Researchers discover methods to find 'needles in haystack' in data

Dec 05, 2005

A Case Western Reserve University research team from physics and statistics has recently created innovative statistical techniques that improve the chances of detecting a signal in large data sets. The new techniques can not only search for the "needle in the haystack" in particle physics, but also have applications in discovering a new galaxy, monitoring transactions for fraud and security risk, identifying the carrier of a virulent disease among millions of people or detecting cancerous tissues in a mammogram.

Case faculty members Ramani Pilla and Catherine Loader from statistics and Cyrus Taylor from physics report their findings in the article, "A New Technique for Finding Needles in Haystacks: A Geometric Approach to Distinguishing between a New Source and Random Fluctuations," December 2, in the journal, Physical Review Letters.

"As haystacks of information grow ever larger--and the needles ever smaller--the search for a signal becomes increasingly difficult to find using traditional approaches. There is a need for sophisticated new statistical methods," the researchers report.

Researchers working with large amounts of data encounter the fundamental problem of determining a real signal from random variation in the data. In many practical problems, a suspected signal may only be a small blip in a noisy experimental background.

The Case team discovered a technique that is built on the principle of comparing a set of summary characteristics for any sub region of the observations with the background variation. From these characteristics, attempts are made to find small regions that appear significantly different from the background--a difference that cannot simply be attributed to random chance.

"Methods used in high-energy particle physics problems traditionally have searched for any departure from a background model; that is, anything that is not a haystack," said Pilla, the project leader. "Our method efficiently incorporates information about the type of disorder expected, thereby enabling us to find the signal of interest more accurately."

At the core of the breakthrough is the idea of posing the problem in terms of a "hypothesis-based testing" paradigm to detect statistical disorder in the data. The method further exploits the flexibility behind a long-established geometric formula in creating a technique that significantly enhances the ability to distinguish a signal.

The researchers said the challenge is two-fold: defining efficient test statistics, and determining the critical cut-off. That is, to help the scientist find what is random variation as opposed to what is the signal. The detection problem involves a large number of comparisons, and the researchers caution that experimentalists should not be fooled into false discoveries by random variation.

"The experimenter wants to control the experiment-wise error rate: if there is nothing in the data, then there must be minimal probability of falsely discovering a signal. On the other hand, we want to maximize our chance of discovering any real signal that may be present in the massive data set," said Loader.

"The probabilistic problem associated with this scenario is reduced to one of finding the areas of certain regions on the surface of high-dimensional spheres," explains Pilla.

The Case researchers then exploit the geometric methods pioneered in 1939 by Harold Hotelling and Hermann Weyl. They tested the statistical techniques by using computer simulated particle physics experiments that mimic the real experiments conducted in colliders to demonstrate that the new technique significantly increased detection probabilities.

"In high-energy particle physics and astrophysics problems, chi-square goodness-of-fit tests are widely employed, although they have relatively low power to detect the signal," notes Taylor. "Through my collaborative work with Professors Pilla and Loader, we will be able to develop powerful statistical tests for detecting a signal from noisy data with high probability, a fundamental problem encountered in many scientific disciplines."

Taylor added that "conducting experiments in a particle collider may cost tens of millions of dollars. Improving efficiency in the analysis of experimental results can lead to enormous cost savings. Furthermore, we can obtain the same results with much smaller experiments, or effectively find much smaller departures from the background model."

"Detecting a real signal (the needle) present in random and chaotic data (the haystack) will lead to scientific success," conclude the researchers.

Source: Case Western Reserve University

Explore further: Generating Mobius strips of light: Researchers experimentally produce these structures from light polarization

add to favorites email to friend print save as pdf

Related Stories

Solving an organic semiconductor mystery

Jan 16, 2015

Organic semiconductors are prized for light emitting diodes (LEDs), field effect transistors (FETs) and photovoltaic cells. As they can be printed from solution, they provide a highly scalable, cost-effective ...

Predicting daily space weather will help keep your GPS on target

Dec 04, 2014

It's well known that severe space weather events – which are quite rare – can have a negative impact on our use of Global Positioning System (GPS) enabled devices. But our research, published in Geophysical Research Letters, shows that another form of space weather – which occurs on ...

Detecting defects in solar cells

Dec 02, 2014

Scientists at the National Physical Laboratory (NPL) have developed a new method for detecting defects in solar cells using a technique called 'compressed sensing'.

Recommended for you

Building the next generation of efficient computers

19 hours ago

UConn researcher Bryan Huey has uncovered new information about the kinetic properties of multiferroic materials that could be a key breakthrough for scientists looking to create a new generation of low-energy, ...

User comments : 0

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more

Click here to reset your password.
Sign in to get notified via email when new comments are made.