Downloads & Free Reading Options - Results

We are now witnessing the increasing availability of event stream data, i.e., a sequence of events with each event typically being denoted by the time it occurs and its mark information (e.g., event type). A fundamental problem is to model and predict such kind of marked temporal dynamics, i.e., when the next event will take place and what its mark will be. Existing methods either predict only the mark or the time of the next event, or predict both of them, yet separately. Indeed, in marked temporal dynamics, the time and the mark of the next event are highly dependent on each other, requiring a method that could simultaneously predict both of them. To tackle this problem, in this paper, we propose to model marked temporal dynamics by using a mark-specific intensity function to explicitly capture the dependency between the mark and the time of the next event. Extensive experiments on two datasets demonstrate that the proposed method outperforms state-of-the-art methods at predicting marked temporal dynamics.

http://uf.catalog.fcla.edu/uf.jsp?st=UF002056070&ix=pm&I=0&V=D&pm=1

We present the multiplicative recurrent neural network as a general model for compositional meaning in language, and evaluate it on the task of fine-grained sentiment analysis. We establish a connection to the previously investigated matrix-space models for compositionality, and show they are special cases of the multiplicative recurrent net. Our experiments show that these models perform comparably or better than Elman-type additive recurrent neural networks and outperform matrix-space models on a standard fine-grained sentiment analysis corpus. Furthermore, they yield comparable results to structural deep models on the recently published Stanford Sentiment Treebank without the need for generating parse trees.

Natural Language Processing (NLP) systems commonly leverage bag-of-words co-occurrence techniques to capture semantic and syntactic word relationships. The resulting word-level distributed representations often ignore morphological information, though character-level embeddings have proven valuable to NLP tasks. We propose a new neural language model incorporating both word order and character order in its embedding. The model produces several vector spaces with meaningful substructure, as evidenced by its performance of 85.8% on a recent word-analogy task, exceeding best published syntactic word-analogy scores by a 58% error margin. Furthermore, the model includes several parallel training methods, most notably allowing a skip-gram network with 160 billion parameters to be trained overnight on 3 multi-core CPUs, 14x larger than the previous largest neural network.

When working with three-dimensional data, choice of representation is key. We explore voxel-based models, and present evidence for the viability of voxellated representations in applications including shape modeling and object classification. Our key contributions are methods for training voxel-based variational autoencoders, a user interface for exploring the latent space learned by the autoencoder, and a deep convolutional neural network architecture for object classification. We address challenges unique to voxel-based representations, and empirically evaluate our models on the ModelNet benchmark, where we demonstrate a 51.5% relative improvement in the state of the art for object classification.

Recurrent neural network (RNN) based character-level language models (CLMs) are extremely useful for modeling out-of-vocabulary words by nature. However, their performance is generally much worse than the word-level language models (WLMs), since CLMs need to consider longer history of tokens to properly predict the next one. We address this problem by proposing hierarchical RNN architectures, which consist of multiple modules with different timescales. Despite the multi-timescale structures, the input and output layers operate with the character-level clock, which allows the existing RNN CLM training approaches to be directly applicable without any modifications. Our CLM models show better perplexity than Kneser-Ney (KN) 5-gram WLMs on the One Billion Word Benchmark with only 2% of parameters. Also, we present real-time character-level end-to-end speech recognition examples on the Wall Street Journal (WSJ) corpus, where replacing traditional mono-clock RNN CLMs with the proposed models results in better recognition accuracies even though the number of parameters are reduced to 30%.

In the article, a neural network analysis of the electrical activity of the human brain, which is included in the concept of "norm". A neural network model (Kohonen neural network) was built, which allows automatically classify electroencephalograms of an organized type.

In this study, it was aimed to predict elementary education teacher candidates' achievements in "Science and Technology Education I and II" courses by using artificial neural networks. It was also aimed to show the independent variables importance in the prediction. In the data set used in this study, variables of gender, type of education, field of study in high school and transcript information of 14 courses including end-of-term letter grades were collected. The fact that the artificial neural network performance in this study was R = 0.84 for the Science and Technology Education I course, and R = 0.84 for the Science and Technology Education II course shows that the network performance overlaps with the findings obtained from the related studies.

A neural network-based parking system with real-time license plate detection and vacant space detection using hyper parameter optimization is presented. When number of epochs increased from 30, 50 to 80 and learning rate tuned to 0.001, the validation loss improved to 0.017 and training object loss improved to 0.040. The model means average precision mAP_0.5 is improved to 0.988 and the precision is improved to 99%. The proposed neural network-based parking system also uses a regularization technique for effective predictive modeling. The proposed modified lasso ridge elastic (LRE) regularization technique provides a 5.21 root mean square error (RMSE) and an R-square of 0.71 with a 4.22 mean absolute error (MAE) indicative of higher accuracy performance compared to other regularization regression models. The advantage of the proposed modified LRE is that it enables effective regularization via modified penalty with the feature selection characteristics of both lasso and ridge.

