Learning in neural networks and stochastic approximation methods with averaging

The problem of adjusting the weights (learning) in multilayer feedforward neural networks (NN) is known to be of a high importance when utilizing NN techniques in various practical applications. The learning procedure is to be performed as fast as possible and in a simple computational fashion, the two requirements which are usually not satisfied practically by the methods developed so far. Moreover, the presence of random inaccuracies are usually not taken into account. In view of these three issues, an alternative stochastic approximation approach discussed in the paper, seems to be very promising.

Stress field control of eruption dynamics at a rift volcano: Nyamuragira, D.R.Congo

Using the record of 30 flank eruptions over the last 110 years at Nyamuragira, we have tested the relationship between the eruption dynamics and the local stress field. There are two groups of eruptions based on their duration (< 80days >) that are also clustered in space and time. We find that the eruptions fed by dykes parallel to the East African Rift Valley have longer durations (and larger volumes) than those eruptions fed by dykes with other orientations. This is compatible with a model for compressible magma transported through an elastic-walled dyke in a differential stress field from an over-pressured reservoir (Woods et al., 2006). The observed pattern of eruptive fissures is consistent with a local stress field modified by a northwest-trending, right lateral slip fault that is part of the northern transfer zone of the Kivu Basin rift segment. We have also re-tested with new data the stochastic eruption models for Nyamuragira of Burt et al. (1994). The time-predictable, pressure-threshold model remains the best fit and is consistent with the typically observed declining rate of sulphur dioxide emission during the first few days of eruption with lava emission from a depressurising, closed, crustal reservoir. The 2.4-fold increase in long-term eruption rate that occurred after 1977 is confirmed in the new analysis. Since that change, the record has been dominated by short-duration eruptions fed by dykes perpendicular to the Rift. We suggest that the intrusion of a major dyke during the 1977 volcano-tectonic event at neighbouring Nyiragongo volcano inhibited subsequent dyke formation on the southern flanks of Nyamuragira and this may also have resulted in more dykes reaching the surface elsewhere. Thus that sudden change in output was a result of a changed stress field that forced more of the deep magma supply to the surface. Another volcano-tectonic event in 2002 may also have changed the magma output rate at Nyamuragira.

Optimal growth strategies when mortality and production rates are size-dependent

Pontryagin's maximum principle from optimal control theory is used to find the optimal allocation of energy between growth and reproduction when lifespan may be finite and the trade-off between growth and reproduction is linear. Analyses of the optimal allocation problem to date have generally yielded bang-bang solutions, i.e. determinate growth: life-histories in which growth is followed by reproduction, with no intermediate phase of simultaneous reproduction and growth. Here we show that an intermediate strategy (indeterminate growth) can be selected for if the rates of production and mortality either both increase or both decrease with increasing body size, this arises as a singular solution to the problem. Our conclusion is that indeterminate growth is optimal in more cases than was previously realized. The relevance of our results to natural situations is discussed.

Latching control of a point absorber

In this paper the authors investigate the use of optimal control techniques for improving the efficiency of the power conversion system in a point absorber wave power device. A simple mathematical model of the system is developed and an optimal control strategy for power generation is determined. They describe an algorithm for solving the problem numerically, provided the incident wave force is given. The results show that the performance of the device is significantly improved with the handwidth of the response being widened by the control strategy.

Optimal steady motions for oriented vehicles

Motivated by the motion planning problem for oriented vehicles travelling in a 3-Dimensional space; Euclidean space E3, the sphere S3 and Hyperboloid H3. For such problems the orientation of the vehicle is naturally represented by an orthonormal frame over a point in the underlying manifold. The orthonormal frame bundles of the space forms R3,S3 and H3 correspond with their isometry groups and are the Euclidean group of motion SE(3), the rotation group SO(4) and the Lorentzian group SO(1; 3) respectively. Orthonormal frame bundles of space forms coincide with their isometry groups and therefore the focus shifts to left-invariant control systems defined on Lie groups. In this paper a method for integrating these systems is given where the controls are time-independent. For constant twist motions or helical motions, the corresponding curves g(t) 2 SE(3) are given in closed form by using the well known Rodrigues’ formula. However, this formula is only applicable to the Euclidean case. This paper gives a method for computing the non-Euclidean screw/helical motions in closed form. This involves decoupling the system into two lower dimensional systems using the double cover properties of Lie groups, then the lower dimensional systems are solved explicitly in closed form.

