Search NASA⌕ Search

SEARCH · Search NASA

Results for “network inference”

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 235 records · Page 13

The circum-Chryse region as a possible example of a hydrologic cycle on Mars: Geologic observations and theoretical evaluation

The transection and superposition relationships among channels, chaos, surface materials units, and other features in the circum-Chryse region of Mars were used to evaluate relative age relationships and evolution of flood events. Channels and chaos in contact (with one another) were treated as single discrete flood-carved systems. Some outflow channel systems form networks and are inferred to have been created by multiple flood events. Within some outflow channel networks, several separate individual channel systems can be traced to a specific chaos which acted as flood-source area to that specific flood channel. Individual flood-carved systems were related to widespread materials units or other surface features that served as stratigraphic horizons. Chryse outflow channels are inferred to have formed over most of the perceivable history of Mars. Outflow channels are inferred to become younger with increasing proximity to the Chryse basin. The relationship of subsequent outflow channel sources to the sources of earlier floods is inferred to disfavor episodic flooding due to the progresssive tapping of a juvenile near-surface water supply. Instead, we propose the circum-Chryse region as a candidate site of past hydrological recycling. The discharge rates necessary to carve the circum-Chryse outflow channels would have inevitably formed temporary standing bodies of H2O on the Martian surface where the flood-waters stagnated and pooled (the Chryse basin is topographically enclosed). These observations and inferences have led us to formulate and evaluate two hypotheses. Our numerical evaluations indicate that of these two hypotheses formulated, the groundwater seep cycle seems by far the more viable. Further observations from forthcoming missions may permit the determination of which mechanisms may have operated to recycle the Chryse flood-waters.

Moore, Jeffrey M.↗

The circum-Chryse region as a possible example of a hydrologic cycle on Mars: Geologic observations and theoretical evaluation

The transection and superposition relationships among channels, chaos, surface materials units, and other features in the circum-Chryse region of Mars were used to evaluate relative age relationships and evolution of flood events. Channels and chaos in contact (with one another) were treated as single discrete flood-carved systems. Some outflow channel systems form networks and are inferred to have been created by multiple flood events. Within some outflow channel networks, several separate individual channel systems can be traced to a specific chaos which acted as flood-source area to that specific flood channel. Individual flood-carved systems were related to widespread materials units or other surface features that served as stratigraphic horizons. Chryse outflow channels are inferred to have formed over most of the perceivable history of Mars. Outflow channels are inferred to become younger with increasing proximity to the Chryse basin. In addition, outflow channels closer to the basin show a greater diversity in age. The relationship of subsequent outflow channel sources to the sources of earlier floods is inferred to disfavor episodic flooding due to the progressive tapping of a juvenile near-surface water supply. Instead, we propose the circum-Chryse region as a candidate site of past hydrological recycling. The discharge rates necessary to carve the circum-Chryse outflow channels would have inevitably formed temporary standing bodies of H2O on the Martian surface where the flood-waters stagnated and pooled (the Chryse basin is topographically enclosed). These observations and inferences have led us to formulate and evaluate two hypotheses: (1) large amounts of the sublimated H2O off the Chryse basin flood lakes precipitated (snowed) onto the flood-source highlands and this H2O was incorporated into the near surface, recharging the H2O sources, making possible subsequent deluges; and (2) ponded flood-water in Chryse basin drained back down an anti basinward dipping subsurface layer accessed along the southern edge of the lake, recharging the flood-source aquifers. H2O not redeposited in the flood-source region was largely lost to the hydrologic cycle. This loss progressively lowered the vitality of the cycle, probably by now killing it. Our numerical evaluations indicate that of the two hypotheses we formulated, the groundwater seep cycle seems by far the more viable. Optimally, approx. 3/4 of the original mass of an ice-covered cylindrical lake (albedo 0.5, 1 km deep, 100-km radius, draining along its rim for one quarter of its circumference into substrata with a permeability of 3000 darcies) can be modeled to have moved underground (on timescales of the order of 10(exp 3) years) before the competing mechanisms of sublimation and freeze down choked off further water removal. Once underground, this water can travel distances equal to the separation between Chryse basin and flood-source sites in geologically short (approx. 10(exp 6) year-scale) times. Conversely, we calculate that optimally only approx. 40% of the H2O carried from Chryse can condense at the highlands, and most of the precipitate would either collect at the base of the highlands/lowlands scarp or sublimate at rates greater than it would accumulate over the flood-source sites. Further observations from forthcoming missions may permit the determination of which mechanisms may have operated to recycle the Chryse flood-waters.