Feedforward Neural Network (FNN)-based language models estimate the probability of the next word based on the history of the last N words, whereas Recurrent Neural Networks (RNN) perform the same task based only on the last word and some context information that cycles in the network. This paper presents a novel approach, which bridges the gap between these two categories of networks. In particular, we propose an architecture which takes advantage of the explicit, sequential enumeration of the word history in FNN structure while enhancing each word representation at the projection layer through recurrent context information that evolves in the network. The context integration is performed using an additional word-dependent weight matrix that is also learned during the training. Extensive experiments conducted on the Penn Treebank (PTB) and the Large Text Compression Benchmark (LTCB) corpus showed a significant reduction of the perplexity when compared to state-of-the-art feedforward as well as recurrent neural network architectures.

Many important NLP problems can be posed as dual-sequence or sequence-to-sequence modeling tasks. Recent advances in building end-to-end neural architectures have been highly successful in solving such tasks. In this work we propose a new architecture for dual-sequence modeling that is based on associative memory. We derive AM-RNNs, a recurrent associative memory (AM) which augments generic recurrent neural networks (RNN). This architecture is extended to the Dual AM-RNN which operates on two AMs at once. Our models achieve very competitive results on textual entailment. A qualitative analysis demonstrates that long range dependencies between source and target-sequence can be bridged effectively using Dual AM-RNNs. However, an initial experiment on auto-encoding reveals that these benefits are not exploited by the system when learning to solve sequence-to-sequence tasks which indicates that additional supervision or regularization is needed.

How to model a pair of sentences is a critical issue in many NLP tasks such as answer selection (AS), paraphrase identification (PI) and textual entailment (TE). Most prior work (i) deals with one individual task by fine-tuning a specific system; (ii) models each sentence's representation separately, rarely considering the impact of the other sentence; or (iii) relies fully on manually designed, task-specific linguistic features. This work presents a general Attention Based Convolutional Neural Network (ABCNN) for modeling a pair of sentences. We make three contributions. (i) ABCNN can be applied to a wide variety of tasks that require modeling of sentence pairs. (ii) We propose three attention schemes that integrate mutual influence between sentences into CNN; thus, the representation of each sentence takes into consideration its counterpart. These interdependent sentence pair representations are more powerful than isolated sentence representations. (iii) ABCNN achieves state-of-the-art performance on AS, PI and TE tasks.

The brain is one of the most studied and highly complex systems in the biological world. It is the information center behind all vertebrate and most invertebrate life, and thus has become a major focus in current research. While many of these studies have concentrated on studying the brain directly, our focus is the structure of the brain itself: at its core an interconnected network of nodes (neurons). A better understanding of the structural aspects of the brain should elucidate some of its functional properties. In this paper we analyze the brain of the nematode Caenorhabditis elegans. Consisting of only 302 neurons, it is one of the better-understood neural networks. Using a Laplacian matrix of the 279-neuron "giant component" of the network, we use an eigenvalue counting function to look for fractal-like self similarity. This matrix representation is also used to plot (in eigenfunction coordinates) both 2 and 3 dimensional visualizations of the neural network. Further analysis examines the small-world properties of the system, including average path length and clustering coefficient. We then test for localization of eigenfunctions, using graph energy and spacial variance. To better understand these results, all of these calculations are also performed on random networks, branching trees, and known fractals, as well as fractals which have been "rewired" to have small-world properties. This analysis is one of many stepping-stones in the research of neural networks. While many of the structures and functions within the brain are known, understanding how the two interact is also important. A firmer grasp on the structural properties of the neural network is a key step in this process

The tonotopic representation of sounds is a well established organizational principle of the auditory cortex. However, given the small extent of auditory cortical regions, mapping tonotopic representation using noninvasive tools such as M/EEG is challenging. Resolving ongoing brain activity at a tonotopic level has been deemed virtually impossible due to volume conduction. However, based on previous data showing robust carrier-frequency decoding, it is clear tonotopic information is present at a noninvasive level. In order to also eavesdrop on ongoing activity, we propose that backward encoding models can be tuned to specific sound frequency bands applied to other data sets and to model neural activity in resulting “feature” (here: sound frequency band) channels.

The tonotopic representation of sounds is a well established organizational principle of the auditory cortex. However, given the small extent of auditory cortical regions, mapping tonotopic representation using noninvasive tools such as M/EEG is challenging. Resolving ongoing brain activity at a tonotopic level has been deemed virtually impossible due to volume conduction. However, based on previous data showing robust carrier-frequency decoding, it is clear tonotopic information is present at a noninvasive level. In order to also eavesdrop on ongoing activity, we propose that backward encoding models can be tuned to specific sound frequency bands applied to other data sets and to model neural activity in resulting “feature” (here: sound frequency band) channels.

Knowing which words have been attended to in previous time steps while generating a translation is a rich source of information for predicting what words will be attended to in the future. We improve upon the attention model of Bahdanau et al. (2014) by explicitly modeling the relationship between previous and subsequent attention levels for each word using one recurrent network per input word. This architecture easily captures informative features, such as fertility and regularities in relative distortion. In experiments, we show our parameterization of attention improves translation quality.

