Showing papers by "École Polytechnique de Montréal published in 2007"

An overview of bilevel optimization

Abstract: This paper is devoted to bilevel optimization, a branch of mathematical programming of both practical and theoretical interest. Starting with a simple example, we proceed towards a general formulation. We then present fields of application, focus on solution approaches, and make the connection with MPECs (Mathematical Programs with Equilibrium Constraints).

Scaling learning algorithms towards AI

Abstract: One long-term goal of machine learning research is to produce methods that are applicable to highly complex tasks, such as perception (vision, audition), reasoning, intelligent control, and other artificially intelligent behaviors. We argue that in order to progress toward this goal, the Machine Learning community must endeavor to discover algorithms that can learn highly complex functions, with minimal need for prior knowledge, and with minimal human intervention. We present mathematical and empirical evidence suggesting that many popular approaches to non-parametric learning, particularly kernel methods, are fundamentally limited in their ability to learn complex high-dimensional functions. Our analysis focuses on two problems. First, kernel machines are shallow architectures, in which one large layer of simple template matchers is followed by a single layer of trainable coefficients. We argue that shallow architectures can be very inefficient in terms of required number of computational elements and examples. Second, we analyze a limitation of kernel machines with a local kernel, linked to the curse of dimensionality, that applies to supervised, unsupervised (manifold learning) and semi-supervised kernel machines. Using empirical results on invariant image recognition tasks, kernel methods are compared with deep architectures, in which lower-level features or concepts are progressively combined into more abstract and higher-level representations. We argue that deep architectures have the potential to generalize in non-local ways, i.e., beyond immediate neighbors, and that this is crucial in order to make progress on the kind of complex tasks required for artificial intelligence.

Large-Population Cost-Coupled LQG Problems With Nonuniform Agents: Individual-Mass Behavior and Decentralized $\varepsilon$ -Nash Equilibria

Abstract: We consider linear quadratic Gaussian (LQG) games in large population systems where the agents evolve according to nonuniform dynamics and are coupled via their individual costs. A state aggregation technique is developed to obtain a set of decentralized control laws for the individuals which possesses an epsiv-Nash equilibrium property. A stability property of the mass behavior is established, and the effect of inaccurate population statistics on an isolated agent is also analyzed by variational techniques.

Passive treatment of acid mine drainage in bioreactors using sulfate-reducing bacteria: critical review and research needs.

Abstract: Acid mine drainage (AMD), characterized by low pH and high concentrations of sulfate and heavy metals, is an important and widespread environmental problem related to the mining industry. Sulfate-reducing passive bioreactors have received much attention lately as promising biotechnologies for AMD treatment. They offer advantages such as high metal removal at low pH, stable sludge, very low operation costs, and minimal energy consumption. Sulfide precipitation is the desired mechanism of contaminant removal; however, many mechanisms including adsorption and precipitation of metal carbonates and hydroxides occur in passive bioreactors. The efficiency of sulfate-reducing passive bioreactors is sometimes limited because they rely on the activity of an anaerobic microflora [including sulfate-reducing bacteria (SRB)] which is controlled primarily by the reactive mixture composition. The most important mixture component is the organic carbon source. The performance of field bioreactors can also be limited by AMD load and metal toxicity. Several studies conducted to find the best mixture of natural organic substrates for SRB are reviewed. Moreover, critical parameters for design and long-term operation are discussed. Additional work needs to be done to properly assess the long-term efficiency of reactive mixtures and the metal removal mechanisms. Furthermore, metal speciation and ecotoxicological assessment of treated effluent from on-site passive bioreactors have yet to be performed.

Design criteria for microstructured-optical-fiber-based surface-plasmon-resonance sensors

Abstract: Design strategies for microstructured-optical-fiber (MOF-) based surface-plasmon-resonance (SPR) sensors are presented. In such sensors, plasmons on the inner surface of the large metallized channels containing analyte can be excited by a fundamental mode of a single-mode microstructured fiber. Phase matching between a plasmon and a core mode can be enforced by introducing air-filled microstructures into the fiber core. Particularly, in its simplest implementation, the effective refractive index of a fundamental mode can be lowered to match that of a plasmon by introducing a small central hole into the fiber core. Resolution of the MOF-based sensors is demonstrated to be as low as 3×10−5 RIU, where RIU means refractive index unit. The ability to integrate large-size microfluidic channels for efficient analyte flow together with a single-mode waveguide of designable modal refractive index is attractive for the development of integrated highly sensitive MOF-SPR sensors operating at any designable wavelength.

