Speaker: Yossi Bar-Yosef

Ph.D. student under the supervision of Prof. Yuval Bistritz

Wednesday, June 14^th, 2017 at 15:00
Room 011, Kitot Bldg., Faculty of Engineering

Variational Maximum Mutual Information for Reducing Mixture Models

Abstract

The Gaussian mixture model (GMM) is a very powerful parametric modeling tool for representing complex data distributions. Consequently, GMMs are widely used in various statistical machine learning applications. In many cases, high-order mixtures (mixtures with large number of components) are used to get an adequate representation of the data, leading to heavy computational and storage demands. This often raises a need to approximate the high-order models by models with fewer components.

We propose a novel approach to this problem based on a parametric realization of the maximum mutual information (MMI) criterion and its approximation by an analytical expression named variational-MMI (VMMI). The suggested VMMI objective can then be maximized by analytically tractable algorithms that optimize the parameters of the requested reduced models. Differently from previous methods that produce a reduced model for each class independently, without considering the relations between the classes, the VMMI optimization produces models with improved discrimination ability - a desirable goal for any classification problem.

The use of VMMI optimization for GMM reduction was evaluated by experiments carried out with two speech related classification tasks:  phone recognition and language recognition. For each task, the VMMI-based parametric model reduction was compared to other state-of-the-art reduction methods. Experimental results show that the new discriminative VMMI optimization produces models that significantly outperform the classification ability of comparable reduced models obtained by standard non-discriminative methods.

School of Mechanical Engineering Seminar
Monday, June 12, 2017 at 14:00
Wolfson Building of Mechanical Engineering, Room 206

Micro-Gap Flow Obstructers for Enhanced Direct Chip Cooling

Amir Gorodetsky

M Sc. Student of Dr. Herman Haustein

Heat dissipation in modern high-power electronics require increasingly higher performance cooling, which traditional air-flow systems cannot provide and novel liquid-based solutions are pursued instead. The present study identifies several liquid flow patterns which significantly increase heat transfer, and examines them through simulations and experiments. It is well known that traveling vortices, due to generation of perpendicular unsteady or periodic flows, target a major inhibitor of heat transfer – the steady thermal boundary layer. Traditionally, micro-channels have been widely employed for micro-electronics cooling. However, for generating the desired vortices and a lower pressure drop, micro-gaps are more suitable – despite having lower heat transfer surface area, they also have reduced wall friction. The vortex laden flow generation and heat transfer can be further enhanced by inlet and local pulsation using active flow control methods.

Numerical DNS (transient laminar 2D) were conducted in order to converge to desirable experimental conditions. The experimental micro test-sections incorporated several novel aspects including a fully transparent system, a sputter-coated transparent heater, laser drilled and micro-CNC elements.

In addition, several novel measurement techniques were employed: High speed IR thermography, High speed microscopy, µ-PIV and image processing. In order to examine MEMS-based active flow control for mitigating local hotspots, a single slot jet system was also created, employing an electromagnetically actuated micro valve and enabling in-situ characterization.

Two main concepts were examined to achieve the desired wall temperature decrease and overall temperature uniformity in a micro-gap flow: bluff-body vortex generators or the addition of impinging slot-jets. The former is very suitable to micro scale flows as vortex shedding (von Karman street) nominally begins at low Reynolds numbers (Re>50). Present investigations showed that vortex shedding onset was only mildly suppressed by the confinement at the micro-scale (200 micron bluff body in a 600 micron channel), while heat transfer increased three-fold at the cost of a mild pressure drop increase. The latter concept improves heat transfer by imposing perpendicular flows – implemented by a row of slot-jets into the micro-gap cross-flow, a configuration previously only examined with a microchannel incorporating circular jets. A significant increase of heat transfer over a plain gap was found, as well as great improvement in wall temperature uniformity. In cooperation with a partner group performing more advanced 3D simulations an optimal configuration for heat transfer was obtained.

Regarding the active flow control aspects of the study, simulations showed that pulsating inlet provides only marginal gains in heat transfer, however local jet throttling experiments have been demonstrated as a potential promising alternative for future devices. 

School of Mechanical Engineering Seminar
Monday, May 29, 2017 at 14:00
Tokhna Building, Room 106

From Israel to Texas and back: 4-years of a petroleum journey

Dr. Nadav Sorek

Texas A&M University, College Station, TX, USA

On December 30, 2011, following the unprecedented gas reservoirs discoveries offshore Israel, the first Oil & Gas engineering program in Israel was launched. Experts from all over the world came to the Technion “to light the torch” of petroleum engineering knowledge in Israel. This presentation will tell the story of following some of these experts all the way to the American center of oil and gas, the state of Texas. The journey included pursuing Petroleum Engineering Ph.D. at Texas A&M University, while switching advisors and research topics, mastering and teaching the field of reservoir simulation and optimization, and spending three summer internships in the oil capital city, Houston. On a more technical note, we will discuss how conceptual modeling addresses key decisions for shale reservoir management and development of new techniques for optimizing conventional oil recovery from water-flooding using reduced order models and novel parameterization techniques. We will conclude with describing the experience in the industry, working with data analytics on hundreds of wells and a century of production data from oil reservoirs all over the world.

‏