[1] Introduction: Dried fruits are one of the most important non-oil exports and the efforts should be made to grow the economy of the country by increasing their exports to world markets. Meanwhile, quince juice contains various minerals including iron, phosphorus, calcium, potassium and rich in vitamins such as vitamins A, C and B vitamins. Drying of food is one of the ways to keep its quality and increase its shelflife. During this process, the removal of moisture through the simultaneous transfer of heat and mass occurs. By transferring heat from the environment to the foodstuff, the heat energy evaporates the surface moisture. The drying process has a great impact on the product. In recent years, new and innovative techniques have been considered that increase the drying rate and maintain the quality of the product and infrared drying is one of these novel techniques.. Infrared systems are emitting electromagnetic waves with a wavelength of 700 nm to 1 mm. The advantage of using infrared is to minimize waste and prevent product quality loss due to reduced drying time can be mentioned. The need to predict product quality in each process makes it necesary to model and discover the relationship between factors that can affect the final quality of the product. Artificial neural networks have been considered as a meta-innovative algorithm for modeling and prediction, which can be favored by the ability of these networks to model and predict processes The complexity and discovery of non-random fluctuations in data and the ability to discover the interactions between variables, economical savings in the use and disconnection of classical model abusive constraints (Togrul et al., 2004), the ability to reduce The effect of non-effective variables on the model by setting internal parameters is the ability to predict the desired parameter variations with minimum parameters (Bowers et al., 2000).   Materials and methods: In this research, quince fruit (Variety of Isfahan) was purchased as the premium product of Isfahan Gardens and was kept at 0 ° C in the cold room prior to further experiments. The fruits were removed from the refrigerator one hour before processing and exposed to ambient temperature. After washing, surface moisture was removed by moisture absorbent paper and turned into slices with a constant thickness of 4 mm. The specimens were subjected to pre-treatment with an osmotic solution (vacuum for 70 minutes at a temperature of 40 ° C for 5 hours). For drying the samples, an infrared convective dryer with three voltages (800.400 and 1200 watts) and a constant speed of 0.5 m / s was used. In this way, the samples were placed under infrared lamps on a plate made from a grid and the weight of the samples was measured in a scale of 10 minutes by means of a scale and recorded on the computer. In order to achieve stable conditions in the system, the dryer was switched on 30 minutes before the process. The distance between the samples and the infrared lamp was fixed in all treatments at 16 cm. The drying process continued to reach a moisture content of 0.22 basis. To perform a puncture tests, quince slices were used in a Brookfield-based American LFRA-4500 tissue analysis device. In order to model these parameters in the drying process, the results of examining the quality of the samples, including the firmness of the tissue as well as the drying time, were used as network outputs. The power, concentration and pressure parameters were considered as network inputs. In this research, a multilayer perceptron network (MLP) was used. Due to its simplicity and high precision, this model has a great application in modeling the drying of agricultural products. Many functions in transmitting numbers from the previous layer to the next layer may be used (Tripathy et al., 2008).   Result & discussion: The results indicated that the stiffness of the tissue is reduced in vacuum conditions with increased power. So, the least amount of stiffness was related to osmotic sample dried at 1200 watts. By increasing the infrared power, the stiffness of the tissue decreases, the reason for this is probably the volume increase phenomenon that occurs during the rapid evaporation of moisture through infrared rays from inside the tissue. The results showed that at the start of the drying process, due to the high moisture content of the product, the moisture loss rate is high. Gradually, with the advent of time and reduced initial moisture content, the rate of moisture reduction naturally decreases. At lower power, the drying time is longer and with increasing power, the drying time decreases due to the increase of the thermal gradient inside the product and consequently the increase in the rate of evaporation of the moisture content of the product. The results of this study showed that the neural artificial network, as a powerful tool, can estimate the stiffness parameters of the tissue and the drying time with high precision. The most suitable neural network structure to predict these parameters with a 3-7-2 topology along with logarithmic activation functions with a total explanation coefficient above 0.9923 represent the best results. Also, by increasing the drying capacity and using osmotic dehydration, the drying time and the stiffness of the tissue samples is decreased.

The goal of this work was to analyze the potential of neural networks and fuzzy logic methods to develop approximate response surfaces as process modeling, that is for mapping of input into output. Structural response was chosen as an example. Each of the many methods surveyed are explained and the results are presented. Future research directions are also discussed.

The goal of this work was to analyze the potential of neural networks and fuzzy logic methods to develop approximate response surfaces as process modeling, that is for mapping of input into output. Structural response was chosen as an example. Each of the many methods surveyed are explained and the results are presented. Future research directions are also discussed.

This dissertation research extends the current knowledge of feature saliency in artificial neural networks (ANN). Feature saliency measures allow for the user to rank order the features based upon the saliency, or relative importance, of the features. Selecting a parsimonious set of salient input features is crucial to the success of any ANN model. In this research, several methodologies were developed using the Signal to Noise Ratio (SNR) Feature Screening Method and its associated SNR Saliency Measure for selecting a parsimonious set of salient features to classify pilot workload in addition to air traffic controller workload. Candidate features were derived from electroencephalography (EEG), electrocardiography (EKG), electro-oculography (EOG), and respiratory gauges. In addition, a new saliency measure was developed that can account for time in Elman Recurrent Neural Networks (RNN). This Partial Derivative Based Spatial Temporal Saliency Measure is used via a Spatial Temporal Feature Screening Method for selecting a parsimonious set of salient features in both time and space. Finally, a technique for investigating the memory capacity of an Elman RNN was developed.

