Na área de simulação de processos, existe uma visível tendência da migração das ferramentas seqüenciais modulares, que hoje são as mais amplamente utilizadas, para as baseadas em equações. Uma das principais vantagens do paradigma baseado em equações ou simultâneo é que este se mostra eficiente na solução de problemas de simulação, otimização, estimação de parâmetros e reconciliação de dados, todos baseados em um mesmo conjunto de modelos, evitando retrabalho de modelagem. Porém, a tecnologia simultânea também apresenta algumas deficiências, onde destacam-se os problemas de robustez tanto na modelagem quanto na obtenção de resultados numéricos. Este trabalho tem como objetivo reunir e desenvolver técnicas que permitam reduzir estas deficiências. Para tanto, as técnicas conhecidas para depuração de sistemas de equações que representam problemas estacionários e dinâmicos foram estudadas em detalhe. Pôde-se observar que para o caso estático os métodos disponíveis para depuração de modelos, já se apresentam em um nível bem desenvolvido. Já para o caso dinâmico, onde há uma maior complexidade, as técnicas conhecidas encontram-se em um nível de desenvolvimento muito menor. Neste ponto encontram-se as principais contribuições deste trabalho. / In the field of process simulation the movement from the sequential modular tools, which are currently the most widely used, to the equation based approach is clear. One of the key advantages of the equation based or simulatneous approach is that using a single model one can solve simulation, optimization, parameter estimation, and optimization problems. This fact avoids modeling rework for each application. However, the simultaneous technology has problems regarding modeling and solving robustness. This work aims to group and develop methods capable of minimize these deficiencies. In order to achieve this goal, available debugging approaches for both steady-state and dynamic system of equations were studied in detail. For the steady-state case well stablished debugging techniques are known. For dynamic models, where the complexity is higher, the analysis and debugging methods are much less mature. This was the source for the major contributions of this work.

Na área de simulação de processos, existe uma visível tendência da migração das ferramentas seqüenciais modulares, que hoje são as mais amplamente utilizadas, para as baseadas em equações. Uma das principais vantagens do paradigma baseado em equações ou simultâneo é que este se mostra eficiente na solução de problemas de simulação, otimização, estimação de parâmetros e reconciliação de dados, todos baseados em um mesmo conjunto de modelos, evitando retrabalho de modelagem. Porém, a tecnologia simultânea também apresenta algumas deficiências, onde destacam-se os problemas de robustez tanto na modelagem quanto na obtenção de resultados numéricos. Este trabalho tem como objetivo reunir e desenvolver técnicas que permitam reduzir estas deficiências. Para tanto, as técnicas conhecidas para depuração de sistemas de equações que representam problemas estacionários e dinâmicos foram estudadas em detalhe. Pôde-se observar que para o caso estático os métodos disponíveis para depuração de modelos, já se apresentam em um nível bem desenvolvido. Já para o caso dinâmico, onde há uma maior complexidade, as técnicas conhecidas encontram-se em um nível de desenvolvimento muito menor. Neste ponto encontram-se as principais contribuições deste trabalho. / In the field of process simulation the movement from the sequential modular tools, which are currently the most widely used, to the equation based approach is clear. One of the key advantages of the equation based or simulatneous approach is that using a single model one can solve simulation, optimization, parameter estimation, and optimization problems. This fact avoids modeling rework for each application. However, the simultaneous technology has problems regarding modeling and solving robustness. This work aims to group and develop methods capable of minimize these deficiencies. In order to achieve this goal, available debugging approaches for both steady-state and dynamic system of equations were studied in detail. For the steady-state case well stablished debugging techniques are known. For dynamic models, where the complexity is higher, the analysis and debugging methods are much less mature. This was the source for the major contributions of this work.