Body-centric modelling, identification, and acceleration tracking control of a quadrotor UAV

This paper presents the mathematical development of a body-centric nonlinear dynamic model of a quadrotor UAV that is suitable for the development of biologically inspired navigation strategies. Analytical approximations are used to find an initial guess of the parameters of the nonlinear model, then parameter estimation methods are used to refine the model parameters using the data obtained from onboard sensors during flight. Due to the unstable nature of the quadrotor model, the identification process is performed with the system in closed-loop control of attitude angles. The obtained model parameters are validated using real unseen experimental data. Based on the identified model, a Linear-Quadratic (LQ) optimal tracker is designed to stabilize the quadrotor and facilitate its translational control by tracking body accelerations. The LQ tracker is tested on an experimental quadrotor UAV and the obtained results are a further means to validate the quality of the estimated model. The unique formulation of the control problem in the body frame makes the controller better suited for bio-inspired navigation and guidance strategies than conventional attitude or position based control systems that can be found in the existing literature.

Investigating music tempo as a feedback mechanism for closed-loop BCI control

The feedback mechanism used in a brain-computer interface (BCI) forms an integral part of the closed-loop learning process required for successful operation of a BCI. However, ultimate success of the BCI may be dependent upon the modality of the feedback used. This study explores the use of music tempo as a feedback mechanism in BCI and compares it to the more commonly used visual feedback mechanism. Three different feedback modalities are compared for a kinaesthetic motor imagery BCI: visual, auditory via music tempo, and a combined visual and auditory feedback modality. Visual feedback is provided via the position, on the y-axis, of a moving ball. In the music feedback condition, the tempo of a piece of continuously generated music is dynamically adjusted via a novel music-generation method. All the feedback mechanisms allowed users to learn to control the BCI. However, users were not able to maintain as stable control with the music tempo feedback condition as they could in the visual feedback and combined conditions. Additionally, the combined condition exhibited significantly less inter-user variability, suggesting that multi-modal feedback may lead to more robust results. Finally, common spatial patterns are used to identify participant-specific spatial filters for each of the feedback modalities. The mean optimal spatial filter obtained for the music feedback condition is observed to be more diffuse and weaker than the mean spatial filters obtained for the visual and combined feedback conditions.

Optimal infinity-quasiconformal immersions

For a Hamiltonian K ∈ C2(RN × n) and a map u:Ω ⊆ Rn − → RN, we consider the supremal functional (1) The “Euler−Lagrange” PDE associated to (1)is the quasilinear system (2) Here KP is the derivative and [ KP ] ⊥ is the projection on its nullspace. (1)and (2)are the fundamental objects of vector-valued Calculus of Variations in L∞ and first arose in recent work of the author [N. Katzourakis, J. Differ. Eqs. 253 (2012) 2123–2139; Commun. Partial Differ. Eqs. 39 (2014) 2091–2124]. Herein we apply our results to Geometric Analysis by choosing as K the dilation function which measures the deviation of u from being conformal. Our main result is that appropriately defined minimisers of (1)solve (2). Hence, PDE methods can be used to study optimised quasiconformal maps. Nonconvexity of K and appearance of interfaces where [ KP ] ⊥ is discontinuous cause extra difficulties. When n = N, this approach has previously been followed by Capogna−Raich ? and relates to Teichmüller’s theory. In particular, we disprove a conjecture appearing therein.