This dissertation research extends the current knowledge of feature saliency in artificial neural networks (ANN). Feature saliency measures allow for the user to rank order the features based upon the saliency, or relative importance, of the features. Selecting a parsimonious set of salient input features is crucial to the success of any ANN model. In this research, several methodologies were developed using the Signal to Noise Ratio (SNR) Feature Screening Method and its associated SNR Saliency Measure for selecting a parsimonious set of salient features to classify pilot workload in addition to air traffic controller workload. Candidate features were derived from electroencephalography (EEG), electrocardiography (EKG), electro-oculography (EOG), and respiratory gauges. In addition, a new saliency measure was developed that can account for time in Elman Recurrent Neural Networks (RNN). This Partial Derivative Based Spatial Temporal Saliency Measure is used via a Spatial Temporal Feature Screening Method for selecting a parsimonious set of salient features in both time and space. Finally, a technique for investigating the memory capacity of an Elman RNN was developed.

This thesis evaluates the capabilities of artificial neural networks in forecasting the take-rates of the Voluntary Separations Incentive/Special Separations Benefit (VSI/SSB) programs for male, Marine Corps Enlisted Personnel in the grades of E-5 and E-6. The Artificial Neural Networks models are compared with the forecasting abilities of a classical regression model. The data are taken from the Headquarters Marine Corps Enlisted Master File which contains military and personal background on each enlisted member of the United States Marine Corps. The classical regression model is a casual model constructed based upon the theory of occupational job choice. The neural network models are presented with all available data elements. Empirical results indicate that artificial neural networks provide forecasting results at least as good as, if not better than, those obtained using classical regression techniques. However, artificial neural networks are limited in their usefulness for policy analysis. Neural networks, Modeling techniques, Voluntary separation programs, VSI, SSB, Marine Corps separations incentives.

Large amount of small Unmanned Aerial Vehicles (sUAVs) are projected to operate in the near future. Potential sUAV applications include, but not limited to, search and rescue, inspection and surveillance, aerial photography and video, precision agriculture, and parcel delivery. sUAVs are expected to operate in the uncontrolled Class G airspace, which is at or below 500 feet above ground level (AGL), where many static and dynamic constraints exist, such as ground properties and terrains, restricted areas, various winds, manned helicopters, and conflict avoidance among sUAVs. How to enable safe, efficient, and massive sUAV operations at the low altitude airspace remains a great challenge. NASA's Unmanned aircraft system Traffic Management (UTM) research initiative works on establishing infrastructure and developing policies, requirement, and rules to enable safe and efficient sUAVs' operations. To achieve this goal, it is important to gain insights of future UTM traffic operations through simulations, where the accurate trajectory model plays an extremely important role. On the other hand, like what happens in current aviation development, trajectory modeling should also serve as the foundation for any advanced concepts and tools in UTM. Accurate models of sUAV dynamics and control systems are very important considering the requirement of the meter level precision in UTM operations. The vehicle dynamics are relatively easy to derive and model, however, vehicle control systems remain unknown as they are usually kept by manufactures as a part of intellectual properties. That brings challenges to trajectory modeling for sUAVs. How to model the vehicle's trajectories with unknown control system? This work proposes to use a neural network to model a vehicle's trajectory. The neural network is first trained to learn the vehicle's responses at numerous conditions. Once being fully trained, given current vehicle states, winds, and desired future trajectory, the neural network should be able to predict the vehicle's future states at next time step. A complete 4-D trajectory are then generated step by step using the trained neural network. Experiments in this work show that the neural network can approximate the sUAV's model and predict the trajectory accurately.

Speaker intent detection and semantic slot filling are two critical tasks in spoken language understanding (SLU) for dialogue systems. In this paper, we describe a recurrent neural network (RNN) model that jointly performs intent detection, slot filling, and language modeling. The neural network model keeps updating the intent estimation as word in the transcribed utterance arrives and uses it as contextual features in the joint model. Evaluation of the language model and online SLU model is made on the ATIS benchmarking data set. On language modeling task, our joint model achieves 11.8% relative reduction on perplexity comparing to the independent training language model. On SLU tasks, our joint model outperforms the independent task training model by 22.3% on intent detection error rate, with slight degradation on slot filling F1 score. The joint model also shows advantageous performance in the realistic ASR settings with noisy speech input.

Full-scale flight-test pilot floor vibration is modeled using neural networks and full-scale wind tunnel test data for low speed level flight conditions. Neural network connections between the wind tunnel test data and the three flight test pilot vibration components (vertical, lateral, and longitudinal) are studied. Two full-scale UH-60A Black Hawk databases are used. The first database is the NASA/Army UH-60A Airloads Program flight test database. The second database is the UH-60A rotor-only wind tunnel database that was acquired in the NASA Ames 80- by 120- Foot Wind Tunnel with the Large Rotor Test Apparatus (LRTA). Using neural networks, the flight-test pilot vibration is modeled using the wind tunnel rotating system hub accelerations, and separately, using the hub loads. The results show that the wind tunnel rotating system hub accelerations and the operating parameters can represent the flight test pilot vibration. The six components of the wind tunnel N/rev balance-system hub loads and the operating parameters can also represent the flight test pilot vibration. The present neural network connections can significantly increase the value of wind tunnel testing.

