Seminars Spring 2011
We have two Seminar Series: Brown Bag Seminars and RUTCOR Colloquia.
Brown Bag Seminars are usually held on Thursdays, 12:00-1:00 PM, in
the RUTCOR lounge. Pizza is served but feel free to bring your own
lunch. The Brown Bag Seminar Series usually hosts speakers from the
RUTCOR community. Unless otherwise noted, all the brown bag seminars
listed below will be held at 12:00 Noon in the RUTCOR lounge.
RUTCOR Colloquia are usually held on Thursdays, 1:30-2:30 PM, in
the Seminar Room or Room 139. The RUTCOR Colloquia Series features
speakers from outside the RUTCOR community. Unless otherwise noted, all
the RUTCOR colloquia listed below will be held at 1:30 PM in Room 139
in the RUTCOR building.
03/03/11 RUTCOR Colloquium - 3:00PM
(Note change of time)
Affiliation:University of Haifa
Title: Intersection Graphs of Paths on a Grid
Abstract: We investigate the class of vertex intersection graphs of paths on a grid, and specifically consider the subclasses that are obtained when each path in the representation has at most k bends (turns). We call such a subclass the Bk-VPG graphs, k ≥ 0. In chip manufacturing, circuit layout is modeled as paths (wires) on a grid, where it is natural to constrain the number of bends per wire for reasons of feasibility and to reduce the cost of the chip.
This is joint work with Andrei Asinowski, Martin
Limouzy, Marina Lipshteyn, and Michal Stern.
03/24/11 RUTCOR Colloquium - 1:30PM
Speaker:Dr. Giovanni Felici
Affiliation:Istituto di Analisi dei Sistemi ed Informatica
Consiglio Nazionale delle Ricerche
Title: Integer and Logic Programming for Knowledge Extraction from Large Data Sets
Abstract: Knowledge Extraction and Data Mining techniques are used
identify relevant patterns and models in large amount of data. Several
methods have been developed and successfully applied for a vast range
of applications. In this talk we focus on techniques that make use of
Integer and Logic Programming to formulate and solve two central and
strongly related problems: Feature Selection and Supervised Learning.
The method proposed for Feature Selection is based on integer formulations that represent the retained information using linear constraints associated with the observed data. Several variants of such model will be discussed; we analyze some formal properties of such variants, compare them with other methods proposed in the literature and describe an effective randomized heuristic algorithm that has been developed and successfully applied for the solution of large scale feature selection problems.
The Supervised Learning method is based on the extraction of separating logic formulas from training data and converts the learning problem into a sequence of instances of a Minimum Cost Satisfiability problem (MinSat), a well studied and hard optimization problem. These two methods, when combined, are capable of extracting very synthetic information with a high semantic value.
We conclude the talk discussing several additional characteristics of the proposed approach, and then describing their applications to two problems that arise in the analysis of DNA and genomic sequences: Micorarray Analysis, for which we show results on a recent study on neurodegeneration of transgenic mice, and Barcode Analysis, a recent project supported by the Barcode of Life Consortium funded by the Sloan Foundation, that aims at classifying the living species through a small fragment of mitochondrial DNA.
03/31/11 RUTCOR Colloquium - 1:30PM
Speaker:Dr. Lisa Hellerstein
Affiliation:Brooklyn Polytechnic University
Title: Min-Cost and Max-Throughput Boolean Function Evaluation
Abstract: We consider the problem of evaluating a fixed Boolean function f(x1,...,xn) on a given input vector (x1,...,xn), when evaluating each variable xi incurs a cost (or takes time) ci, and each variable xi is equal to 1 with probability pi.
Boolean function evaluation problems arise in a number
of fields, including databases (parallel filter ordering), artificial intelligence (satisficing search), and VLSI testing. We survey selected results on both Min-cost and Max-throughput versions of the Boolean Function Evaluation problem.
04/14/11 RUTCOR BrownBag - 12:00 Noon
Speaker: Dr. Michael Katehakis
Affiliation: MSIS, Rutgers School of Business
Title: NON-PARAMETRIC UP-AND-DOWN EXPERIMENTATION REVISITED
Abstract: Let Y(x) be a random variable such that
P(Y(x) = 1) = F(x) and P(Y(x) = 0) = 1- F(x)
where F(x) is a distribution function. It is sometimes of interest to estimate a given $\alpha-$ quantile of F(x) with observations distributed like Y(x) where the choice of x is under control.
Two approaches for obtaining solutions to this problem are the Robbins-Monro stochastic approximation and Derman’s non-parametric up-and-down procedure.
In recent work on the non-parametric up-and-down procedure we have modified the method by establishing a stopping criterion with predetermined confidence probability of correct detection. Further we have established worst case performance bounds and convergence rates.
In this talk we also discuss applications in inventory control under unknown demand distribution, tailored testing, drug dosage determination.
This is joint work with Cyrus Derman, Columbia University