On the cutting stock problem under stochastic demand

This paper addresses the one-dimensional cutting stock problem when demand is a random variable. The problem is formulated as a two-stage stochastic nonlinear program with recourse. The first stage decision variables are the number of objects to be cut according to a cutting pattern. The second stage decision variables are the number of holding or backordering items due to the decisions made in the first stage. The problem`s objective is to minimize the total expected cost incurred in both stages, due to waste and holding or backordering penalties. A Simplex-based method with column generation is proposed for solving a linear relaxation of the resulting optimization problem. The proposed method is evaluated by using two well-known measures of uncertainty effects in stochastic programming: the value of stochastic solution-VSS-and the expected value of perfect information-EVPI. The optimal two-stage solution is shown to be more effective than the alternative wait-and-see and expected value approaches, even under small variations in the parameters of the problem.

Stochastic Assessment Of Environmental Flows In Semiarid Environments

The regimen of environmental flows (EF) must be included as terms of environmental demand in the management of water resources. Even though there are numerous methods for the computation of EF, the criteria applied at different steps in the calculation process are quite subjective whereas the results are fixed values that must be meet by water planners. This study presents a friendly-user tool for the assessment of the probability of compliance of a certain EF scenario with the natural regimen in a semiarid area in southern Spain. 250 replications of a 25-yr period of different hydrological variables (rainfall, minimum and maximum flows, ...) were obtained at the study site from the combination of Monte Carlo technique and local hydrological relationships. Several assumptions are made such as the independence of annual rainfall from year to year and the variability of occurrence of the meteorological agents, mainly precipitation as the main source of uncertainty. Inputs to the tool are easily selected from a first menu and comprise measured rainfall data, EF values and the hydrological relationships for at least a 20-yr period. The outputs are the probabilities of compliance of the different components of the EF for the study period. From this, local optimization can be applied to establish EF components with a certain level of compliance in the study period. Different options for graphic output and analysis of results are included in terms of graphs and tables in several formats. This methodology turned out to be a useful tool for the implementation of an uncertainty analysis within the scope of environmental flows in water management and allowed the simulation of the impacts of several water resource development scenarios in the study site.

On Real-Time Monitoring And Control For Efficient Management Of Drinking Water Networks: Barcelona Case Study

Drinking water utilities in urban areas are focused on finding smart solutions facing new challenges in their real-time operation because of limited water resources, intensive energy requirements, a growing population, a costly and ageing infrastructure, increasingly stringent regulations, and increased attention towards the environmental impact of water use. Such challenges force water managers to monitor and control not only water supply and distribution, but also consumer demand. This paper presents and discusses novel methodologies and procedures towards an integrated water resource management system based on advanced ICT technologies of automation and telecommunications for largely improving the efficiency of drinking water networks (DWN) in terms of water use, energy consumption, water loss minimization, and water quality guarantees. In particular, the paper addresses the first results of the European project EFFINET (FP7-ICT2011-8-318556) devoted to the monitoring and control of the DWN in Barcelona (Spain). Results are split in two levels according to different management objectives: (i) the monitoring level is concerned with all the aspects involved in the observation of the current state of a system and the detection/diagnosis of abnormal situations. It is achieved through sensors and communications technology, together with mathematical models; (ii) the control level is concerned with computing the best suitable and admissible control strategies for network actuators as to optimize a given set of operational goals related to the performance of the overall system. This level covers the network control (optimal management of water and energy) and the demand management (smart metering, efficient supply). The consideration of the Barcelona DWN as the case study will allow to prove the general applicability of the proposed integrated ICT solutions and their effectiveness in the management of DWN, with considerable savings of electricity costs and reduced water loss while ensuring the high European standards of water quality to citizens.

Reservoir Management For Flood And Drought Protection Using Infinite Horizon Model Predictive Control

Model Predictive Control (MPC) is a control method that solves in real time an optimal control problem over a finite horizon. The finiteness of the horizon is both the reason of MPC's success and its main limitation. In operational water resources management, MPC has been in fact successfully employed for controlling systems with a relatively short memory, such as canals, where the horizon length is not an issue. For reservoirs, which have generally a longer memory, MPC applications are presently limited to short term management only. Short term reservoir management can be effectively used to deal with fast process, such as floods, but it is not capable of looking sufficiently ahead to handle long term issues, such as drought. To overcome this limitation, we propose an Infinite Horizon MPC (IH-MPC) solution that is particularly suitable for reservoir management. We propose to structure the input signal by use of orthogonal basis functions, therefore reducing the optimization argument to a finite number of variables, and making the control problem solvable in a reasonable time. We applied this solution for the management of the Manantali Reservoir. Manantali is a yearly reservoir located in Mali, on the Senegal river, affecting water systems of Mali, Senegal, and Mauritania. The long term horizon offered by IH-MPC is necessary to deal with the strongly seasonal climate of the region.