This article is from Frontiers in Computational Neuroscience , volume 8 . Abstract The antennal lobe (AL), olfactory processing center in insects, is able to process stimuli into distinct neural activity patterns, called olfactory neural codes. To model their dynamics we perform multichannel recordings from the projection neurons in the AL driven by different odorants. We then derive a dynamic neuronal network from the electrophysiological data. The network consists of lateral-inhibitory neurons and excitatory neurons (modeled as firing-rate units), and is capable of producing unique olfactory neural codes for the tested odorants. To construct the network, we (1) design a projection, an odor space, for the neural recording from the AL, which discriminates between distinct odorants trajectories (2) characterize scent recognition, i.e., decision-making based on olfactory signals and (3) infer the wiring of the neural circuit, the connectome of the AL. We show that the constructed model is consistent with biological observations, such as contrast enhancement and robustness to noise. The study suggests a data-driven approach to answer a key biological question in identifying how lateral inhibitory neurons can be wired to excitatory neurons to permit robust activity patterns.

Book Source: Digital Library of India Item 2015.225022 dc.contributor.author: Om Prakash dc.date.accessioned: 2015-07-10T15:28:14Z dc.date.available: 2015-07-10T15:28:14Z dc.date.digitalpublicationdate: 2005-09-08 dc.identifier.barcode: 5990010120112 dc.identifier.origpath: /rawdataupload/upload/0120/114 dc.identifier.copyno: 1 dc.identifier.uri: http://www.new.dli.ernet.in/handle/2015/225022 dc.description.scannerno: 15 dc.description.scanningcentre: IIIT, Allahabad dc.description.main: 1 dc.description.tagged: 0 dc.description.totalpages: 66 dc.format.mimetype: application/pdf dc.language.iso: English dc.publisher: Indian Institute Of Technology Kanpur dc.rights: Out_of_copyright dc.source.library: Indian Institute Of Technology Kanpur dc.subject.classification: Technology dc.subject.classification: Engineering. Technology In General dc.subject.classification: Mechanical Engineering In General. Nuclear Technology. Electrical Engineering. Machinery dc.title: Modeling Of Performence Of An Artillery Rocket Using Neural Networks

Neural networks are a very useful tool for analyzing and modeling complex real world systems. Applying neural network simulations to real world problems generally involves large amounts of data and massive amounts of computation. To efficiently handle the computational requirements of large problems, we have implemented at Los Alamos a highly efficient neural network compiler for serial computers, vector computers, vector parallel computers, and fine grain SIMD computers such as the CM-2 connection machine. This paper describes the mapping used by the compiler to implement feed-forward backpropagation neural networks for a SIMD (Single Instruction Multiple Data) architecture parallel computer. Thinking Machines Corporation has benchmarked our code at 1.3 billion interconnects per second (approximately 3 gigaflops) on a 64,000 processor CM-2 connection machine (Singer 1990). This mapping is applicable to other SIMD computers and can be implemented on MIMD computers such as the CM-5 connection machine. Our mapping has virtually no communications overhead with the exception of the communications required for a global summation across the processors (which has a sub-linear runtime growth on the order of O(log(number of processors)). We can efficiently model very large neural networks which have many neurons and interconnects and our mapping can extend to arbitrarily large networks (within memory limitations) by merging the memory space of separate processors with fast adjacent processor interprocessor communications. This paper will consider the simulation of only feed forward neural network although this method is extendable to recurrent networks.

This project utilizes the neurocomputing technology towards modeling semiconductor fabrication processes for which analytical descriptions do not exist. Using data measured on GaAs fabrication lines of microwave circuits, partial fabrication stages as well as the complete process have been modeled. The developed models allow yield estimation and the determination as to which devices/wafers should be continued in the fabrication line. Subsequently, sensitivity analysis can be performed on process input factors to reveal which inputs carry more importance in producing final electronic devices having targeted specifications. The concept of neural network models of fabrication process has also been applied for achieving improved yield of fabricated devices. Process data have been evaluated for principal components and reduced neural network models developed. Perceptron networks have then been inverted and process inputs recentered to maximize the yield. To achieve this, optimization has been performed in the reduced input space. The principal component analysis allows for re-adjustment of actual inputs for maximum yield. The software DESCENT, developed as a part of this project, can be used as a tool for practical design centering for maximum yield. It should be noted that results of modeling and centering, including the DESCENT package, are available to model and improve yield of other fabrication and manufacturing techniques.

With support from the Air Force Office of Scientific Research, we made significant theoretical and empirical advances in our understanding of cognitive control. We discovered new phenomena and developed theories to account for them. We developed theories of cognitive control and visual attention that integrated mathematical psychology with cognitive science and with neuroscience. We published our work in major journals, including Psychological Review, Journal of Neuroscience, and Journal of Experimental Psychology. We trained several graduate students and postdoctoral fellows and strengthened our collaboration with Tom Palmeri.

A dataset of 854 small unmanned aerial system (SUAS) flight experiments from 2005-2009 is analyzed to determine significant factors that contribute to mishaps. The data from 29 airframes of different designs and technology readiness levels were aggregated. 20 measured parameters from each flight experiment are investigated, including wind speed, pilot experience, number of prior flights, pilot currency, etc. Outcomes of failures (loss of flight data) and damage (injury to airframe) are classified by logistic regression modeling and artificial neural network analysis. From the analysis, it can be concluded that SUAS damage is a random event that cannot be predicted with greater accuracy than guessing. Failures can be predicted with greater accuracy (38.5% occurrence, model hit rate 69.6%). Five significant factors were identified by both the neural networks and logistic regression. SUAS prototypes risk failures at six times the odds of their commercially manufactured counterparts. Likewise, manually controlled SUAS have twice the odds of experiencing a failure as those autonomously controlled. Wind speeds, pilot experience, and pilot currency were not found to be statistically significant to flight outcomes. The implications of these results for decision makers, range safety officers and test engineers are discussed.

