Like the first GWDAW, hosted by MIT in December 1996, GWDAW'98 was organized as a workshop. There were only a small number of short (typically 15 minutes) invited talks, each of which introduced or commented on an outstanding challenge in gravitational wave data analysis. Each talk was followed by extended, moderated discussion (typically 45 minutes) among all the participants. Each invited speaker was admonished by the organizing committee to speak to the future: to look, not backward to the achievements of yesterday, but forward to the challenges ahead and how they might be addressed.
To provide focus for the meeting, the organizing committee identified four different areas of study that would benefit from intense, focused discussion in a workshop setting. Speakers were chosen and the workshop was organized around the subjects of data diagnostics, upper limits and confidence intervals, LISA data analysis challenges, and collaborative data analysis.
The workshop's first morning began an orientation: a series of status reports given by representatives of the four different interferometer projects currently under construction -- GEO 600 (B. S. Sathyaprakash), LIGO (A. Lazzarini), TAMA 300 (N. Kanda and T. Tanaka), and VIRGO (A. Vicere') -- with an emphasis on their developing plans for data handling and analysis.
Following lunch, the participants turned to the discussion of data diagnostics: the use of the data channel itself as a diagnostic monitor of both the instrument's health and the usefulness of the signal channel for analysis. Dr. S. Vitale (University of Trento) described a new, automated data quality monitoring system that has been developed and installed on the AURIGA cryogenic acoustic detector. Two components of this system were of particular interest to the workshop participants: the first was the requirements it placed on the distribution of the signal channel output over hour-long intervals; the second was the requirement of consistency in the excitation amplitude of the antenna's two resonant modes.
Following this discussion, Dr. S. Mukherjee (Penn State) described the development of a Kalman Filter that can extract from the signal channel of an interferometric detector the amplitude and phase of the mirror suspension violin modes. Passing gravitational waves do not excite these modes, while technical mechanical noise sources that move the interferometer test masses do; thus, the suspension modes are a sensitive monitor of mechanical disturbances masquerading as gravitational waves. The wide-ranging discussion that followed included the use of the Kalman predictor as a means of reducing the data dynamic range and, possibly, the removal of the lines before analysis for gravitational wave signals.
The afternoon was rounded out by a presentation by Drs. N. Kanda (Miyagi University of Education) and D. Tatsumi (Institute for Cosmic Ray Research, University of Tokyo), who outlined of the TAMA detector calibration scheme and discussed some of the the effects of calibration error on the ability to reliably detect and identify the parameters of gravitational wave sources.
Despite their unprecedented sensitivity and bandwidth, there is no guarantee that the first generation of interferometric detectors will detect directly any sources. Nevertheless these instruments can -- at the very least -- set interesting and provocative upper limits on source strengths and rates. The second morning of the workshop was given-over to a discussion of the use of data from gravitational wave detectors to set upper limits.
Four talks punctuated a very animated discussion. The morning began with a tutorial, including several worked examples, on the construction of confidence intervals and credible sets, given by Dr. S. Finn (Penn State University). Two lessons were evident in this presentation: the first, that the analysis involved in the construction of upper limits and confidence intervals is as involved as that which goes into deciding upon detection or measuring parameter values, and the second, that the procedures for constructing confidence intervals and credible sets involve choices, and that these affect quantitatively the upper limits or confidence intervals derived.
Following this tutorial, Dr. S. Mohanty (Penn State University) described how, even in the absence of a known waveform, one can set upper-limits on the gravitational wave strength from gamma-ray bursts or other potential gravitational wave sources associated with an astrophysical trigger. After a lengthy discussion period, Dr. B. Allen (University of Wisconsin, Milwaukee) discussed an analysis of the LIGO 40M prototype November 1994 data set for binary inspiral signals. Focusing on the largest ``detected'' signal in the data set, this collaboration placed what might be termed an upper limit on the possible sets of upper limits that could arise from a more detailed analysis. Finally, Dr. M. Papa (Albert Einstein Institute) inaugurated a discussion of several different methods for searching for weak periodic gravitational wave signals in long stretches of data.
The afternoon of the first day focused on data analysis challenges associated with LISA: the proposed Laser Interferometer Space Antenna. LISA is currently being considered by NASA as a joint ESA/NASA mission, with a potential launch data as early as 2007. LISA has a much greater immediate potential for detecting directly gravitational waves from known astrophysical sources; however, the challenges of LISA data analysis are different than for ground-based interferometers and it is important to show now that LISA's promise can be realized through the analysis of the data it would collect. Four talks, spread over an afternoon of discussion, focused the workshop's attention on LISA's current status (Dr. S. Vitale, University of Trento), data analysis challenges (Dr. R. Stebbins, JILA and University of Colorado), ability to direct the attention of astronomers to imminent activity in different parts of the sky (Dr. A. Vecchio, Albert Einstein Institute), and what turns out to be the unimportance of gravitational lensing of cosmological sources for LISA (Dr. C. Cutler, Albert Einstein Institute).
The final session of the conference was devoted to an animated discussion on collaborative data analysis. There is a general agreement that the joint analysis of data from all simultaneously operating detectors will lead to more constraining upper limits, greater confidence in reported detections, greater measurement precision, and more information about observed sources. There are, however, practical impediments to collaborative data analysis. These include the different ``personalities'' of the experimental apparatus, deriving from their great complexity When different teams design, build, commission and characterize their instruments, where resides the common knowledge required to carry out an analysis that makes sophisticated use of the multiple data streams?
This problem has been tackled by the acoustic detector community and
some of the lessons they have learned were discussed by Dr. G.
Pizzella (University of Rome "Tor Vergata"). Other communities face
the identical problem: Dr. B. Barish (Caltech) described the mechanism
that the neutrino detection community has developed for data sharing.
Finally, Dr. R. Weiss (MIT) closed out the workshop with a final
presentation on the organization of data analysis in the COBE project
and his own perspective on the importance of cooperative data
analysis. Weiss identified what may be one of the more difficult
problems we face: the cultural transition from the model of the
scientist as an individual entrepreneur, who keeps hold of an idea and
the credit for it, to the collaborative model, where ideas are shared,
the community accepts the credit for the accomplishments, and the
participants take their reward from being part of the community.