Spelling suggestions: "subject:"higherorder"" "subject:"highorder""
181 |
Living on the slippery slope : the nature, sources and logic of vaguenessZardini, Elia January 2008 (has links)
According to the dominant approach in the theory of vagueness, the nature of the vagueness of an expression ‘F’ consists in its presenting borderline cases in an appropriately ordered series: objects which are neither definitely F nor definitely not F (where the notion of definiteness can be semantic, ontic, epistemic, psychological or primitive). In view of the various problems faced by theories of vagueness adopting the dominant approach, the thesis proposes to reconsider the naive theory of vagueness, according to which the nature of the vagueness of an expression consists in its not drawing boundaries between any neighbouring objects in an appropriately ordered series. It is argued that expressions and concepts which do present this feature play an essential role in our cognitive and practical life, allowing us to conceptualize—in a way which would otherwise be impossible—the typically coarse-grained distinctions we encounter in reality. Despite its strong initial plausibility and ability to explain many phenomena of vagueness, the naive theory is widely rejected because thought to be shown inconsistent by the sorites paradox. In reply, it is first argued that accounts of vagueness based on the dominant approach are themselves subject to higher-order sorites paradoxes. The paradox is then solved on behalf of the naive theory by rejecting the unrestricted transitivity of the consequence relation on a vague language; a family of logics apt for reasoning with vague expressions is proposed and studied (using models with partially ordered values). The characteristic philosophical and logical consequences of this novel solution are developed and defended in detail. In particular, it is shown how the analysis of what happens in the attempt of surveying a sorites series and deciding each case allows the naive theory to recover a "thin" notion of a borderline case.
|
182 |
高階動差對投資組合之影響黃奕栩, Huang, I Hsu Unknown Date (has links)
自Markowitz(1952)提出平均數-變異數準則以來,對於該準則適宜性的討論即不曾停止過。許多實證上資料顯示資產報酬率分配不為常態,而越來越多學者也對於高於二階以上之高階動差對投資決策之影響提出證實。本文利用臺灣八大類股指數報酬率分配資料,運用多目標規劃求解法進行實證,發現臺灣股票市場呈現顯著峰態性質,此外,本文樣本外試驗結果亦指出,平均數-變異數-偏態-峰態架構下之最適投資組合的報酬率高於傳統平均數-變異數架構下之最適投資組合以及大盤報酬。
|
183 |
Saturation methods for global model-checking pushdown systemsHague, Matthew January 2009 (has links)
Pushdown systems equip a finite state system with an unbounded stack memory, and are thus infinite state. By recording the call history on the stack, these systems provide a natural model for recursive procedure calls. Model-checking for pushdown systems has been well-studied. Tools implementing pushdown model-checking (e.g. Moped) are an essential back-end component of high-profile software model checkers such as SLAM, Blast and Terminator. Higher-order pushdown systems define a more complex memory structure: a higher-order stack is a stack of lower-order stacks. These systems form a robust hierarchy closely related to the Caucal hierarchy and higher-order recursion schemes. This latter connection demonstrates their importance as models for programs with higher-order functions. We study the global model-checking problem for (higher-order) pushdown systems. In particular, we present a new algorithm for computing the winning regions of a parity game played over an order-1 pushdown system. We then show how to compute the winning regions of two-player reachability games over order-n pushdown systems. These algorithms extend the saturation methods of Bouajjani, Esparza and Maler for order-1 pushdown systems, and Bouajjani and Meyer for higher-order pushdown systems with a single control state. These techniques begin with an automaton recognising (higher-order) stacks, and iteratively add new transitions until the automaton becomes saturated. The reachability result, presented at FoSSaCS 2007 and in the LMCS journal, is the main contribution of the thesis. We break the saturation paradigm by adding new states to the automaton during the iteration. We identify the fixed points required for termination by tracking the updates that are applied, rather than by observing the transition structure. We give a number of applications of this result to LTL model-checking, branching-time model-checking, non-emptiness of higher-order pushdown automata and Büchi games. Our second major contribution is the first application of the saturation technique to parity games. We begin with a mu-calculus characterisation of the winning region. This formula alternates greatest and least fixed point operators over a kind of reachability formula. Hence, we can use a version of our reachability algorithm, and modifications of the Büchi techniques, to compute the required result. The main advantages of this approach compared to existing techniques due to Cachat, Serre and Vardi et al. are that it is direct and that it is not immediately exponential in the number of control states, although the worst-case complexity remains the same.