Talk by Zoran Tiganj of Indiana University.  Given to the Redwood Center for Theoretical Neuroscience at UC Berkeley. Abstract: Building artificial agents that can mimic human learning and reasoning has been a longstanding objective in artificial intelligence. I will discuss some of the empirical data and computational models from neuroscience and cognitive science that could help us advance towards this goal. Specifically, I will talk about the importance of structured representations of knowledge, particularly about mental or cognitive maps for time, space, and concepts. I will present data from recent behavioral and neural studies, which suggest that the brain maintains a scale-invariant mental timeline of the past and uses it to construct a compressed mental timeline of the future. From the computational perspective, these findings illustrate how associative learning can play a role in building structured representations of knowledge. Finally, I will discuss possible strategies to incorporate these findings into building artificial agents, especially in memory-augmented and attention-based neural networks.

Here we demonstrate that the activity of neural ensembles can be quantitatively modeled. We first show that an ensemble dynamical model (EDM) accurately approximates the distribution of voltages and average firing rate per neuron of a population of simulated integrate-and-fire neurons. EDMs are high-dimensional nonlinear dynamical models. To faciliate the estimation of their parameters we present a dimensionality reduction method and study its performance with simulated data. We then introduce and evaluate a maximum-likelihood method to estimate connectivity parameters in networks of EDMS. Finally, we show that this model an methods accurately approximate the high-gamma power evoked by pure tones in the auditory cortex of rodents. Overall, this article demonstrates that quantitatively modeling brain activity at the ensemble level is indeed possible, and opens the way to understanding the computations performed by neural ensembles, which could revolutionarize our understanding of brain function.

The convolutional neural networks (CNNs) have proven to be a powerful tool for discriminative learning. Recently researchers have also started to show interest in the generative aspects of CNNs in order to gain a deeper understanding of what they have learned and how to further improve them. This paper investigates generative modeling of CNNs. The main contributions include: (1) We construct a generative model for the CNN in the form of exponential tilting of a reference distribution. (2) We propose a generative gradient for pre-training CNNs by a non-parametric importance sampling scheme, which is fundamentally different from the commonly used discriminative gradient, and yet has the same computational architecture and cost as the latter. (3) We propose a generative visualization method for the CNNs by sampling from an explicit parametric image distribution. The proposed visualization method can directly draw synthetic samples for any given node in a trained CNN by the Hamiltonian Monte Carlo (HMC) algorithm, without resorting to any extra hold-out images. Experiments on the challenging ImageNet benchmark show that the proposed generative gradient pre-training consistently helps improve the performances of CNNs, and the proposed generative visualization method generates meaningful and varied samples of synthetic images from a large-scale deep CNN.

A practical means of reducing the impact of blast loads on buildings is to introduce a barrier wall between the explosive device and the building. The height and location of the barrier wall are key design variables in terms of effectively reducing the peak positive and negative overpressure and impulse on the building. Until recently, set-ups that included a barrier between the explosive device and the building could only be modeled with consistent accuracy by using numeric simulation techniques. Unfortunately, these models require many hours of processing time to complete a simulation run, even for the fastest of today's computers. This has led several researchers to consider the use of advanced empirical modeling methods, specifically artificial neural networks, to overcome problems of computationally expensive simulations. Neural networks have the potential to make predictions of the influence of a barrier on blast propagation in a matter of seconds using a desktop computer, thus making it easier for designers to home-in on an optimal solution. Artificial neural networks appear to be well suited to this application, performing well for problems that are strongly non-linear and comprise many independent variables. This paper reports on past and on-going research in this field at AFRL Tyndall, using both scaled-live experimental data and simulated data to develop the neural models. The design and validation of these models are presented, and their successes and deficiencies are discussed. The paper concludes with an overview of current and future research plans to take this work to a state suitable for use in the field, and to extend it to problems comprising significantly more complicated configurations of structures than a barrier positioned between the explosive device and a building.

Research works are ongoing in mixing the biologically synthesized oil with the diesel for reducing the effect of global warming and climate change. From the review study, it is noted that the blended biodiesels require more assert about their practical viability. So, the non-edible Tamanu oil is synthesized and it is blended with diesel and its emission characteristics, engine performance and combustion characteristics are studied in our previous work. This paper attempts to model the diesel engine fueled with tamanu oil biodiesel blend. The proposed model exploits the context of neural network and the firefly algorithm is used to train it. After analyzing the various characteristics of the diesel engine, the acquired data is subjected to a proposed FF-NM method. The simulated results are statistically evaluated and the proposed modeling method is proved to be better than the other NM.

Models to predict tree survival and mortality can help to understand vegetation dynamics and to predict effects of climate change on native forests. The objective of the present study was to use Artificial Neural Networks, based on the competition index and climatic and categorical variables, to predict tree survival and mortality in Semideciduous Seasonal Forests in the Atlantic Forest biome. Numerical and categorical trees variables, in permanent plots, were used. The Agricultural Reference Index for Drought (ARID) and the distance-dependent competition index were the variables used. The overall efficiency of classification by ANNs was higher than 92% and 93% in the training and test, respectively. The accuracy for classification and number of surviving trees was above 99% in the test and in training for all ANNs. The classification accuracy of the number of dead trees was low. The mortality accuracy rate (10.96% for training and 13.76% for the test) was higher with the ANN 4, which considers the climatic variable and the competition index. The individual tree-level model integrates dendrometric and meteorological variables, representing a new step for modeling tree survival in the Atlantic Forest biome.