Properties of nanoparticles generated during femtosecond laser machining in air and water

Abstract: Femtosecond laser ablation is used to generate nanoparticle aerosols and colloids from solid targets of various materials (Ti, Ag, Au, Co, etc.) in air and water ambience. We determine the influence of different laser parameters (pulse energy, pulse overlap) and properties of media (air, airbrush, water) on the rate of production and size distribution of the laser-generated nanoparticles. It is shown that the pulse overlap and laser fluence are the parameters determining the nanoparticle size. At optimum conditions the nanoparticle productivity can be increased by 150–300%. The generation of multimaterial nanoparticle dispersions is demonstrated. Being free of toxic impurities, the laser-produced nanoparticles may be promising for biomedical applications.

Development of Millimeter-Wave Planar Diplexers Based on Complementary Characters of Dual-Mode Substrate Integrated Waveguide Filters With Circular and Elliptic Cavities

Abstract: A novel high-performance millimeter-wave planar diplexer is developed based on the complementary characters of substrate integrated waveguide (SIW) dual-mode filters with circular and elliptic cavities by making the tradeoff between the isolation, insertion loss, and selectivity. The responses of the dual-mode SIW circular and elliptic cavities are first investigated. It can be found that the upper side response of the circular cavity and the lower side response of the elliptic cavity are very steep. The diplexers with high isolation performance are then designed based on the complementary response characters of circular and elliptic cavities. A diplexer with two dual-mode SIW circular and elliptic cavities is designed and fabricated with a normal printed circuit board process. The measured insertion losses are 1.95 and 2.09 dB in the upper and lower passbands centered at 26 and 25 GHz with the fractional bandwidths of 5.2% and 5.4%. The isolation is lower than - 50dB

Design of an Electronically Beam Scanning Reflectarray Using Aperture-Coupled Elements

Abstract: The design of a C-band electronically beam scanning reflectarray is presented. The reconfigurable reflectarray element consists of a microstrip patch, printed on a flexible membrane substrate, aperture-coupled to a transmission line loaded with two varactor diodes. The designed element allows continuous tuning of the reflected signal's phase over a 360deg range with a maximum loss of 2.4 dB at 5.4 GHz. The measured results on a 30-element reflectarray breadboard show that by varying the bias voltages on each element the main beam can be steered to large angles, up to 40deg from broadside in the H-plane. A loss analysis of the proposed reflectarray is also presented

Kinetics of lactic acid fermentation on glucose and corn by Lactobacillus amylophilus

Abstract: The biosynthesis of ammonium lactate, a product of lactic acid fermentation was studied from corn and glucose at five different pH values of 5.4 to 7.8. In the glucose fermentations, a 100% conversion of substrate was obtained resulting in a maximum lactic acid production yield of 93.2%. The optimum pH for the maximum volumetric rate of lactic acid biosynthesis (1.56 g dm−3 h−1) was between 6.0 and 6.5. The corn fermentations were slower than the glucose fermentations with a resulting lactic acid yield of 67.5%. Hydrolysis of corn by enzymatic or chemical methods as well as the use of ammonium hydroxide for pH control increased both the final concentration and the rates of lactic acid production. An enhanced yield of more than 90% was finally obtained in the corn fermentations. A logistic model adequately described the kinetics of biomass growth, lactic acid production and sugar utilization in the glucose fermentations at different pH values. The dynamics of lactic acid formation in the corn fermentations were also successfully described by the developed model. The dependence of the model parameters on pH was investigated.