|
184 |
The development of self-interference of split HOLZ (SIS-HOLZ) lines for measuring z-dependent atomic displacement in crystalsNorouzpour, Mana 01 May 2017 (has links)
Measuring atomic displacement inside crystals has been an important field of interest for decades especially in semiconductor industry for its effect on the crystal structure and symmetry, subsequently on the bandgap structure. There are three different image based, diffraction based, and electron holography based techniques using transmission electron microscope (TEM). These methods enable measuring atomic displacement inside specimen. However, among all TEM techniques offering nano-scale resolution measurements, convergent beam electron diffraction (CBED) patterns show the highest sensitivity to the atomic displacement. Higher order Laue zone (HOLZ) lines split by small variations of lattice constant allowing the atomic displacement measurement through the crystal. However it is a cumbersome measurement and it can only reveal the atomic displacement in two dimensions. Therefore, the atomic displacement information at each depth through the specimen thickness is still missing. This information can be obtained by recovering the phase information across the split HOLZ line. The phase profile across the split HOLZ line can be retrieved by the electron interferometry method. The phase of the diffracted beam is the required information to reconstruct the atomic displacement profile through the specimen thickness. In this work, we first propose a novel technique of self-interference of split HOLZ line based on the diffracted beam interferometry which recovers the phase information across the split HOLZ line. The experimental details of the technique have been examined to report the parameters in order to implement the method. Regarding the novelty of the technique and the lack of the of a reference phase profile to discuss the results, phase profile simulation was a main contribution. For simulating the phase profile across the split HOLZ line the Howie-Whelan formula supporting the kinematical theory of diffraction is used. Accordingly, the analytical approach to simulate the phase profiles across the split HOLZ line for three various suggested atomic displacements are studied. Also, the effect of some parameters such as the atomic displacement amplitude, the specimen thickness, and the g reflection is investigated on the phase profile. This study leads to an equation used for fitting the experimental results with the simulated phase profile. Consequently, self-interference of split HOLZ line (SIS-HOLZ) is studied as a method of reconstructing the phase profile across the split HOLZ line which carries the information of atomic displacement through the specimen thickness. / Graduate / 0548 / 0794 / mananrp@uvic.ca
|
185 |
Caractérisation des structures de septines hautement organisées chez la drosophile et leur interaction avec le cytosquelette d’actineDragieva, Zlatina 06 1900 (has links)
Les septines sont des GTPases conservées dérégulées dans le cancer et les maladies neurodégénératives. Elles servent de protéines d’échafaudage et forment une barrière de diffusion à la membrane plasmique et au corps central lors de la cytokinèse. Elles interagissent avec l’actine et s’organisent en complexes qui polymérisent et forment des structures hautement organisées (anneaux et filaments). Leur dynamique d’assemblage et leur rôle dans la cellule restent à être élucidés. La Drosophile est un modèle simple pour l’étude des septines puisqu’on n’y retrouve que 5 gènes (sep1, sep2, sep4, sep5, peanut) comparativement aux 13 gènes chez l’humain.
À l’aide d’un anticorps contre Pnut, nous avons identifié des structures tubulaires dans 30% des cellules S2 de Drosophile. Mon projet a comme but de caractériser ces tubes en élucidant leurs constituants, leur comportement et leurs propriétés pour mieux clarifier le mécanisme par lequel les septines forment des structures hautement organisées et interagissent avec le cytosquelette d’actine.
Par immunofluorescence, j’ai pu démontrer que ces tubes sont cytoplasmiques, en mitose ou interphase, ce qui suggère qu’ils ne sont pas régulés par le cycle cellulaire. Pour investiguer la composition et les propriétés dynamiques de ces tubes, j’ai généré une lignée cellulaire exprimant Sep2-GFP qui se localise aux tubes et des ARNi contre les cinq septines. Trois septines sont importantes pour la formation de ces tubes et anneaux notamment Sep1, Sep2 et Pnut. La déplétion de Sep1 cause la dispersion du signal GFP en flocons, tandis que la déplétion de Sep2 ou de Pnut mène à la dispersion du signal GFP uniformément dans la cellule. Des expériences de FRAP sur la lignée Sep2-GFP révèlent un signal de retour très lent, ce qui indique que ces structures sont très stables.
