop VisHash, a visual similarity preserving signature that works well on technical diagrams. Importantly, we demonstrate the high level of precision of VisHash for image retrieval compared with competing image hashes of large sets of real drawings from patents and technical images from the web. VisHash is available as open-source code to incorporate into image search and indexing workflows.
11843-15
Author(s): Rajiv Mandya Nagaraju, Cory Merkel, Rochester Institute of Technology (United States)
On demand
Show Abstract + Hide Abstract
Producing a better segmentation is crucial in scene understanding. Semantic Segmentation is a vital task for applications such as autonomous driving, robotics, medical image understanding. Efficient high and low-level context manipulation is a key for competent pixel-level classification. In our implementation, We use a two bridged network. The first bridge manipulates the subtle differences between images and produces the mutual gate vector to understand the low-level features in the images better. The second bridge uses the dilated convolutional network to avoid attrition of the size of the image while gathering a better understanding of the image's high-level features. The initial experiments have yielded an initial mean IoU of 70.1% and pixel accuracy of 94.4% on the cityscapes dataset and 34.6% on the ADE20K dataset.
11843-16
Author(s): Sria Biswas, Balasubramanyam Appina, Indian Institute of Information Technology, Design and Manufacturing, Kancheepuram (India); Roopak R. Tamboli, Indian Institute of Technology Hyderabad (India); Peter Andras Kara, Budapest Univ. of Technology and Economics (Hungary); Aniko Simon, Sigma Technology (Hungary)
On demand
Show Abstract + Hide Abstract
In this paper, we introduce an exhaustive analysis regarding the practical applications of objective quality metrics for stereoscopic 3D imaging. Our contribution addresses each and every state-of-the-art objective metric in the scientific literature, separately for image and video quality. The study differentiates the metrics by input requirements and supervision, and examines performance via statistical measures. The paper focuses on the actual practical applications of the predictive models, and highlights relevant criteria, along with general feasibility, suitability and usability. The analysis of the investigated use cases also addresses potential future research questions and specifies the appropriate directives for quality-focused, user-centric development.
11843-18
Author(s): James P. Theiler, Christopher X. Ren, Los Alamos National Lab. (United States)
On demand
Show Abstract + Hide Abstract
Gaussianization is a recently suggested approach for density estimation from data drawn from a decidedly non-Gaussian, and possibly high dimensional, distribution. The key idea is to learn a transformation that, when applied to the data, leads to an approximately Gaussian distribution. The density, for any given point in the original distribution, is then given by the determinant of the transformation's Jacobian at that point, multiplied by the (analytically known) density of the Gaussian for the transformed data. In this work, we investigate the use of distilled machine learning to provide a compact implementation of the Gaussianization transform (which in usual practice is obtained iteratively), thereby enabling faster computation, better controlled regularization, and more direct estimation of the Jacobian. While density estimation underlies many statistical analyses, our interest is in hyperspectral detection problems.
11843-19
Author(s): Marian Anghel, Patrick Kelly, Nicolas Hengartner, Los Alamos National Lab. (United States)
On demand
Show Abstract + Hide Abstract
Quantifying the predictive uncertainty of Neural Network (NN) models remains a dificult, unsolved problem especially since the ground truth is usually not available. In this work we evaluate many regression uncertainty estimation models and discuss their accuracy using training sets where the uncertainty is known exactly. We compare three regression models, a homoscedastic model, a heteroscedastic model, and a quantile model and show that: while all models can learn an accurate estimation of response, the accurate estimation of uncertainty is very difficult; the quantile model has the best performance in estimating uncertainty; model bias is confused with uncertainty and it is very difficult to disentangle the two when we have only one measurement per training point; improved accuracy of the estimated uncertainty is possible, but the experimental cost for learning uncertainty is very large since it requires multiple estimations of the response almost everywhere in the input space.
11843-20
Author(s): Michael Grunwald, Matthias Hermann, Fabian Freiberg, Matthias O. Franz, Hochschule Konstanz (Germany)
On demand
Show Abstract + Hide Abstract
Parametric texture models have been applied successfully to synthesize artificial images. Psychophysical studies show that under defined conditions observers are unable to differentiate between model-generated and original natural textures. In industrial applications the reverse case is of interest: a texture analysis system should decide if human observers are able to discriminate between a reference and a novel texture. Here, we implemented a biologically-inspired novelty detection approach. Assuming that the features used for texture synthesis are important for human texture perception, we compare psychophysical as well as learnt texture representations based on activations of a pretrained CNN in a novelty detection scenario. Based on a digital print inspection scenario we show that psychophysical texture representations are able to outperform CNN-encoded features.