Na área de simulação de processos, existe uma visível tendência da migração das ferramentas seqüenciais modulares, que hoje são as mais amplamente utilizadas, para as baseadas em equações. Uma das principais vantagens do paradigma baseado em equações ou simultâneo é que este se mostra eficiente na solução de problemas de simulação, otimização, estimação de parâmetros e reconciliação de dados, todos baseados em um mesmo conjunto de modelos, evitando retrabalho de modelagem. Porém, a tecnologia simultânea também apresenta algumas deficiências, onde destacam-se os problemas de robustez tanto na modelagem quanto na obtenção de resultados numéricos. Este trabalho tem como objetivo reunir e desenvolver técnicas que permitam reduzir estas deficiências. Para tanto, as técnicas conhecidas para depuração de sistemas de equações que representam problemas estacionários e dinâmicos foram estudadas em detalhe. Pôde-se observar que para o caso estático os métodos disponíveis para depuração de modelos, já se apresentam em um nível bem desenvolvido. Já para o caso dinâmico, onde há uma maior complexidade, as técnicas conhecidas encontram-se em um nível de desenvolvimento muito menor. Neste ponto encontram-se as principais contribuições deste trabalho. / In the field of process simulation the movement from the sequential modular tools, which are currently the most widely used, to the equation based approach is clear. One of the key advantages of the equation based or simulatneous approach is that using a single model one can solve simulation, optimization, parameter estimation, and optimization problems. This fact avoids modeling rework for each application. However, the simultaneous technology has problems regarding modeling and solving robustness. This work aims to group and develop methods capable of minimize these deficiencies. In order to achieve this goal, available debugging approaches for both steady-state and dynamic system of equations were studied in detail. For the steady-state case well stablished debugging techniques are known. For dynamic models, where the complexity is higher, the analysis and debugging methods are much less mature. This was the source for the major contributions of this work.

No description available.

Yes / Weight initialization of neural networks has an important influence on the learning process, and the selection of initial weights is related to the activation interval of the activation function. It is proposed that an improved and extended weight initialization method for neural network with asymmetric activation function as an extension of the linear interval tolerance method (LIT), called ‘GLIT’ (generalized LIT), which is more suitable for higher-dimensional inputs. The purpose is to expand the selection range of the activation function so that the input falls in the unsaturated region, so as to improve the performance of the network. Then, a tolerance solution theorem based upon neural network system is given and proved. Furthermore, the algorithm is given about determining the initial weight interval. The validity of the theorem and algorithm is verified by numerical experiments. The input could fall into any preset interval in the sense of probability under the GLIT method. In another sense, the GLIT method could provide a theoretical basis for the further study of neural networks. / The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This paper is partly supported by National Science Foundation of China under Grants (62073226, 61603262), Liaoning Province Natural Science Foundation (2020-KF-11-09, 2021-KF-11-05), Shen-Fu Demonstration Zone Science and Technology Plan Project (2020JH13, 2021JH07), Central Government Guides Local Science and Technology Development Funds of Liaoning Province (2021JH6).

Approved for public release; distribution in unlimited. / While a necessity of all operating systems, the code that initializes a system can be notoriously difficult to understand. This thesis explores the most common architectures used for bringing an operating system to its initial state, once the operating system gains control from the boot loader. Specifically, the ways in which the OpenBSD and Linux operating systems handle initialization are dissected. With this understanding, a set of threats relevant to the initialization sequence was developed. A thorough study was also made to determine the degree to which initialization code adheres to widely accepted software engineering principles. Based upon this threat analysis and the observed strengths and weaknesses of existing systems, a set of recommendations for initialization sequence architecture and implementation have been developed. These recommendations can serve as a guide for future operating system development. / Civilian, Naval Postgraduate School