J’ai aussi démontré une relation entre l’actine et les septines. Le traitement avec la Latrunculin A (un inhibiteur de la polymérisation de l’actine) ou la Jasplakinolide (un stabilisateur des filaments d’actine) mène à la dépolymérisation rapide (< 30 min) des tubes en anneaux flottants dans le cytoplasme, même si ces tubes ne sont pas reconnus suite à un marquage de la F-actine. L’Actin05C-mCherry se localise aux tubes, tandis que le mutant déficient de la polymérisation, Actin05C-R62D-mCherry perd cette localisation. On observe aussi que la déplétion de la Cofiline et de l’AIP1 (ce qui déstabilise l’actine) mène au même phénotype que le traitement avec la Latrunculine A ou la Jasplakinolide. Alors on peut conclure qu’un cytosquelette d’actine dynamique est nécessaire pour la formation et le maintien des tubes de septines.
Les futures études auront comme but de mieux comprendre l’organisation des septines en structures hautement organisées et leur relation avec l’actine. Ceci sera utile pour l’élaboration du réseau d’interactions des septines qui pourra servir à expliquer leur dérégulation dans le cancer et les maladies neurodégénératives. / Septins are highly conserved GTP-binding proteins deregulated in diseases such as cancer and neurodegenerative diseases. Septins scaffold other proteins and act as diffusion barriers at the plasma membrane and midbody during cytokinesis. They interact with the actin cytoskeleton and have been observed to form ordered complexes that can polymerize into higher-order structures such as filaments and rings. The principles of assembly and disassembly of such filaments and rings and their cellular roles are yet to be elucidated. Drosophila offers a simple system, as there are only 5 septin genes: peanut, sep1, sep2, sep4, and sep5 in contrast to 13 found in humans.
We have previously found that Drosophila S2 cells contain unusual tubular structures that label with Peanut antibody. The goal of my Master’s project has been to characterise these structures, by defining their constituents, behaviours and properties, in the hope that this will shed light on the mechanisms by which septins can form higher-ordered structures and how they interact with other cytoskeletal elements such as actin.
Using fluorescence microscopy, I show that these tubes are cytoplasmic and present in 30% of cells, both during mitosis and interphase, suggesting they are not cell cycle regulated. To investigate their composition and dynamic properties, I generated S2 cell lines stably expressing Sep2-GFP, which localizes to septin tubes and double stranded RNAs against all septins. The products of three septin genes were found to be essential for the assembly of septin tubes: Sep1, Sep2, and Pnut. The depletion of Sep1 led to the dispersal of the GFP signal into cytoplasmic clumps, whereas the depletion of Sep2 and Pnut led to its uniform distribution through the cell. FRAP analysis of Sep2-GFP revealed only slow recovery after many hours, indicating that the structures are very stable.
I also discovered an unusual relationship between septin tubes and the actin cytoskeleton. Although the tubes did not label with conventional F-actin probes (Phalloidin, LifeAct), treatment with inhibitors of F-actin assembly (Latrunculin A) or disassembly (Jasplakinolide) led to their rapid (<30 min) dispersal into scattered rings. Furthermore, overexpressed Actin05C-mCherry localised to the septin tubes, while a polymerization-deficient mutant Actin05C-R62D-mCherry did not. Depletion of the actin severing protein Cofilin and the actin capping protein AIP1 also disrupted septin tubes dispersing them into cytoplasmic rings. A dynamic actin cytoskeleton is thus required for the formation and/or maintenance of higher ordered structures such as rings and tubes.
Conclusion and Relevance: Ongoing studies aim to further elucidate how septins organize into such ordered structures and how actin regulates the process. This will clarify the septin network of interactions and facilitate the comprehension of their implication in cancer and neurodegenerative diseases.