Models to predict tree survival and mortality can help to understand vegetation dynamics and to predict effects of climate change on native forests. The objective of the present study was to use Artificial Neural Networks, based on the competition index and climatic and categorical variables, to predict tree survival and mortality in Semideciduous Seasonal Forests in the Atlantic Forest biome. Numerical and categorical trees variables, in permanent plots, were used. The Agricultural Reference Index for Drought (ARID) and the distance-dependent competition index were the variables used. The overall efficiency of classification by ANNs was higher than 92% and 93% in the training and test, respectively. The accuracy for classification and number of surviving trees was above 99% in the test and in training for all ANNs. The classification accuracy of the number of dead trees was low. The mortality accuracy rate (10.96% for training and 13.76% for the test) was higher with the ANN 4, which considers the climatic variable and the competition index. The individual tree-level model integrates dendrometric and meteorological variables, representing a new step for modeling tree survival in the Atlantic Forest biome.

Microelectromechanical systems (MEMS) are a broad and rapidly expanding field that is currently receiving a great deal of attention because of the potential to significantly improve the ability to sense, analyze, and control a variety of processes, such as heating and ventilation systems, automobiles, medicine, aeronautical flight, military surveillance, weather forecasting, and space exploration. MEMS are very small and are a blend of electrical and mechanical components, with electrical and mechanical systems on one chip. This research establishes reliability estimation and prediction for MEMS devices at the conceptual design phase using neural networks. At the conceptual design phase, before devices are built and tested, traditional methods of quantifying reliability are inadequate because the device is not in existence and cannot be tested to establish the reliability distributions. A novel approach using neural networks is created to predict the overall reliability of a MEMS device based on its components and each component's attributes. The methodology begins with collecting attribute data (fabrication process, physical specifications, operating environment, property characteristics, packaging, etc.) and reliability data for many types of microengines. The data are partitioned into training data (the majority) and validation data (the remainder). A neural network is applied to the training data (both attribute and reliability); the attributes become the system inputs and reliability data (cycles to failure), the system output. After the neural network is trained with sufficient data. the validation data are used to verify the neural networks provided accurate reliability estimates. Now, the reliability of a new proposed MEMS device can be estimated by using the appropriate trained neural networks developed in this work.

Field-programmable gate array (FPGA) is a prominent device in developing the internet of things (IoT) application since it offers parallel computation, power efficiency, and scalability. The identification and authentication of these FPGAbased IoT applications are crucial to secure the user-sensitive data transmitted over IoT networks. Physical unclonable function (PUF) technology provides a great capability to be used as device identification and authentication for FPGAbased IoT applications. Nevertheless, conventional PUF-based authentication suffers a huge overhead in storing the challenge-response pairs (CRPs) in the verifier’s database. Therefore, in this paper, the FPGA implementation of the Arbiter-PUF model using an artificial neural network (ANN) is presented. The PUF model can generate the CRPs on-the-fly upon the authentication request (i.e., by a prover) to the verifier and eliminates huge storage of CRPs database in the verifier. The architecture of ANN (i.e., Arbiter-PUF model) is designed in Xilinx system generator and subsequently converted into intellectual property (IP). Further, the IP is programmed in Xilinx Artix-7 FPGA with other peripherals for CRPs generation and validation. The findings show that the Arbiter-PUF model implementation on FPGA using the ANN technique achieves approximately 98% accuracy. The model consumes 12,196 look-up tables (LUTs) and 67 mW power in FPGA.

The purposes of this research were: (1) validating Kim's (2007) simulation method by applying analytic methods and (2) comparing the two different Robust Parameter Design methods with three measures of performance (label accuracy for enemy, friendly, and clutter). Considering the features of CID, input variables were defined as two controllable (threshold combination of detector and classifier) and three uncontrollable (map size, number of enemies and friendly). The first set of experiments considers Kim's method using analytical methods. In order to create response variables, Kim's method uses Monte Carlo simulation. The output results showed no difference between simulation and the analytic method. The second set of experiments compared the measures of performance between a standard RPD used by Kim and a new method using Artificial Neural Networks (ANNs). To find optimal combinations of detection and classification thresholds, Kim's model uses regression with a combined array design, whereas the ANNs method uses ANN with a crossed array design. In the case of label accuracy for enemy, Kim's solution showed the higher expected value, however it also showed a higher variance. Additionally, the model's residuals were higher for Kim's model.