WebAssembly (Wasm) has emerged as a new technology for the web that enables complex and interactive web applications, while utilizing a compact and platform-independent bytecode format. Due to its flexibility, portability, and built-in security, it has since evolved to be used in many other embeddings, such as internet-of-things, server applications, and even mobile applications. While a goal of Wasm is near-native performance, research has found that its performance is not as great as initially expected. Due to this, projects like The WebAssembly Pre-Initializer (Wizer) have emerged as potential solutions to this problem. Wizer is a tool developed by the coalition Bytecode Alliance with the purpose of speeding up the startup time, or Critical Path to Interactive (CPTI), of a Wasm module by pre-initializing it and saving a snapshot of the Wasm instance state into a new Wasm module. Wizer has been evaluated using two benchmark programs. However, no larger-scale investigation into the CPTI improvement brought by its pre-initialization has been conducted. Furthermore, saving a snapshot of the module is likely to result in a larger module in terms of file size, leading to increased compile time or, for use cases where it is relevant, network latency. This project investigates, mainly within the field of Wasm in non-web environments, the extent to which Wizer is able to improve CPTI for a Wasm module. The purpose of this is to allow both Wasm maintainers and developers to form an opinion whether pre-initialization could be standardized for use in Wasm compilers and toolchains, or whether pre-initialization should be applied to their Wasm module based on its CPTI before pre-initialization. Results are obtained by compiling a number of sample software down to Wasm, measuring their CPTI in terms of elapsed CPU cycles both without and with pre-initialization using Wizer, and comparing them. This is made possible through an extension to the Sightglass benchmarking framework also developed by Bytecode Alliance. The results show that pre-initialization using Wizer increases the CPTI if the Wasm module cannot be compiled to native CPU instructions in advance. However, if compilation can be done in advance, Wizer is able to reduce the CPTI of a Wasm module by a factor of between two to six times, depending on how it is initialized. / WebAssembly (Wasm) har framträtt som en ny teknik för webben som möjligör komplexa och interaktiva webbapplikationer, genom ett kompakt och platformsoberoende bytekodformat. Tack vare teknikens flexibilitet, portabilitet och inbyggd säkerhet, har den även utvecklats till att användas i andra samanhang, exampelvis i sakernas internet, serverapplikationer, och även mobilapplikationer. Trots att ett mål med Wasm är prestanda jämförbar med native applikationer, har forskning funnit att den inte presterat så väl som man tidigare trott. Därför har project som The WebAssembly Pre-Initializer (Wizer) framträtt som möjliga lösningar till detta problem. Wizer är ett verktyg utvecklat av koalitionen Bytecode Alliance med syftet att snabba upp uppstartstider, även kallat Critical Path to Interactive (CPTI), av en Wasm modul genom att förinitialisera den och spara en ögonblicksbild av Wasm instansens tillstånd som en ny Wasm modul. Wizer har evaluerats genom två testprogram. Dock har inte någon storskalig undersökning utförts inom CPTI-förbättringen som dess förinitialisering kan medföra. Dessutom är det sannorlikt att sparandet av en ögonblicksbild av en modul leder till en filstorleksmässigt större modul, vilket gör att kompileringstiden, och även nätverkslatensen i användningsfall där det förekommer, kan öka. Det här projektet undersöker, huvudsakligen inom Wasm utanför webbläsaren, omfattningen av Wizers förbättring av CPTI för en Wasm modul. Syftet med detta är att möjliggöra för Wasm designers och utvecklare att förstå hurvida förinitialisering skulle kunna bli standardiserat i Wasm kompilerare eller verktyg, eller om förinitialisering borde tillämpas på en Wasm modul under utveckling baserat på dens CPTI innan förinitialisering. Resultat samlas genom att kompilera flera exempelprogram ner till Wasm, mäta deras CPTI genom passerade CPU cykler både med och utan förinitialisering med Wizer, och jämföra mätningarna. Detta möjligörs genom en utökning av testramverket Sightglass som också utvecklats av Bytecode Alliance. Resultaten visar att förinitialisering med Wizer ökar CPTI om Wasm modulen inte kan i förväg kompileras till instruktioner som kan köras på CPU:n. Om kompilering i förväg dock är möjlig kan Wizer minska CPTI för en given Wasm modul med en faktor av två upp till sex gånger, beroende på den typ av förinitialisering som den gör.

The Swedish Armed Forces setup of the Weather Research and Forecasting Model (WRF) has problems to forecast low clouds in stably stratified conditions when the ground is covered by snow. The aim of this thesis is to understand what causes this deficit. Simulations during January and February 2018 are here compared with observations from Sodankylä in northern Finland. It is revealed that neither type of planetary boundary layer parameterization chosen nor vertical or horizontal interpolation are responsible for the deficiency. Instead, our experiments show that, to first order, poor initialization of Stratocumulus (Sc) clouds from the host model, Atmospheric Model High Resolution (HRES), of the Integrated Forecast System (IFS) is the missing link. In situations when Sc clouds are missing in the IFS analysis, although they exist in reality, we use information from vertical soundings from Sodankylä. In the initialization process we used the fact that liquid potential temperature is constant in a well-mixed cloud. Initializing cloud water and cloud ice from IFS HRES and from soundings with different methods improves the model performance and the formation of very low artificial clouds at the first model level is prohibited.

The goal of this work is to improve the robustness and generalization of deep learning models, using a similar approach to the unsupervised "innate learning" strategy in visual development. A series of research studies are presented to demonstrate how an unsupervised machine learning efficient coding approach can create filters similar to the receptive fields of the primary visual cortex (V1) in the brain, and these filters are capable of pretraining convolutional neural networks (CNNs) to enable faster training times and higher accuracy with less dependency on the source data. Independent component analysis (ICA) is used for unsupervised feature extraction as it has shown success in both applied machine learning and modeling biological neural receptive fields. This pretraining applies equally well to various forms of visual input, including natural color images, black and white, binocular, and video to drive the V1-like Gabor filters in the brain. For efficient processing of typical visual scenes, the filters that ICA produces are developed by encoding natural images. These filters are then used to initialize the kernels in the first layer of a CNN to train on the CIFAR-10 dataset to perform image classification. Results show that the ICA initialization for a custom made CNN produces models with a test accuracy up to 12% better than the standard model in the first 10 epochs, which for specific accuracy thresholds reduces the number of training epochs by approximately 40% (to reach 60% accuracy) and 50% (to reach 70% accuracy). Additionally, this pre-training results in marginally higher accuracy even after extensive training over 50 epochs. This proposed method of unsupervised machine learning to pre-train weights in deep learning improves both training time and accuracy, which is why it is observed in biological networks and is finding increased application in applied deep learning.

Learning with neural networks depends on the particular parametrization of the functions represented by the network, that is, the assignment of parameters to functions. It also depends on the identity of the functions, which get assigned typical parameters at initialization, and, later, the parameters that arise during training. The choice of the activation function is a critical aspect of the network design that influences these function properties and requires investigation. This thesis focuses on analyzing the expected behavior of networks with maxout (multi-argument) activation functions. On top of enhancing the practical applicability of maxout networks, these findings add to the theoretical exploration of activation functions beyond the common choices. We believe this work can advance the study of activation functions and complicated neural network architectures. We begin by taking the number of activation regions as a complexity measure and showing that the practical complexity of deep networks with maxout activation functions is often far from the theoretical maximum. This analysis extends the previous results that were valid for deep neural networks with single-argument activation functions such as ReLU. Additionally, we demonstrate that a similar phenomenon occurs when considering the decision boundaries in classification tasks. We also show that the parameter space has a multitude of full-dimensional regions with widely different complexity and obtain nontrivial lower bounds on the expected complexity. Finally, we investigate different parameter initialization procedures and show that they can increase the speed of the gradient descent convergence in training. Further, continuing the investigation of the expected behavior, we study the gradients of a maxout network with respect to inputs and parameters and obtain bounds for the moments depending on the architecture and the parameter distribution. We observe that the distribution of the input-output Jacobian depends on the input, which complicates a stable parameter initialization. Based on the moments of the gradients, we formulate parameter initialization strategies that avoid vanishing and exploding gradients in wide networks. Experiments with deep fully-connected and convolutional networks show that this strategy improves SGD and Adam training of deep maxout networks. In addition, we obtain refined bounds on the expected number of linear regions, results on the expected curve length distortion, and results on the NTK. As the result of the research in this thesis, we develop multiple experiments and helpful components and make the code for them publicly available.