Key words: actin, septin, tube, ring, higher-order, hexamer, cofilin, AIP1
|
186 |
Modèles d'évaluation et d'allocations des actifs financiers dans le cadre de non normalité des rendements : essais sur le marché françaisHafsa, Houda 12 November 2012 (has links)
Depuis quelques années, la recherche financière s'inscrit dans une nouvelle dynamique. La nécessité de mieux modéliser le comportement des rendements des actifs financiers et les risques sur les marchés pousse les chercheurs à trouver des mesures de risque plus adéquates. Ce travail de recherche se situe dans cette évolution, ayant admis les caractéristiques des séries financières par des faits stylisés tels que la non normalité des rendements. A travers cette thèse nous essayons de montrer l'importance d'intégrer des mesures de risque qui tiennent compte de la non normalité dans le processus d'évaluation et d'allocation des actifs financiers sur le marché français. Cette thèse propose trois chapitres correspondant chacun à un article de recherche académique. Le premier article propose de revisiter les modèles d'évaluation en prenant en compte des moments d'ordres supérieurs dans un cadre de downside risk. Les résultats indiquent que les downside co-moments d'ordres supérieurs sont déterminants dans l'explication des variations des rendements en coupe transversale. Le second chapitre propose de mettre en relation la rentabilité financière et le risque mesuré par la VaR ou la CVaR. Nous trouvons que la VaR présente un pouvoir explicatif plus élevé que celui de la CVaR et que l'approche normale est plus intéressante que l'approche basée sur l'expansion de Cornish-Fisher (1937). Ces deux résultats contredisent les prédictions théoriques mais nous avons pu démontrer qu'ils sont inhérents au marché français. Le troisième chapitre propose une autre piste, nous revisitons le modèle moyenne-CVaR dans un cadre dynamique et en présence des coûts de transaction / This dissertation is part of an ongoing researches looking for an adequate model that apprehend the behavior of financial asset returns. Through this research, we propose to analyze the relevance of risk measures that take into account the non-normality in the asset pricing and portfolio allocation models on the French market. This dissertation is comprised of three articles. The first one proposes to revisit the asset pricing model taking into account the higher-order moments in a downside framework. The results indicate that the downside higher order co-moments are relevant in explaining the cross sectional variations of returns. The second paper examines the relation between expected returns and the VaR or CVaR. A cross sectional analysis provides evidence that VaR is superior measure of risk when compared to the CVaR. We find also that the normal estimation approach gives better results than the approach based on the expansion of Cornish-Fisher (1937). Both results contradict the theoretical predictions but we proved that they are inherent to the French market. In the third paper, we review the mean-CVaR model in a dynamic framework and we take into account the transaction costs. The results indicate that the asset allocation model that takes into account the non-normality can improve the performance of the portfolio comparing to the mean-variance model, in terms of the average return and the return-to CVaR ratio. Through these three studies, we think that it is possible to modify the risk management framework to apprehend in a better way the risk of loss associated to the non-normality problem
|
187 |
Vícedětné rodiny: faktory a kontexty / Multi-child families: patterns and backgroundFaberová, Anna January 2018 (has links)
Multi-child families: patterns and background Abstract The scope of this thesis is to analyse fertility rate in the Czech Republic aiming at the third- and higher-order fertility rates. The objective consists in finding, defining and assessing the determinants and factors that may affect the higher-order fertility rates. As the pivotal source of the input data, the individual anonymized data from the Czech Republic Population and Housing Census held in 2011 were used. Only the data referring to women born in the period 1920-1971 that gave at least two live births were selected for the purpose of the detailed fertility rates analysis. The analysis was carried out by way of nominal logistic regression. The gain of the thesis is a detailed evaluation of the higher- order fertility rates pursuant to the four factors or determinants - concerning the women in question - that have been selected for this purpose, i.e.: highest level of completed education; size of the municipality of residence; religious belief; and characteristics according to the fertile partnerships. In particular the mutual links among the factors in question and the higher-order fertility rates are specified in the thesis. Keywords: multi-child families, fertility, higher-order fertility, generation, census, nominal logistic regression
|
188 |
Efeito de não linearidades estruturais na resposta aeroelástica de aerofólios / Effect of structural nonlinearities in the aeroelastic response of airfoilsPereira, Daniel de Almeida 04 August 2015 (has links)
