•   Terence Yeoh
• The Aerospace Corporation
  

•  Failure Analysis
• Microfocus X-ray System
• Computed tomorgaphy

•  Used array facilities and existing scripts
  

•  Compute sinogram using scipy array packages
  

•  Energized states of matter with radically different chemical, electrical, and physical attributes
• have non-linear voltage spikes that appear random
• Need statistical analysis of events
• Tin whisker electrical shorts can develop into sustained plasmas
• When these develop they can kill a spacecraft
• They were tasked with evaluating space shuttle risks
• Initial assessment by NASA indicated low risk
• Testing showed risk much higher

•  Wavelet decomposition breaks the complex waveform into manageable bites
• freq/time maps
• separate random noise without throwing out noise envelopes
  

•  Used for automated signal processing
• clustering algorithms are necessary for sorting data
  

•  Delegate.py
• forking code
• PyWavelets
• wavelet transforms
• Psyco
• "JIT" compiled Code
  

•  Single system image clustering solution
• http://www.openmosix.org
• soon to be shut down
• Cluster aces as one large multiprocessor system
• Useful for Monte Carlo simulations or GA code
  

•  specify population size, islanding, migrants
• determine fitness function
• iterate over generations, specify exit criteria
• Fitness function is written in python
• They have used this code for automated alignment
• remove jitter from slice to slice
• use to be done by hand

•  A specializing compiler for python
• Useful for accelerating scipy's fitness function
• speedups come at the cost of memory space
• compiled functions do not relase memory until after the program is complete

•  all popular wavelets supported
• supports all useful transorms
  

• inherently random events
• best suited for wavelet analysis
• Neural networks did not seem to work
• They want to do arbitrary waveform analysis
• take plasmas regardless of media and find the most important features

•   Titus Brown
• http://ged.cse.msu.edu
  

• regulatory genomics intro
• a bioinformatics solution
• project maintenance
• sociological considerations
• future plans
  

• Same materials, different architecture
  
• regulatory regions control gene expression

• regulatory regions
• binding sites
• seem impossible to find
• genomes are big, regulatory regions are small
• no obvious statistical signature unlike protein coding regions
• no good way to test predictions without doing experiments
• slow, expensive, difficult

•  look for regions conserved between two or more species
• can use very simple sequence matching algorithm
  

•  Use for displaying data
• correlations done by eye
  

•  Allows system that lets biologists:
• establish sequence analyses with custom params
• run them on someone else's computer server
• visualize and interact with the result via a client GUI
• Aimed at bench biologists
• Intended to be extensible
• all open source
• use python, linux, Quixote (web server)
  

•   James Taylor
• NYU
  

•  http://g2.bx.psu.edu
• an open-source framework for integrating tools and databases into a workspace
• a web-based serve we provide integrating many popular tools for comparitive genomics

•  HTML inputs generated from abstract parameter description
• user doesn't write HTML
  
• tool help generated from a simple text format
• template for generating command line from parameter values
• output datasets generated by tool
• functional tests to be run with the full stack in place
• running functional tests for a specific tool on the command line
• test results on the command line and in an HTML report
• template language for building complex operations
• conditional groups, grouping constructs can be nested
• describe interface at a very high level
• command line tool expects a configuration file
• configuration file is generated based upon user input

•  flexible exection environment
• jobs are submitted as JobRunner
• Pluggable queuing policies
  

•  RGenetics
• http://regenetics.org

•  mapping "reads" onto protein databases
• need to do much faster

•   Gregory Benson and Alexey S. Fedosov
• http://www.cs.usfca.edu/river
  

• RIVER 

•  Reliable Virtual Resources
• Python framwork for parallel and distributed programming
• prototype parallel programming systems
• River Core
• Discovery
• Process naming and creation
• message passing
• state management
• River Extensions
• RPC/RMI, Trickle, MPI, MapReduce
• Benefits
• small, easy to use core interface
• written in python
• dynamic typing for rapid prototyping
• python goodies
• heterogeneous(python as vm)
• state capture at language VM level
• integrated check point migration
• Motivation
• parallel programming is still hard
• the future: more cores, largest clusters
• apps will have to utilize multiple processors
• apps will have to tolerate failures
• find the next set of programming models
• incrementally improve current models