Multistep stochastic mirror descent for risk-averse convex stochastic programs based on extended polyhedral risk measures

We consider risk-averse convex stochastic programs expressed in terms of extended polyhedral risk measures. We derive computable con dence intervals on the optimal value of such stochastic programs using the Robust Stochastic Approximation and the Stochastic Mirror Descent (SMD) algorithms. When the objective functions are uniformly convex, we also propose a multistep extension of the Stochastic Mirror Descent algorithm and obtain con dence intervals on both the optimal values and optimal solutions. Numerical simulations show that our con dence intervals are much less conservative and are quicker to compute than previously obtained con dence intervals for SMD and that the multistep Stochastic Mirror Descent algorithm can obtain a good approximate solution much quicker than its nonmultistep counterpart. Our con dence intervals are also more reliable than asymptotic con dence intervals when the sample size is not much larger than the problem size.

Flexible information acquisition and optimal Tobin tax in tractable dynamic global games

My dissertation focuses on dynamic aspects of coordination processes such as reversibility of early actions, option to delay decisions, and learning of the environment from the observation of other people’s actions. This study proposes the use of tractable dynamic global games where players privately and passively learn about their actions’ true payoffs and are able to adjust early investment decisions to the arrival of new information to investigate the consequences of the presence of liquidity shocks to the performance of a Tobin tax as a policy intended to foster coordination success (chapter 1), and the adequacy of the use of a Tobin tax in order to reduce an economy’s vulnerability to sudden stops (chapter 2). Then, it analyzes players’ incentive to acquire costly information in a sequential decision setting (chapter 3). In chapter 1, a continuum of foreign agents decide whether to enter or not in an investment project. A fraction λ of them are hit by liquidity restrictions in a second period and are forced to withdraw early investment or precluded from investing in the interim period, depending on the actions they chose in the first period. Players not affected by the liquidity shock are able to revise early decisions. Coordination success is increasing in the aggregate investment and decreasing in the aggregate volume of capital exit. Without liquidity shocks, aggregate investment is (in a pivotal contingency) invariant to frictions like a tax on short term capitals. In this case, a Tobin tax always increases success incidence. In the presence of liquidity shocks, this invariance result no longer holds in equilibrium. A Tobin tax becomes harmful to aggregate investment, which may reduces success incidence if the economy does not benefit enough from avoiding capital reversals. It is shown that the Tobin tax that maximizes the ex-ante probability of successfully coordinated investment is decreasing in the liquidity shock. Chapter 2 studies the effects of a Tobin tax in the same setting of the global game model proposed in chapter 1, with the exception that the liquidity shock is considered stochastic, i.e, there is also aggregate uncertainty about the extension of the liquidity restrictions. It identifies conditions under which, in the unique equilibrium of the model with low probability of liquidity shocks but large dry-ups, a Tobin tax is welfare improving, helping agents to coordinate on the good outcome. The model provides a rationale for a Tobin tax on economies that are prone to sudden stops. The optimal Tobin tax tends to be larger when capital reversals are more harmful and when the fraction of agents hit by liquidity shocks is smaller. Chapter 3 focuses on information acquisition in a sequential decision game with payoff complementar- ity and information externality. When information is cheap relatively to players’ incentive to coordinate actions, only the first player chooses to process information; the second player learns about the true payoff distribution from the observation of the first player’s decision and follows her action. Miscoordination requires that both players privately precess information, which tends to happen when it is expensive and the prior knowledge about the distribution of the payoffs has a large variance.