A aeroelasticidade estuda a interação mútua entre os efeitos aerodinâmicos e estruturais. É sabido que essa relação muitas vezes se comporta de maneira não linear, causando diversos problemas, tais como flutter, oscilações em ciclo limite, bifurcações e caos. Tais fenômenos são difíceis de serem diagnosticados, podendo causar problemas graves à estrutura das aeronaves e também inviabilizar as suas operações. Dentre as principais fontes de não linearidades em sistemas aeroelásticos, pode-se citar as de origem aerodinâmica e estrutural. As de origem estrutural, por sua vez, podem ter caráter distribuído ou concentrado. Sabe-se que os efeitos estruturais concentrados denominados enrijecimento e folga são os de maior impacto na aeroelasticidade não linear. Desse modo, o objetivo desse trabalho é estudar a interação não linear entre duas não linearidades estruturais, ou seja, o enrijecimento associado à rigidez em torção e a folga presente nas articulações das superfícies de controle de seções típicas aeroelásticas. Experimentos em túnel de vento são realizados utilizando um dispositivo que permite variar a intensidade do efeito de enrijecimento e do tamanho da folga na articulação da superfície de comando. O modelo numérico de seção típica aeroelástica também é utilizado e validado com dados experimentais. Análises por meio de diagramas de bifurcação de Hopf e técnicas baseadas em espectros de potência são utilizadas. Todas as respostas aeroelásticas foram caracterizadas através de ferramentas de análise nos domínios do tempo e da frequência, como técnica de reconstrução de espaço de estados e os espectros de alta ordem (HOS), os quais são importantes na identificação dos tipos de acoplamentos não lineares. Resultados indicam que a combinação dos efeitos de enrijecimento e folga são responsáveis pelo comportamento subcrítico das bifurcações de Hopf e que a intensidade do enrijecimento tem influência direta nas amplitudes de ciclo limite. / Aeroelasticity is the field of engineering that deals with the mutual interaction between the aerodynamic and structural dynamics effects. It is known that this relationship often shows nonlinear behavior, causing various problems such as flutter, limit cycle oscillations, bifurcations and chaos. Such phenomena are difficult to predict and can cause serious problems to the aircraft structure and also they can jeopardize their operations. The unsteady aerodynamic and structural dynamics provide the main sources of nonlinearities in aeroelastic systems. Structural nonlinearities can be treated as distributed or concentrated effects. It is know that the nonlinear concentrated structural effects referred as hardening and freeplay have a significant impact on nonlinear aeroelasticity. The objective of this work is to analyze an aeroelastic system under the influence of combined structural nonlinearities, i.e., the hardening nonlinearity in the pitch airfoil motion and the freeplay nonlinearity in the control surface hinge. Wind tunnel experiments are carried out using one device that allows to vary the intensity of the hardening effect and the size of the freeplay gap in the control surface hinge. The numerical model of the typical aeroelastic section is also used and validated with experimental data. All aeroelastic responses are characterized by analytical tools in time and frequency domains. It was used the state space reconstruction technique and the higher order spectral analysis (HOS) to identify types of nonlinear couplings. The results indicate that the combination of hardening and freeplay effects are responsible for inducing the subcritical behavior on the Hopf bifurcations and that the intensity of the stiffness has a direct influence on the limit cycle amplitudes.
|
189 |
On syzygies of algebraic varieties with applications to moduliAgostini, Daniele 17 September 2018 (has links)
Diese Dissertation beschäftigt sich mit asymptotischen Syzygien und Gleichungen Abelscher Varietäten, sowie mit deren Anwendung auf zyklische Überdeckungen von Kurven von Geschlecht zwei.
Was asymptotischen Syzygien angeht, zeigen wir für beliebige Geradenbündel auf projektiven Schemata: Wenn die asymptotischen Syzygien von Grad p eines Geradenbündels verschwinden, dann ist das Geradenbündel p-sehr ampel. Darüber hinaus verwenden wir die Bridgeland-King-Reid-Haiman Korrespondenz, um zu zeigen, dass dieses Ergebnis auch umgekehrt wahr ist, wenn es um eine glatte Fläche und kleine p geht. Dies dehnt Ergebnisse von Ein-Lazarsfeld und Ein-Lazarsfeld-Yang aus. Wir verwenden unsere Ergebnisse, um zu untersuchen, wie Syzygien verwendet werden können, um den Grad der Irrationalität einer Varietät zu begrenzen.
Ferner, beweisen wir eine Vermutung von Gross and Popescu über Abelsche Flächen, deren Ideal durch Quadriken und Kubiken erzeugt wird.
Außerdem verwenden wir die projektive Normalität einer Abelschen Fläche, um die Prym Abbildung, die mit zyklischen Überdeckungen von Geschlecht zwei Kurven assoziert ist, zu untersuchen. Wir zeigen, dass das Differential der Abbildung generisch injektiv ist, wenn der Grad der Überdeckung mindestens sieben ist. Wir dehnen damit Ergebnisse von Lange und Ortega aus. Abschließend zeigen wir, dass das Differential genau für bielliptische Überdeckungen nicht injectiv ist. / In this thesis we study asymptotic syzygies of algebraic varieties and equations of abelian surfaces, with applications
to cyclic covers of genus two curves.