• Extend python's rapid development capabilities to parallel systems
• facilitate short design/development cycle
  

•  Virtual machines (VMs)
• Python + River Core
• Virtual resources (VRs)
• Named with UUIDs
• code, data, thread, and message queue
• discover/allocate/deploy
• flexible code execution
• super flexible messaging
• Only need to specify a destination
• pass arbitrary attributes and values
• must be able to be pickled
  
• state management
• designed from the beginning
• encapsulate local state in VR
• only hooks to outside UUIDs
• per-VR queues hold in-transit messages
• Transparent migration and check pointing
• internal and external support
• Coordinated checkpointing
• algorithm is extensible

•  One net thread, one control VR
• control handles VR create, state
• TCP-based, connection caching (scalable)
• broadcast-based discovery
• super flexible messaging
• queue matching
• serialization (Pickle)

• Use stackless python
• CPyhon without the execution stack 
• Keep soft VR state separate from hard VR state
• two Vr classes: VR and VRI (internal)
• VR has a reference to host VRI
• Stackless: run VR as a tasklet in a VRI thread
• generate atomic system calls (VRI calls)
• state capture: unlink VRI reference from VR

•  remote access and invocation (RAI)
• RPC, RMI, and remote data access
• Create and access functions, objects, data on remote VRs
• built on top of the River core

•  simple task framing language
• put code/data on remote VMs
• execute sequentially or in parallel
• parallel invocation
• fork/join paradigm
• dynamic scheduling

•  partial implementation of MPI 1.2 in River
• most p-to-p and collectives
• easy to read and understand
• model C MPI interface
• experiment with different algorithms
  

•   Christfried Webers
• http://elefant.developer.nicta.com.au/
• Statistical Machine Learning Group
• NICTA
• Austrailian Gov
  

•  task
• training
• use training data sets
  
• testing
• use test data sets
• supervised learning
• use labels to help the learning algorithm
• does not need to deal with large data sets
• does need to deal with large data structures

•   Ondrej Certik

• Density Functional Theory calculations
• Scipy nonlinear solvers
• Finite Element Method using python-petsc and libmesh
• SymPy - the symbolic manipulation package in Python

Slides on the scipy web site

• Eric Jones
  

•  Scientists often know how to module their problems in software
• Its exploring the algorithms that is hard
• Need to make the exploration easier
• Enthough has more ideas than solitions
• Use Traits to make algorithms interactive
• complex problems, simple algorithms
• oil drilling
  
• only have seismic data to work with and what geoscientists have told them
  
• given physics for every type of area.
• simple formulas for each area
• combine simple formulas to build up algorithms
• Have created a stochastic modeling tool
• Approach to providing a better analysis
• code blocks are sets of executable instructions
• blocks are executed in contexts
• contexts with events
• events fire when data changes
• interacting with a variable
• dependency analysis
• how does variable change effect model
  
• create a "shadow" context to refer back to the original context for all static values
• function context accepts only functions
• data context excepts everything else
• prevents data context from getting cluttered

•   Bill Smith

•  treat text as data
• text in the sense as literature
• gaining prominence
• commercial tools include tools

•  I want to know everything about gene xyz
• name is a central identifier and search key but
• names are really complicated
• origins in other species, environments
• biologist's sense of humor
• duplicate common words
• ambiguous/nonlinear
• small differences have a large impact on meaning
• problem with simple lexical searches

 Offer capability to automate the expansion of a scientist's literature search.  Increase recall --maintain precision.

• Develop good set of synonymous protein names
• extract all citations
• extract articles and abstracts
• do fuzzy search
• count up and rate the matches

• Couldn't do this with simple regex searching
• Developed a multiple-pass approach
• index source test
• search test
• tabluate

• This is a parallel problem

•   Sargis Dallakyan
• http://mgltools.scripps.edu
  

•  Multiple tools available