Analysis of squat and stoop dynamic liftings: muscle forces and internal spinal loads

Abstract: Despite the well-recognized role of lifting in back injuries, the relative biomechanical merits of squat versus stoop lifting remain controversial. In vivo kinematics measurements and model studies are combined to estimate trunk muscle forces and internal spinal loads under dynamic squat and stoop lifts with and without load in hands. Measurements were performed on healthy subjects to collect segmental rotations during lifts needed as input data in subsequent model studies. The model accounted for nonlinear properties of the ligamentous spine, wrapping of thoracic extensor muscles to take curved paths in flexion and trunk dynamic characteristics (inertia and damping) while subject to measured kinematics and gravity/external loads. A dynamic kinematics-driven approach was employed accounting for the spinal synergy by simultaneous consideration of passive structures and muscle forces under given posture and loads. Results satisfied kinematics and dynamic equilibrium conditions at all levels and directions. Net moments, muscle forces at different levels, passive (muscle or ligamentous) forces and internal compression/shear forces were larger in stoop lifts than in squat ones. These were due to significantly larger thorax, lumbar and pelvis rotations in stoop lifts. For the relatively slow lifting tasks performed in this study with the lowering and lifting phases each lasting ∼2 s, the effect of inertia and damping was not, in general, important. Moreover, posterior shift in the position of the external load in stoop lift reaching the same lever arm with respect to the S1 as that in squat lift did not influence the conclusion of this study on the merits of squat lifts over stoop ones. Results, for the tasks considered, advocate squat lifting over stoop lifting as the technique of choice in reducing net moments, muscle forces and internal spinal loads (i.e., moment, compression and shear force).

Extrusion foaming of semi-crystalline PLA and PLA/thermoplastic starch blends.

Abstract: Low density open-cell foams were obtained from polylactic acid (PLA) and from blends of PLA with thermoplastic starch (TPS) using CO(2) as a blowing agent. Two unexpected features were found. First, a 2D cavitation process in the fractured cell walls was unveiled. Elliptical cavities with dimensions in the 100-300 nm range were aligned perpendicular to large cell cracks clearly exhibiting 2D crazing prior to macroscopic cell rupture. Secondly, a significant crystallization rate increase associated with the CO(2) foaming of PLA was discovered. While the PLA used in this study crystallized very slowly in isothermal crystallization, the PLA foams exhibited up to 15% crystallinity, providing evidence that CO(2) plasticization and the biaxial stretching upon foam expansion provided conditions that could increase the crystallization rate by several orders of magnitude.

Frustrated 2D Molecular Crystallization

Abstract: Grafting isophthalic acid groups to linear connectors produces tetracarboxylic acids 2−4, which are extended analogues of trimesic acid (1). Normal pairwise association of -COOH groups induces trimesic acid to form a hexagonal network held together by six hydrogen bonds per molecule. In contrast, analogues 2−4 are designed to form two polymorphs, parallel network II and Kagome network III, which are linked by eight hydrogen bonds per molecule. The particular connectivity of these networks allows a smooth transition from one to the other without introducing discontinuities in hydrogen bonding. DFT calculations suggest that subtle differences in hydrogen bonding favor parallel motif II for short tetraacid 2 and Kagome motif III for long tetraacid 4, whereas the two motifs are closely similar in energy for intermediate tetraacid 3. These preferences were confirmed by using STM to image the adsorption of compounds 2−4 on graphite. 2D crystallization of tetraacid 3 is frustrated, presumably because the two mot...

Robust H∞ control of descriptor discrete-time markovian jump systems

Abstract: This paper considers the stochastic stability and the robust control of descriptor discrete-time systems with Markovian jumping parameters. In terms of linear matrix inequalities, a necessary and sufficient condition is proposed, which ensures a discrete-time descriptor Markovian jump system to be regular, causal and stochastically stable. A robust admissibility condition and a robust bounded real lemma are also developed. Based on these, a sufficient condition on the existence of a state-feedback controller which guarantees the robust admissibility and the performance is also given by employing the linear matrix inequality technique. A robustly stabilizing state feedback controller can be constructed through the numerical solutions of linear matrix inequalities. Finally, an example is provided to demonstrate the effectiveness of the proposed approach.

System level assessment of an optical NoC in an MPSoC platform

Abstract: In the near future, Multi-Processor Systems-on-Chip (MPSoC) will become the main thrust driving the evolution of integrated circuits. MPSoCs introduce new challenges, mainly due to growing communication through their interconnect structure. Current electrical interconnects will face hard challenges to overcome such data flows. Integrated optical interconnect is a potential technological improvement to reduce these problems. The main contributions of this paper are i) the optical network integration in a system-level MPSoC platform and ii) the quantitative evaluation of optical interconnect for MPSoC design using a multimedia application.