First, we show that vanishing of asymptotic p-th syzygies implies p-very ampleness for line bundles on arbitrary projective schemes. For smooth surfaces we prove that the converse holds, when p is small, by studying the Bridgeland-King-Reid-Haiman correspondence for the Hilbert scheme of points. This extends previous results of Ein-Lazarsfeld and Ein-Lazarsfeld-Yang. As an application of our results, we show how to use syzygies to bound the irrationality of a variety.
Furthermore, we confirm a conjecture of Gross and Popescu about abelian surfaces whose ideal is generated by quadrics and cubics.
In addition, we use projective normality of abelian surfaces to study the Prym map associated to cyclic covers of genus two curves. We show that the differential of the map is generically injective as soon as the degree of the cover is at least seven, extending a previous result of Lange and Ortega. Moreover, we show that the differentials fails to be injective precisely at bielliptic covers.
|
190 |
Structured higher-order algorithmic differentiation in the forward and reverse mode with application in optimum experimental designWalter, Sebastian 07 May 2012 (has links)
In dieser Arbeit werden Techniken beschrieben, die es erlauben (höhere) Ableitungen und Taylorapproximationen solcher Computerprogramme effizient zu berechnen. Auch inbesondere dann, wenn die Programme Algorithmen der numerischen linearen Algebra (NLA) enthalten. Im Gegensatz zur traditionellen algorithmischen Differentiation (AD), bei der die zugrunde liegenden Algorithmen um zusätzliche Befehlere erweitert werden, sind in dieser Arbeit die Zerlegungen durch definierende Gleichungen charakterisiert. Basierend auf den definierenden Gleichungen werden Strukturausnutzende Algorithmen hergeleitet. Genauer, neuartige Algorithmen für die Propagation von Taylorpolynomen durch die QR, Cholesky und reell-symmetrischen Eigenwertzerlegung werden präsentiert. Desweiteren werden Algorithmen für den Rückwärtsmodus der AD hergeleitet, welche im Wesentlichen nur die Faktoren der Zerlegungen benötigen. Im Vergleich zum traditionellen Ansatz, bei dem alle Zwischenergebnisse gespeichert werden, ist dies eine Reduktion von O(N^3) zu O(N^2) für Algorithmen mit O(N^3) Komplexität. N ist hier die Größe der Matrix. Zusätzlich kann bestehende, hoch-optimierte Software verwendet werden. Ein Laufzeitvergleich zeigt, dass dies im Vergleich zum traditionellen Ansatz zu einer Beschleunigung in der Größenordnung 100 führen kann. Da die NLA Funktionen als Black Box betrachtet werden, ist desweiteren auch der Berechnungsgraph um Größenordnungen kleiner. Dies bedeutet, dass Software, welche Operator Overloading benutzt, weniger Overhead hervorruft und auch weniger Speicher benötigt. / This thesis provides a framework for the evaluation of first and higher-order derivatives and Taylor series expansions through large computer programs that contain numerical linear algebra (NLA) functions. It is a generalization of traditional algorithmic differentiation (AD) techniques in that NLA functions are regarded as black boxes where the inputs and outputs are related by defining equations. Based on the defining equations, structure-exploiting algorithms are derived. More precisely, novel algorithms for the propagation of Taylor polynomials through the QR, Cholesky,- and real-symmetric eigenvalue decomposition are shown. Recurrences for the reverse mode of AD, which require essentially only the returned factors of the decomposition, are also derived. Compared to the traditional approach where all intermediates of an algorithm are stored, this is a reduction from O(N^3) to O(N^2) for algorithms with O( N^3) complexity. N denotes the matrix size. The derived algorithms make it possible to use existing high-performance implementations. A runtime comparison shows that the treatment of NLA functions as atomic can be more than one order of magnitude faster than an automatic differentiation of the underlying algorithm. Furthermore, the computational graph is orders of magnitudes smaller. This reduces the additional memory requirements, as well as the overhead, of operator overloading techniques to a fraction.
|
Page generated in 0.0357 seconds