Moore, Jeffrey M.↗

The Circum-Chryse Region as a Possible Example of a Hydrologic Cycle on Mars: Geologic Observations and Theoretical Evaluation

The transection and superposition relationships among channels, chaos, surface materials units, and other features in the circum-Chryse region of Mars were used to evaluate relative age relationships and evolution of flood events. Channels and chaos in contact (with one another) were treated as single discrete flood-carved systems. Some outflow channel systems form networks and are inferred to have been created by multiple flood events. Within some outflow channel networks, several separate individual channel systems can be traced to a specific chaos which acted as flood-source area to that specific flood channel. Individual flood-carved systems were related to widespread materials units or other surface features that served as stratigraphic horizons. Chryse outflow channels are inferred to have formed over most of the perceivable history of Mars. Outflow channels are inferred to become younger with increasing proximity to the Chryse basin. In addition, outflow channels closer to the basin show a greater diversity in age. The relationship of subsequent outflow channel sources to the sources of earlier floods is inferred to disfavor episodic flooding due to the progressive tapping of a juvenile near-surface water supply. Instead, we propose the circum-Chryse region as a candidate site of past hydrological recycling. The discharge rates necessary to carve the circum-Chryse outflow channels would have inevitably formed temporary standing bodies of H2O on the Martian surface where the flood-waters stagnated and pooled (the Chryse basin is topographically enclosed). These observations and inferences have led us to formulate and evaluate two hypotheses: (1) large amounts of the sublimated H2O off the Chryse basin flood lakes precipitated (snowed) onto the flood-source highlands and this H2O was incorporated into the near surface, recharging the H2O sources, making possible subsequent deluges; and (2) ponded flood-water in Chryse basin drained back down an anti basinward dipping subsurface layer accessed along the southern edge of the lake, recharging the flood-source aquifers. H2O not redeposited in the flood-source region was largely lost to the hydrologic cycle. This loss progressively lowered the vitality of the cycle, probably by now killing it. Our numerical evaluations indicate that of the two hypotheses we formulated, the groundwater seep cycle seems by far the more viable. Optimally, approximately 3/4 of the original mass of an ice-covered cylindrical lake (albedo 0.5, 1 km deep, 100-km radius, draining along its rim for one quarter of its circumference into substrata with a permeability of 3000 darcies) can be modeled to have moved underground (on timescales of the order of 10(exp 3) years) before the competing mechanisms of sublimation and freeze down choked off further water removal. Once underground, this water can travel distances equal to the separation between Chryse basin and flood-source sites in geologically short (approximately 10(exp 6) year-scale) times. Conversely, we calculate that optimally only approximately 40% of the H2O carried from Chryse can condense at the highlands, and most of the precipitate would either collect at the base of the highlands/lowlands scrap or sublimate at rates greater than it would accumulate over the flood-source sites. Further observations from forthcoming missions may permit the determination of which mechanisms may have operated to recycle the Chryse flood-waters.

Moore, Jeffrey M.↗

Normalizing flows for likelihood-free inference with fusion simulations

Fluid-based scrape-off layer transport codes, such as UEDGE, are heavily utilized in tokamak analysis and design, but typically require user-specified anomalous transport coefficients to match experiments. Determining the uniqueness of these parameters and the uncertainties in them to match experiments can provide valuable insights to fusion scientists. Here, we leverage recent work in the area of likelihood-free inference (‘simulation-based inference’) to train a neural network, which enables accurate statistical inference of the anomalous transport coefficients given experimental plasma profile input. UEDGE is treated as a black-box simulator and runs multiple times with anomalous transport coefficients sampled from priors, and the neural network is trained on these simulations to emulate the posterior. The neural network is trained as a normalizing flow model for density estimation, allowing it to accurately represent complicated, high-dimensional distribution functions. With a fixed simulation budget, we compare a single-round procedure to a multi-round approach that guides the training simulations toward a specific target observation. Finally, we discuss the future possibilities for use of amortized models, which train on a wide range of simulations and enable fast statistical inference for results during experiments.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

A quantitative comparison of the fingerprint of twinned microstructures through surface and three-dimensional techniques

Assessing the fingerprint of a material’s microstructure is key for supporting materials design. With the emergence of a wide range of 3D characterization techniques, it is critical to understand the main differences in fingerprints reconstructed from 2D and 3D datasets. To this end, we introduce a graph-based microstructure reconstruction framework that enables structural comparisons of twin domain networks in high purity Ti using 3D and 2D electron backscatter diffraction. Insights into the structure of the twin networks are facilitated by combining statistical analysis of twin crystallography with visual and graphical analysis of the novel graph abstractions of the twins. We demonstrate that compared to 3D reconstructions, conventional 2D views of twinning miss key aspects of the microstructure including the high interconnectivity of domains into networks that span the full reconstruction volume. The reduced cross-grain and in-grain twin connectivity typically observed in 2D has notable implications on our understanding of how twinning mediates the plastic response of microstructures and how twin networks evolve. It is thus clear that 3D characterization is critical for accurately inferring both twin network morphologies as well as the key unit processes facilitating network formation.

36 MATERIALS SCIENCE↗

Geometry-aware training of factorized layers in tensor Tucker format

Reducing parameter redundancies in neural network architectures is crucial for achieving feasible computational and memory requirements during train and inference of large networks. Given its easy implementation and flexibility, one promising approach is layer factorization, which reshapes weight tensors into a matrix format and parameterizes it as the product of two rank-r matrices. However, this family of approaches often requires an initial full-model warm-up phase, prior knowledge of a feasible rank, and it is sensitive to parameter initialization.In this work, we introduce a novel approach to train the factors of a Tucker decomposition of the weight tensors. Our training proposal proves to be optimal in locally approximating the original unfactorized dynamics and stable for the initialization. Furthermore, the rank of each mode is dynamically updated during training.We provide a theoretical analysis of the algorithm, showing convergence, approximation and local descent guarantees. The method's performance is further illustrated through a variety of experiments, showing remarkable training compression rates and comparable or even better performance than the full baseline and alternative layer factorization strategies.

Zangrando, Emanuele [Gran Sasso Science Institute ↗

Multiverse: Bayesian model selection for neural networks

Multiverse is a code repository for a set of tools for Bayesian model selection for neural networks. The goal of the tools is to provide capabilities for selecting among prior and model specifications for Bayesian neural networks. Multiverse will include tools for creating Bayesian neural networks, evaluating the Bayesian model evidence, and performing inference in Bayesian neural networks. These components are written in Python, a high-level programming language that takes advantage of the Python ecosystem of high-quality open-source packages for machine learning and signal processing.

Klein, Natalie↗

The internet worm

In November 1988 a worm program invaded several thousand UNIX-operated Sun workstations and VAX computers attached to the Research Internet, seriously disrupting service for several days but damaging no files. An analysis of the work's decompiled code revealed a battery of attacks by a knowledgeable insider, and demonstrated a number of security weaknesses. The attack occurred in an open network, and little can be inferred about the vulnerabilities of closed networks used for critical operations. The attack showed that passwork protection procedures need review and strengthening. It showed that sets of mutually trusting computers need to be carefully controlled. Sharp public reaction crystalized into a demand for user awareness and accountability in a networked world.

Denning, Peter J.↗

Reservoir Modeling by Data Integration via Intermediate Spaces and Artificial Intelligence Tools in MPS Simulation Frameworks

Conditioning stochastic simulations are very important in many geostatistical applications that call for the introduction of nonlinear and multiple-point data in reservoir modeling. Here, a new methodology is proposed for the incorporation of different data types into multiple-point statistics (MPS) simulation frameworks. Unlike the previous techniques that call for an approximate forward model (filter) for integration of secondary data into geologically constructed models, the proposed approach develops an intermediate space where all the primary and secondary data are easily mapped onto. Definition of the intermediate space, as may be achieved via application of artificial intelligence tools like neural networks and fuzzy inference systems, eliminates the need for using filters as in previous techniques. The applicability of the proposed approach in conditioning MPS simulations to static and geologic data is verified by modeling a real example of discrete fracture networks using conventional well-log data. The training patterns are well reproduced in the realizations, while the model is also consistent with the map of secondary data.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Coupled flow accumulation and atmospheric blocking govern flood duration

Abstract We present a physically based Bayesian network model for inference and prediction of flood duration that allows for a deeper understanding of the nexus of antecedent flow regime, atmospheric blocking, and moisture transport/release mechanisms. Distinct scaling factors at the land surface and regional atmospheric levels are unraveled using this Bayesian network model. Land surface scaling explains the variability in flood duration as a function of cumulative exceedance index, a new measure that represents the evolution of the flood in the basin. Dynamic atmospheric scaling explains the cumulative exceedance index using the interaction between atmospheric blocking system and the synergistic model of wind divergence and atmospheric water vapor. Our findings underline that the synergy between a large persistent low-pressure blocking system and a higher rate of divergent wind often triggers a long-duration flood, even in the presence of moderate moisture supply in the atmosphere. This condition in turn causes an extremely long-duration flood if the basin-wide cumulative flow prior to the flood event was already high. Thus, this new land-atmospheric interaction framework integrates regional flood duration scaling and dynamic atmospheric scaling to enable the coupling of ‘horizontal’ (for example, streamflow accumulation inside the basin) and ‘vertical’ flow of information (for example, interrelated land and ocean-atmosphere interactions), providing an improved understanding of the critical forcing of regional hydroclimatic systems. This Bayesian model approach is applied to the Missouri River Basin, which has the largest system of reservoirs in the United States. Our predictive model can aid in decision support systems for the protection of national infrastructure against long-duration flood events.

Najibi, Nasser (ORCID:0000000205157192)↗

From IMT Device Measurements to Network-Level Consequences: When Learning Suppresses Beyond-LIF Neuron Dynamics

Emerging neuromorphic devices such as insulator--metal transition (IMT) devices exhibit complex temporal dynamics, including slow internal state memory, hysteresis, and burst-like firing, which are poorly captured by conventional leaky integrate-and-fire (LIF) neurons. However, it remains unclear when such dynamics influence learning and inference at the network level, particularly under commonly used unsupervised plasticity rules. We present a controlled, full-stack co-design study spanning experimental characterization of individual IMT devices, compact neuron model development, and large-scale spiking network simulations with identical architectures and learning rules. Rather than optimizing benchmark accuracy, our goal is to diagnose when neuron-level dynamics survive learning and competition, and when they are suppressed, to inform the co-design of devices, networks, and learning rules that can exploit beyond-LIF complexity.

42 ENGINEERING↗

Cross Inference of Throughput Profiles Using Micro Kernel Network Method

Dedicated network connections are being increasingly deployed in cloud, centralized and edge computing and data infrastructures, whose throughput profiles are critical indicators of the underlying data transfer performance. Due to the cost and disruptions to physical infrastructures, network emulators, such as Mininet, are often used to generate measurements needed to estimate throughput profiles, typically expressed as a function of the connection round trip time. The profiles estimated using measurements from such emulated networks are usually inaccurate for high bandwidth and high latency connections, since they do not accurately reflect the critical network transport dynamics mainly due to computing and memory constraints of the host. We present a machine learning (ML) method to estimate the throughput profiles using emulation measurements to closely match the testbed and production network profiles. In particular, we propose a micro Kernel Network (mKN) that provides baseline throughput measurements on the host running Mininet emulations, which are used to learn a regression map that converts them to the corresponding testbed measurement estimates. Once initially learned, this map is applied to measurements from subsequent network emulations on the same host. We present experimental measurements to illustrate this approach, and derive generalization equations for the proposed mKN-ML method. Using a four-site scenario emulation, we show the effectiveness of this method in providing accurate concave throughput profiles from inaccurate convex or non-smooth ones indicated by Mininet emulation.

Rao, Nageswara↗

Scalable optimal Bayesian classification of single-cell trajectories under regulatory model uncertainty

Single-cell gene expression measurements offer opportunities in deriving mechanistic understanding of complex diseases, including cancer. However, due to the complex regulatory machinery of the cell, gene regulatory network (GRN) model inference based on such data still manifests significant uncertainty. The goal of this paper is to develop optimal classification of single-cell trajectories accounting for potential model uncertainty. Partially-observed Boolean dynamical systems (POBDS) are used for modeling gene regulatory networks observed through noisy gene-expression data. We derive the exact optimal Bayesian classifier (OBC) for binary classification of single-cell trajectories. The application of the OBC becomes impractical for large GRNs, due to computational and memory requirements. To address this, we introduce a particle-based single-cell classification method that is highly scalable for large GRNs with much lower complexity than the optimal solution. The performance of the proposed particle-based method is demonstrated through numerical experiments using a POBDS model of the well-known T-cell large granular lymphocyte (T-LGL) leukemia network with noisy time-series gene-expression data.

59 BASIC BIOLOGICAL SCIENCES↗

SU-F-E-09: Respiratory Signal Prediction Based On Multi-Layer Perceptron Neural Network Using Adjustable Training Samples

Purpose: Dynamic tracking of moving organs, such as lung and liver tumors, under radiation therapy requires prediction of organ motions prior to delivery. The shift of moving organ may change a lot due to huge transform of respiration at different periods. This study aims to reduce the influence of that changes using adjustable training signals and multi-layer perceptron neural network (ASMLP). Methods: Respiratory signals obtained using a Real-time Position Management(RPM) device were used for this study. The ASMLP uses two multi-layer perceptron neural networks(MLPs) to infer respiration position alternately and the training sample will be updated with time. Firstly, a Savitzky-Golay finite impulse response smoothing filter was established to smooth the respiratory signal. Secondly, two same MLPs were developed to estimate respiratory position from its previous positions separately. Weights and thresholds were updated to minimize network errors according to Leverberg-Marquart optimization algorithm through backward propagation method. Finally, MLP 1 was used to predict 120∼150s respiration position using 0∼120s training signals. At the same time, MLP 2 was trained using 30∼150s training signals. Then MLP is used to predict 150∼180s training signals according to 30∼150s training signals. The respiration position is predicted as this way until it was finished. Results: In this experiment, the two methods were used to predict 2.5 minute respiratory signals. For predicting 1s ahead of response time, correlation coefficient was improved from 0.8250(MLP method) to 0.8856(ASMLP method). Besides, a 30% improvement of mean absolute error between MLP(0.1798 on average) and ASMLP(0.1267 on average) was achieved. For predicting 2s ahead of response time, correlation coefficient was improved from 0.61415 to 0.7098.Mean absolute error of MLP method(0.3111 on average) was reduced by 35% using ASMLP method(0.2020 on average). Conclusion: The preliminary results demonstrate that the ASMLP respiratory prediction method is more accurate than MLP method and can improve the respiration forecast accuracy.

60 APPLIED LIFE SCIENCES↗

Predictability and reduced order modeling in stochastic reaction networks.

Many systems involving chemical reactions between small numbers of molecules exhibit inherent stochastic variability. Such stochastic reaction networks are at the heart of processes such as gene transcription, cell signaling or surface catalytic reactions, which are critical to bioenergy, biomedical, and electrical storage applications. The underlying molecular reactions are commonly modeled with chemical master equations (CMEs), representing jump Markov processes, or stochastic differential equations (SDEs), rather than ordinary differential equations (ODEs). As such reaction networks are often inferred from noisy experimental data, it is not uncommon to encounter large parametric uncertainties in these systems. Further, a wide range of time scales introduces the need for reduced order representations. Despite the availability of mature tools for uncertainty/sensitivity analysis and reduced order modeling in deterministic systems, there is a lack of robust algorithms for such analyses in stochastic systems. In this talk, we present advances in algorithms for predictability and reduced order representations for stochastic reaction networks and apply them to bistable systems of biochemical interest. To study the predictability of a stochastic reaction network in the presence of both parametric uncertainty and intrinsic variability, an algorithm was developed to represent the system state with a spectral polynomial chaos (PC) expansion in the stochastic space representing parametric uncertainty and intrinsic variability. Rather than relying on a non-intrusive collocation-based Galerkin projection [1], this PC expansion is obtained using Bayesian inference, which is ideally suited to handle noisy systems through its probabilistic formulation. To accommodate state variables with multimodal distributions, an adaptive multiresolution representation is used [2]. As the PC expansion directly relates the state variables to the uncertain parameters, the formulation lends itself readily to sensitivity analysis. Reduced order modeling in the time dimension is accomplished using a Karhunen-Loeve (KL) decomposition of the stochastic process in terms of the eigenmodes of its covariance matrix. Subsequently, a Rosenblatt transformation relates the random variables in the KL decomposition to a set of independent random variables, allowing the representation of the system state with a PC expansion in those independent random variables. An adaptive clustering method is used to handle multimodal distributions efficiently, and is well suited for high-dimensional spaces. The spectral representation of the stochastic reaction networks makes these systems more amenable to analysis, enabling a detailed understanding of their functionality, and robustness under experimental data uncertainty and inherent variability.

59 BASIC BIOLOGICAL SCIENCES↗

wa-hls4ml and lui-gnn: A benchmark and GNN-based surrogate model for hls4ml resource and latency estimation

As machine learning (ML) increasingly serves as a tool for addressing real-time challenges in scientific applications, the development of advanced tooling has significantly reduced the time required to iterate on various designs. These advancements have solved major obstacles, but also exposed new challenges. For example, processes that were not previously considered bottlenecks, such as model synthesis, are now becoming limiting factors in the rapid iteration of designs. To reduce these emerging constraints, multiple efforts are being launched toward designing an ML-based surrogate model that estimates resource usage of synthesized accelerator architectures. This model would reduce the design iteration time, especially when designing within a set of given hardware constraints. This approach shows considerable potential, but as it stands, the effort is early and would benefit from coordination and standardization to assist future work as it emerges. We introduce wa-hls4ml, a benchmark for ML accelerator resource and latency estimation, and its corresponding initial dataset of more than 100,000 fully connected neural networks, all synthesized using hls4ml and targeting Xilinx FPGAs. In addition to the resource utilization and latency data provided, the dataset includes generated artifacts and log files for many of the synthesized neural networks, in order to support future research in ML-based code generation. The benchmark evaluates the performance of resource and latency predictors against several common ML model architectures, primarily originating from scientific domains, as exemplar models, as well as the average performance across a subset of the dataset. We measure the performance of a given predictor model through multiple metrics, including $R^2$ score and SMAPE on regression tasks, as well as inference time to further characterize the estimator under test. Additionally, we introduce the latency/utilization inference graph neural network (lui-gnn), a surrogate model that uses a graph neural network to represent input architectures in the form of a directed graph. This graph representation allows for a diverse set of model architectures to all be effectively handled by a surrogate model. We present the architecture and performance of the model, as evaluated by the new proposed benchmark, including SMAPE, $R^2$ score, and inference times, and find that lui-gnn generally predicts latency and utilization for the 75\% quantile within several percent of the synthesized resources on the synthetic test dataset, indicating that this approach of estimating resource and latency via a surrogate models has promise and warrants further research.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Inferring safety critical events from vehicle kinematics in naturalistic driving environment: Application of deep learning Algorithms

Advances in sensing technology has enabled the collection of countless terabytes of second-by-second kinematics data. Such data provides opportunities for real-time monitoring of driving behavior and identification of safety critical events (SCEs) including crashes and near crashes. The concept of volatility is relevant in this context, which identifies instability and erratic variations in driving behavior prior to involvement in SCEs. This study utilized vehicle kinematics from a large-scale naturalistic driving data to develop a deep learning approach based on 1D convolutional neural networks (CNN) for inferring SCEs. The data are unique in the sense that such accurate pre-crash data at high fidelity are not available in traditional crash repositories. This study contributes to the literature by providing a first attempt at predicting responses to SCEs by developing deep learning-based CNN architectures using novel driving volatility based kinematic thresholds for a sample of 9553 events. The key contribution lies in developing a volatility-based CNN input layout that is acceptable to CNN schemes and represents the motion kinematics such as speed, acceleration and volatility measures. Several 1D-CNN architectures were developed using layers, numbers of convolutions, layer patterns, and kernels. Shallow and deep architectures were tested, revealing higher accuracy of shallow architectures in detecting SCEs. The optimal number of epochs were identified using an early stopping method while the CNN performance was improved by increasing the number of epochs. The ensemble CNN had the highest predictive accuracy of 95.6% for detection of crashes and near crashes, which was 2.5% higher than the optimal CNN using 20% hold out test data. The ensemble CNN also outperformed classical machine learning models and model performance reported in past studies on detection of SCEs. Finally, these results have implications for identification of safety hotspots and providing real-time alerts and warnings in connected and highly automated vehicle environment including society of automotive engineers levels 3–5.

42 ENGINEERING↗