A 1-V CMOS Current Reference With Temperature and Process Compensation

Abstract: A 1-V current reference fabricated in a standard CMOS process is described. Temperature compensation is achieved from a bandgap reference core using a transimpedance amplifier in order to generate an intermediate voltage reference, VREF. This voltage applied to the gate of a carefully sized nMOS output transistor provides a reference drain current, IREF , nearly independent of temperature by mutual compensation of mobility and threshold voltage variations. The circuit topology allows for compensation of threshold voltage variation due to process parameters as well. The current reference has been fabricated in a standard 0.18-mum CMOS process. Results from nineteen samples measured over a temperature range of 0degC to 100degC , showed values of IREF of 144.3 muA plusmn 7% and VREF of 610.9 mV plusmn 2% due to the combined effect of temperature and process variations.

Design and Experimental Verification of Compact Frequency-Selective Surface With Quasi-Elliptic Bandpass Response

Abstract: A compact frequency-selective surface (FSS) having a quasi-elliptic bandpass response is presented in this paper. This was realized by vertically cascading substrate integrated waveguide (SIW) cavities. A single SIW cavity FSS has been fully studied and approximate analytical formulas are introduced to calculate its resonant frequencies. Two different resonances can be excited by a plane wave in the single SIW cavity FSS. According to theories of the cascading cavity filter and dual-mode filter, cross coupling can be realized in cascading SIW cavity FSSs, thus a compact FSS with a quasi-elliptic bandpass response is implemented. A Ka-band sample was fabricated by a printed circuit board (PCB) process. Experiments were carried out to validate this design method. Measured results are in agreement with predicted ones. The proposed quasi-elliptic FSS presents a number of advantages, namely, high selectivity, stable performance, and much reduced volume.

The ambivalence of ridesharing

Abstract: Ridesharing is quite a popular topic of discussion among transport authority personnel. It is perceived to be a viable alternative to classical modes of transportation, and receives a great deal of political support from transport planners. However, not much objective information is available on ridesharing behaviors. We use travel survey data to study the evolution of the ridesharing market in an urban area. Our study is based on data from four large-scale OD surveys conducted in the Greater Montreal Area (1987, 1993, 1998 and 2003). In the latest survey conducted in Montreal, car passengers were asked to identify the driver who gave them the opportunity to travel in this way. Their answers were classified according to the type of driver; for instance, a member of their household, a neighbor or a co-worker. We use this information to calibrate a model matching car passengers and car drivers belonging to the same household. This will be referred to as IHHR (intra-household ridesharing). Preliminary results reveal that approximately 70% of all trips made by car passengers are the result of IHHR. Furthermore, around 15% of those trips are questionable, in that they were exclusively generated for another individual’s purposes, consequently generating an additional trip for the journey back home. Moreover, this percentage increased over time. Objective data regarding ridesharing and its evolution in an urban area will undoubtedly help decision makers gain a clearer profile of this means of travel and help to realign attitudes on the issue.

Hydrostatic, Temperature, Time-Displacement Model for Concrete Dams

Abstract: This paper presents frequency domain solution algorithms of the one-dimensional transient heat transfer equation that describes temperature variations in arch dam cross sections. Algorithms are developed to compute the temperature T(x,t) , spatial distribution, and time evolution for the “direct” problem, where the temperature variations are specified at the upstream and downstream faces, and for the “inverse” problem, where temperatures have been measured at thermometers located inside instrumented dam sections. The resulting nonlinear temperature field is decomposed in an effective average temperature, Tm (t) , and a linear temperature difference, Tg (x,t) , from which the dam thermal displacement response can be deducted. The proposed frequency domain solution procedures are able to reproduce an arbitrary transient heat response by appending trailing temperatures at the end of thermal signals, thus transforming a periodic heat transfer problem in a transient one. The frequency domain solution procedure...

Optimal condition based maintenance with imperfect information and the proportional hazards model

Abstract: Condition based maintenance (CBM) is based on collecting observations over time, in order to assess equipment's state, to prevent its failure and to determine the optimal maintenance strategies. In this paper, we derive an optimal CBM replacement policy when the state of equipment is unknown but can be estimated based on observed condition. We use a proportional hazards model (PHM) to represent the system's degradation. Since equipment's state is unknown, the optimization of the optimal maintenance policy is formulated as a partially observed Markov decision process (POMDP), and the problem is solved using dynamic programming. Practical advantages of combining the PHM with the POMDP are shown.