From etsb at amath.washington.edu Sat Oct 9 18:26:19 2010
From: etsb at amath.washington.edu (Eric Shea-Brown)
Date: Tue Jun 12 13:43:54 2018
Subject: [Amath-seminars] Mon Oct 11: Optimization seminar by Burjorjee
Message-ID: <8C9AA93A-378C-47ED-B6E9-5627D1D53D90@amath.washington.edu>
Keki Burjorjee (Amazon.com)
Optimization by hyperclimbing: A new theory on the workings of genetic
algorithms
Research seminar
Monday, October 11, 2010
3:00pm, 403 Paul G. Allen Center for Computer Science & Engineering
Abstract
Genetic algorithms (GAs) have been used to obtain "quick-and-dirty"
solutions to a wide range of optimization problems, including high
dimensional ones with rugged, dynamic, and even stochastic cost
functions. Unfortunately the adaptive capacity of these algorithms is
currently not well understood. In this talk I will present the
hyperclimbing hypothesis, a new explanation that I proposed in my
doctoral thesis. A hyperclimbing heuristic works by decimating a
search space, i.e. by progressively limiting sampling within a series
of nested subsets. At each step, this heuristic sifts through
potentially vast numbers of coarse partitions of the subset it
"inhabits", and identifies ones that partition this set into subsets
whose expected fitness values are significantly variegated. The
hyperclimbing hypothesis holds that GAs work by implementing
hyperclimbing extraordinarily efficiently. Proof-of-concept evidence
for the hyperclimbing hypothesis comes from the use of a novel
analytic technique involving the exploitation of "algorithmic
symmetry". In addition to this evidence, I will present the results of
a relatively demanding test of validity-one involving the application
of a genetic algorithm with a simple tweak to large, random instances
of an over-constrained MAX-3SAT problem.
http://www.cs.washington.edu/htbin-post/mvis/mvis?ID=984
From akb6 at washington.edu Mon Oct 25 17:56:40 2010
From: akb6 at washington.edu (Andrea Barreiro)
Date: Tue Jun 12 13:43:54 2018
Subject: [Amath-seminars] CANCELLED: Boeing Distinguished lecture - October
28 - James D. Murray
References:
Message-ID: <4D4FFED6-4C82-41F5-999D-B81BADBC75A3@washington.edu>
Dear all,
I'm sorry to be the bearer of bad news but our first Boeing lecture of the quarter has been cancelled;
Prof. Murray has had to cancel his trip to Seattle next week.
Our next speaker will be Jim Keener on November 4; look for more details from me soon.
Andrea
Begin forwarded message:
> From: Andrea Barreiro
> Date: October 22, 2010 11:49:17 AM PDT
> To: amath-current@amath.washington.edu, amath-seminars@amath.washington.edu
> Subject: Boeing Distinguished lecture - October 28 - James D. Murray
>
> Dear all,
>
> We will open our Boeing lecture series next Thursday with a familiar face: James D. Murray, professor emeritus of AMATH and author of a favorite math biology text. See you there!
>
> --------------------------------------------------------------------------------------------------
>
>
> Boeing Distinguished Lecture Series
>
> James D. Murray (Princeton University, University of Washington, and University of Oxford)
>
> Vignettes from a Mathematician?s Odyssey in Biology and Medicine:
> from pilot ejection injuries to the benefits of cannibalism
>
> Thursday, October 28, 2010
> 220 Guggenheim, 4:00 pm
> (Reception to follow on 4th floor of Guggenheim)
>
> Abstract:
> I shall describe a series of problems for which the models are both practical and the mathematics simple. Pilot ejection can cause extreme stresses on the spine which frequently result in major injuries. A basic model of the physical process shows how such injuries could occur. Many diseases require a careful assessment of medication level: liver disease is a prime example. I shall describe the medical dilemma and show how the model provided a scientific method for determining the appropriate medication level. It is now used in England. Breathalysers are notoriously inaccurate. I shall describe two models which show that it is not possible to design a better breathalyzer. Prostate specific antigen (PSA) blood tests are considered indicators of possible prostate cancer but are notoriously misunderstood and misinterpreted. Our model explains a known anomaly and highlights inadequacies in current government guidelines. Cannibalism is widespread in nature and generally of benefit to the species. I shall describe a model for salamander cannibalism and then describe how important cannibalism has been for humans. Finally I shall show how certain spatial patterns in development are influenced by geometry and scale and, time permitting, how the woods on Bainbridge Island can exhibit jungle like aspects.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From akb6 at washington.edu Mon Nov 1 10:11:08 2010
From: akb6 at washington.edu (Andrea Barreiro)
Date: Tue Jun 12 13:43:54 2018
Subject: [Amath-seminars] Boeing Distinguished lecture - November 4 - Jim
Keener
Message-ID:
Dear all,
We will open our Boeing lecture series this Thursday with Jim Keener. See you there!
-----------------------------------------------------------------------------------------------
Boeing Distinguished Lecture Series
James Keener (University of Utah)
Mechanisms of length regulation of flagella in Salmonella
Thursday, November 4, 2010
220 Guggenheim, 4:00 pm
(Reception to follow on 4th floor of Guggenheim)
The construction of flagellar motors in motile bacteria such as Salmonella is a carefully regulated genetic
process. Among the structures that are built are the hook and the
filament. The length of the hook is tightly controlled while the
length of filaments is less so. However, if a filament is broken off
it will regrow, while a broken hook will not regrow.