Trajectory reconstruction of pedestrian is of paramount importance to understand crowd dynamics and human movement pattern, which will provide insights to improve building design, facility management and route planning. Camera-based tracking methods have been widely explored with the rapid development of deep learning techniques. When moving to indoor environment, many challenges occur, including occlusions, complex environments and limited camera placement and coverage. Therefore, we propose a novel indoor trajectory reconstruction method using building information modeling (BIM) and graph neural network (GNN). A spatial graph representation is proposed for indoor environment to capture the spatial relationships of indoor areas and monitoring points. Closed circuit television (CCTV) system is integrated with BIM model through camera registration. Pedestrian simulation is conducted based on the BIM model to simulate the pedestrian movement in the considered indoor environment. The simulation results are embedded into the spatial graph for training of GNN. The indoor trajectory reconstruction is implemented as GNN conducts edge classification on the spatial graph

The overall objective of this ongoing effort is to provide the capability to model and simulate rotorcraft aeromechanics behaviors in real-time. This would be accomplished by the addition of an aeromechanics element to an existing high-fidelity, real-time helicopter flight simulation. As a first step, the peak vertical vibration at the pilot floor location was considered in this neural-network-based study. The flight conditions considered were level flights, rolls, pushovers, pull-ups, autorotations, and landing flares. The NASA/Army UH-60A Airloads Program flight test database was the source of raw data. The present neural network training databases were created in a physically consistent manner. Two modeling approaches, with different physical assumptions, were considered. The first approach involved a maneuver load factor that was derived using the roll-angle and the pitch-rate. The second approach involved the three pilot control stick positions. The resulting, trained back-propagation neural networks were small, implying rapid execution. The present neural-network-based approach involving the peak pilot vibration was utilized in a quasi-static manner to simulate an extreme, time-varying pull-up maneuver. For the above pull-up maneuver, the maneuver load factor approach was better for real-time simulation, i.e., produced greater fidelity, as compared to the control stick positions approach. Thus, neural networks show promise for use in high-fidelity, real-time modeling of rotorcraft vibration.

Neural network relationships between the full-scale, flight test hub accelerations and the corresponding three N/rev pilot floor vibration components (vertical, lateral, and longitudinal) are studied. The present quantitative effort on the UH-60A Black Hawk hub accelerations considers the lateral and longitudinal vibrations. An earlier study had considered the vertical vibration. The NASA/Army UH-60A Airloads Program flight test database is used. A physics based maneuver-effect- factor (MEF), derived using the roll-angle and the pitch-rate, is used. Fundamentally, the lateral vibration data show high vibration levels (up to 0.3 g's) at low airspeeds (for example, during landing flares) and at high airspeeds (for example, during turns). The results show that the advance ratio and the gross weight together can predict the vertical and the longitudinal vibration. However, the advance ratio and the gross weight together cannot predict the lateral vibration. The hub accelerations and the advance ratio can be used to satisfactorily predict the vertical, lateral, and longitudinal vibration. The present study shows that neural network based representations of all three UH-60A pilot floor vibration components (vertical, lateral, and longitudinal) can be obtained using the hub accelerations along with the gross weight and the advance ratio. The hub accelerations are clearly a factor in determining the pilot vibration. The present conclusions potentially allow for the identification of neural network relationships between the experimental hub accelerations obtained from wind tunnel testing and the experimental pilot vibration data obtained from flight testing. A successful establishment of the above neural network based link between the wind tunnel hub accelerations and the flight test vibration data can increase the value of wind tunnel testing.

Statistical Analysis Plan (SAP) As part of the larger TIP project, 61 SAD patients and 41 healthy controls underwent an emotion regulation task with negative and neutral faces during fMRI scanning. We will use dynamic causal modeling (DCM) to shed light on potential disturbances in the effective connectivity of emotion regulation networks in social anxiety disorder (SAD).

This work presents an original and innovative approach by combining greenhouse cooling with artificial intelligence models to ensure food security. The study is divided into two parts: Experimental and theoretical. 

Electricity consumption is increasing gradually and this trend will continue in the future. In addition, rapid network control systems using the resources offered by power electronics and control microelectronics have been recently studied and developed, and are currently in normal application for some, for others, in pilot applications or as prototypes. This paper attempts to show that these systems are referred to by the general acronym flexible alternative current transmission systems (FACTS) similarly dethroned the traditional systems while offering better solutions and solving the energy quality problem such as the hybrid system (unified power flow controller (UPFC), or universal phase shifter regulator (UPSR)) which opens up new perspectives for more efficient operation of networks by continuous and rapid action on the various parameters of the network (voltage, phase shift, and impedance); thus, the power transits will be better controlled and the voltages better held, which will make it possible to increase the stability margins or tend towards the thermal limits of the lines. In this work, we used a classic control (PI-decoupled) and others while offering more flexibility of control thanks to the development of strategies identification/control based on generalized predictive control (GPC) with neural network to ensure robust control with advanced algorithms.

Click here to view the University of Florida catalog record

Click here to view the University of Florida catalog record

We developed physiologically relevant, neural networks to model time-varying neuronal population operations in the motor cortex and spinal cord, dealing with movements in space. We also developed a model of the interactions between these two networks dealing with generating time-varying motoneuronal outputs for movements in space. The novelty of our approach consisted in (a) the realistic nature of the elements in our networks, (b) the massive and asymmetric interconnectivity among network elements, (c) the physiologically relevant design of the networks, including the communication by spike trains among network elements and rules of connectivity based on experimental findings, (d) the dynamical behavior of the networks, and (e) the time-varying performance of the networks. Finally, we were able to reliably decode and transform the neuronal ensemble activity recorded in behaving animals for controlling an simulated arm. This demonstration suggests that the use of biologically inspired neural networks to transform raw cortical signals into the motor output of a multijoint artificial limb is both feasible and practical time-varying performance of the networks.