11843-22
Author(s): Egor Sedov, Novosibirsk State Univ. (Russian Federation), Aston Institute of Photonic Technologies, Aston Univ. (United Kingdom); Yaroslav Prylepskiy, Aston Institute of Photonic Technologies, Aston Univ. (United Kingdom); Igor Chekhovskoy, Novosibirsk State Univ. (Russian Federation); Sergei Turitsyn, Aston Institute of Photonic Technologies, Aston Univ. (United Kingdom), Novosibirsk State Univ. (Russian Federation)
On demand
Show Abstract + Hide Abstract
In this work we demonstrate that the high-accuracy computation of the continuous nonlinear spectrum can be performed by using artificial neural networks. We propose the artificial neural network (NN) architecture that can efficiently perform the nonlinear Fourier (NF) optical signal processing. The NN consists of sequential convolution layers and fully connected output layers. This NN predicts only one component of the continuous NF spectrum, such that two identical NNs have to be used to predict the real and imaginary parts of the reflection coefficient. To train the NN, we precomputed 94035 optical signals. 9403 signals were used for validation and excluded from training. The final value of the relative error for the entire validation dataset was less than 0.3%. Our findings highlight the fundamental possibility of using the NNs to analyze and process complex optical signals, when the conventional algorithms can fail to deliver an acceptable result.
11843-49
Author(s): Zhicheng Cao, Heng Zhao, Xidian University (China); Shufen Cao, Case Western Reserve University (United States); Liaojun Pang, Xidian University (China)
On demand
11843-51
Author(s): Zhicheng Cao, Xidian University (China); Xing Chen, Siemens Ltd, China (China); Shufen Cao, Case Western Reserve University (United States); Liaojun Pang, Xidian University (China)
On demand
Session 2: Big Data, Simulations, and Physics
In person / Livestream: 4 August 2021 • 2:20 PM - 2:40 PM PDT | Conv. Ctr. Room 7B
Session Chair: Bob Hainsey, SPIE (United States)
11843-26
Author(s): Aaron Dant, Steve Kacenjar, Ron Neely, ASRC Federal Holding, Co. (United States)
On demand | Presented Live 4 August 2021
Show Abstract + Hide Abstract
Synthetic generation of images provides a viable approach to training machine learning models when real data is sparse.  However, the process of rendering useful synthetic training data takes significant time. This paper examines the utility of a self-directed feedback training method that improves the convergence rate when training such models by using small batches of training data and observing the classification performance for each class being simulated. The performance results determine adjustments to subsequent training cases. Training of a person re-identification model with and without this feedback control is assessed.
11843-28
Author(s): Naveenta Gautam, Indian Institute of Technology, Delhi (India); Amol Choudhary, Brejesh Lall, Indian Institute of Technology Delhi (India)
On demand
11843-23
Author(s): Daniel Wang, New Mexico Consortium (United States), Los Alamos National Lab. (United States); Howard Pritchard, New Mexico Consortium (United States); Garrett T. Kenyon, Los Alamos National Lab. (United States)
On demand
Show Abstract + Hide Abstract
Effective up-sampling based on convolutional sparse coding may offer an efficient method for generating high quality, high resolution computational fluid dynamics simulations from lower quality, lower resolution simulations. Sparse coding, which incorporates lateral inhibition, implements neural network architecture that are more biologically accurate than classic deep learning algorithms. Sparse coding neural networks can up-sample and extrapolate spatiotemporally high resolution frames from a low-resolution or decimated input. The quality of the reconstruction can be compared against both an original non-decimated input and the reconstruction quality of competing deep learning networks.
11843-24
Author(s): Nga Nguyen-Fotiadis, Los Alamos National Lab. (United States); Garry R. Maskaly, Lawrence Livermore National Lab. (United States); Andy S. Liao, Christopher L. Fryer, Deborah A. Shutt, Vincent P. Chiravalle, Kyle S. Hickmann, Los Alamos National Lab. (United States)
On demand
Show Abstract + Hide Abstract
We demonstrate and analyze a new methodology for detecting and characterizing shocks in high energy density physics (HEDP) experiments using observed radiographs. Our method consists of simulating many variations of an HEDP experiment using a multi-physics modeling code to produce synthetic radiographs that emulate the actual experimental data collection setup. Shock contours are defined by the peaks of the density derivative values obtained at each radial coordinate of an X-rayed cylindrical object, giving us a ground truth of shock position that would not be available directly from the observed radiograph. We investigate four different state-of-the-art deep convolutional neural networks, Xception, ResNet152, VGG19, and U-Net, for use in regressing the HEDP radiograph to the shock position. We find that the different network architectures are better tuned for locating distinct shock characteristics, equivalent to detecting shock-waves at multiple scales.
11843-25
Author(s): Adam Good, Howard Pritchard, Lissa Moore, Garrett T. Kenyon, Los Alamos National Lab. (United States)
On demand
Session 3: Consumer Applications and E-Commerce
In person / Livestream: 4 August 2021 • 2:40 PM - 3:00 PM PDT | Conv. Ctr. Room 7B
Session Chair: Bob Hainsey, SPIE (United States)
11843-29
Author(s): Daniel Adams, Cory Merkel, Rochester Institute of Technology (United States)
On demand
Show Abstract + Hide Abstract
Smart assistant usage has increased significantly with the AI boom and growth of IoT, though many smart assistants underperform when tasked with interpreting atypical speech input. Dysarthria, heavy accents, and Deaf and hard-of-hearing speech characteristics prove difficult for smart assistants to interpret despite the large amounts of diverse data used to train automatic speech recognition models. We utilize a Transformer model pre-trained on the Librispeech dataset and fine-tuned on the Torgo database. We also develop a method for utilizing proprietary automatic speech recognition models prior to interpretation on smart assistants utilizing a Raspberry Pi 4 and a Google Home smart assistant device. The most accurate model shows a 20.2% character error rate with a corresponding 29.0% word error rate on a subset of medium intelligibility samples. We also highlight the newly developed speech data collection web application, My-Voice.
11843-31
Author(s): Gaurav Anand, Siyun Wang, Karl Ni, Etsy, Inc. (United States)
On demand
Show Abstract + Hide Abstract
Visual search can aid an e-commerce platform by providing appropriate recommendations where semantic labeling and associated metadata is missing. In this work, we detail the bootstrapping of our pipeline that powers visually similar recommendations. While a common approach leverages learned representation from classification tasks using convolutional neural networks, the crux of the problem are the attributes and their ontology. The image representations are learned by a ResNet model architecture, trained from scratch on 3000+ classes. To scale the nearest neighbors on millions of items, we leverage quantization schemes like HNSW, IVF, and PCA. These are incrementally inferenced for new items on multiple GPUs and optimized for data throughput, and indexed in the ANN. Finally, in order to verify the appropriateness, we use an extensive human evaluation pipeline and quality control. In this work, we share all lessons learned from the experiments we conducted for a successful launch.
11843-32
Author(s): Chris Xu, Raphael Louca, Karl Ni, Etsy, Inc. (United States)
On demand
Wednesday Surf Rock Chill and Beer Reception
In person: 4 August 2021 • 4:30 PM - 5:30 PM PDT | Conv. Ctr. West Terrace (Upper Level)
Join other attendees for some light appetizers as you relax to the vibes of a California surf band. Network with company representatives and other technical professionals and enjoy some San Diego sunshine.
Conference Chair
Lawrence Livermore National Lab. (United States)
Conference Chair
Univ. of Dayton (United States)
Conference Chair
NVIDIA Corp. (United States)
Conference Co-Chair
Lawrence Livermore National Lab. (United States)
Conference Co-Chair
Old Dominion Univ. (United States)
Program Committee
St. Jude Children's Research Hospital (United States)
Program Committee
BeamIO (United States)
Program Committee
NVIDIA Corp. (United States)
Program Committee
Pyxeda, Inc. (United States)
Program Committee
Lawrence Livermore National Lab. (United States)
Program Committee
Lawrence Livermore National Lab. (United States)
Program Committee
Univ. of Texas San Antonio Chapter (United States)
Program Committee
Lawrence Livermore National Lab. (United States)
Program Committee
Univ. of Dayton (United States)
Program Committee
Etsy, Inc. (United States)
Program Committee
Lawrence Livermore National Lab. (United States)
Program Committee
Manar D. Samad
Tennessee State Univ. (United States)
Program Committee
ArchiFiction, Inc. (United States)
Program Committee
Los Alamos National Lab. (United States)