The question that will be addressed in this talk is how Salmonella
detects and regulates the length of these structures. This is related
to the more general question of how physical properties (such as size
or length) can be detected by chemical signals and what those
mechanisms are.
In this talk, I will present mathematical models for the regulation
of hook and filament length. The model for hook length regulation is based on the hypothesis that the hook length is determined by the rate of secretion of the length regulatory molecule FliK and a cleavage reaction with the gatekeeper molecule FlhB. A stochastic model for this interaction is built and analyzed, showing excellent agreement with hook length data. The model for filament length regulation is based on the hypothesis that the growth of filaments is diffusion limited and is measured by negative feedback involving the regulatory protein FlgM. Thus, the model includes diffusion on a one-dimensional domain with a moving boundary, coupled with a negative feedback chemical network. The model shows excellent qualitative agreement with data, although there are some interesting unresolved issues related to the quantitative results.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From akb6 at washington.edu Tue Nov 30 10:53:01 2010
From: akb6 at washington.edu (Andrea Barreiro)
Date: Tue Jun 12 13:43:54 2018
Subject: [Amath-seminars] Boeing Distinguished Lecture - Doug Martin,
UW - Dec. 2
Message-ID: <4B7B547D-2960-4A18-BCFA-DC7DAA52491D@washington.edu>
Dear all,
Our final Boeing lecture of the quarter will be this Thursday, Dec. 2, at 4 pm, featuring a member of the UW Statistics department. See you all there!
- Andrea
-------------------------------------------------------------------------------------------------------------------------
Boeing Distinguished Lecture Series
Doug Martin (University of Washington)
Utility functions, risk measures and tail risk management
Thursday, December 2, 2010
220 Guggenheim, 4:00 pm
(Reception to follow on 4th floor of Guggenheim)
We first briefly review the classical expected utility maximization and Markowitz mean-variance optimization approaches to portfolio construction that date back to 1944 and 1952 respectively. Then we describe a set of modern coherence axioms for risk measures that were first introduced 1997, and since then have become a focal point for research on new risk measures and their applications to portfolio optimization and risk management. The industry standard Value-at-risk (VaR) fails to be coherent, while expected tail loss (ETL) is a highly informative coherent tail risk measure. We show that replacing the classical mean-variance approach with mean-ETL portfolio optimization can lead to quite superior portfolio performance. However, risk budgeting is an attractive transparent alternative to black-box portfolio optimization that becomes particularly attractive when based on a tail risk measure such as ETL. The key to portfolio construction with tail risk measures is a combination of: (1) Euler?s formula for obtaining an additive risk decomposition, and (2) reverse mean-ETL optimization to obtain implied returns assuming portfolio weights are optimal. The combination of percentage risk contribution and miss-match between implied returns and forecast returns allows for a transparent and effective approach to portfolio rebalancing, as is illustrated by considerable reduction in loss of a fund-of-hedge funds portfolio during the financial markets melt-down of late 2008 and 2009. Finally I offer a few comments on re-visiting expected utility maximization and on using utility functions to construct so-called spectral risk measures.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From akb6 at washington.edu Thu Dec 2 13:09:37 2010
From: akb6 at washington.edu (Andrea Barreiro)
Date: Tue Jun 12 13:43:54 2018
Subject: [Amath-seminars] REMINDER - Boeing Distinguished Lecture - Doug
Martin, UW - TODAY, Dec. 2
Message-ID: <73D1C302-4F45-4A1C-8984-63BADB03476E@washington.edu>
Dear all,
Just a reminder that our final Boeing lecture of the quarter will be TODAY, Thursday, Dec. 2, at 4 pm.
There will be a reception afterwards on the 4th floor of Guggenheim.
- Andrea
-------------------------------------------------------------------------------------------------------------------------
Boeing Distinguished Lecture Series
Doug Martin (University of Washington)
Utility functions, risk measures and tail risk management
Thursday, December 2, 2010
220 Guggenheim, 4:00 pm
(Reception to follow on 4th floor of Guggenheim)
We first briefly review the classical expected utility maximization and Markowitz mean-variance optimization approaches to portfolio construction that date back to 1944 and 1952 respectively. Then we describe a set of modern coherence axioms for risk measures that were first introduced 1997, and since then have become a focal point for research on new risk measures and their applications to portfolio optimization and risk management. The industry standard Value-at-risk (VaR) fails to be coherent, while expected tail loss (ETL) is a highly informative coherent tail risk measure. We show that replacing the classical mean-variance approach with mean-ETL portfolio optimization can lead to quite superior portfolio performance. However, risk budgeting is an attractive transparent alternative to black-box portfolio optimization that becomes particularly attractive when based on a tail risk measure such as ETL. The key to portfolio construction with tail risk measures is a combination of: (1) Euler?s formula for obtaining an additive risk decomposition, and (2) reverse mean-ETL optimization to obtain implied returns assuming portfolio weights are optimal. The combination of percentage risk contribution and miss-match between implied returns and forecast returns allows for a transparent and effective approach to portfolio rebalancing, as is illustrated by considerable reduction in loss of a fund-of-hedge funds portfolio during the financial markets melt-down of late 2008 and 2009. Finally I offer a few comments on re-visiting expected utility maximization and on using utility functions to construct so-called spectral risk measures.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: