Frame and fortune
Mining for trends at the help desk
Photography lessons
Fleet management made simple
Remembrance: Josef Raviv, 1934 - 1999
Frame and fortune
IBM Research's Pro/3000 digital imaging system is no stranger to the art world, having been employed to digitize collections of major museums. Now, the system has been enlisted in the tricky business of matching masterpieces of painting to suitable frames.
Museum buyers and private collectors come to Julius Lowy Frame & Restoring Co. of New York to have paintings mounted in costly frames that are themselves works of art. In the past, Lowy's customers had to hold a frame up to a painting, squint, and imagine how the two might go together. With a collection of more than 3,500 valuable antique frames, finding a match could be daunting. A frame must almost always be altered to fit the art, and the decision, once executed, is final. "Before cutting down or expanding a frame that costs six figures, customers want to have an accurate idea of what they are buying," notes IBM researcher Alan Cole. To make that possible, the image applications group at IBM's Thomas J. Watson Research Center has worked closely with Lowy to create an end-to-end solution integrating the Pro/3000 and special software.
Lowy's frames have been directly digitized, creating images that offer high resolution
and faithful color. Customers can now browse the huge frame collection electronically,
selecting by period, price and style. More important, once the system scans the artwork,
it can yield exact-scale renditions of the art and frame together, showing the frame as
it would actually be cut or extended to fit. "This system has completely changed the way
we do business," says Lowy president Larry Shar.-- Gil Bassak
More on Pro/3000 applications
Mining for trends at the help desk
Data mining -- the use of computers to find patterns in masses of corporate data -- has reached a new level of sophistication at two of the 400 PC help centers IBM operates around the globe. The centers in Kawasaki, Japan, and Raleigh, North Carolina, are trying out a "text mining" system that uses natural-language algorithms to sift through written summaries that call operators make after each call. Developed by a group of researchers under Tetsuya Nasukawa at IBM's Tokyo Research Laboratory (TRL), the system is unearthing trends in customers' questions and complaints. The result is greater customer satisfaction and lower-cost service.
Ordinary data mining simply looks for keywords, but the text-mining system -- dubbed TAKMI (an abbreviation for Text Analysis and Knowledge Mining but also a Japanese word meaning "skilled craftsman") -- spots grammatical relationships, as well. Knowing which word is the subject, which the verb, and which the object, TAKMI can categorize calls according to whether they are, say, complaints or questions and according to the product that is causing difficulty. For instance, the system recognizes that the statements "Win98 crashes Internet Explorer 4.5" and "Internet Explorer 4.5 crashes Win98" belong in different categories, even though the keywords are the same. To help analysts visualize the significance of its findings, TAKMI generates a color-coded chart that shows which problems are most common with which products.
In the current trials, TAKMI is cutting costs on two fronts. One is call volume. "Typically in the industry, a large help center might receive between 500,000 and a million calls a year, at a cost of $20 to $50 per call," says Koichi Takeda, project manager for text mining research at TRL. "Analysts at Kawasaki and Raleigh are catching problems early, or discovering others that were being overlooked," he says. "This information is helping them compile new FAQs [frequently asked questions], which are then posted on the appropriate Web page. And we know that FAQs reduce the number of calls." Call operators can also give customers new information that can shorten calls and make follow-up calls less frequent.
Other savings come from faster processing. Takeda says TAKMI enables analysts in Raleigh to pore through thousands of calls a week, instead of mere hundreds, as before. Hence problems and trends become evident sooner.
The TRL group is now working to broaden the use of TAKMI to encompass product development, planning and marketing.
-- John Boyd
Photography lessons
Several interactive touch screens at New York City's Museum of Modern Art are serving as a laboratory for studying how technology can help people learn. While museum visitors use the systems to explore images from MoMA's treasure trove of photographs, IBM researchers are busy observing how the visitors learn about art. The learning systems were developed by the cognitive human-computer interaction group at IBM's Thomas J. Watson Research Center, as part of a project called Discovery Learning. The project's goal, says manager Lauretta Jones, is to understand the ways people learn with computers, and to use that knowledge to develop new types of human-computer interfaces that make learning easier.
The MoMA project is an outgrowth of one Jones and her colleagues began in 1992 at the World's Fair in Seville, Spain, where IBM had placed 230 "Guest Services" kiosks around the grounds. "It was a tremendous laboratory," says Jones. "For six months, we had thousands of people using the system every day. We could watch how people used it, collect data and make changes."
Last year, when MoMA asked IBM to help the museum display its digital library of 30,000 photographic images, Jones saw another opportunity to study learning in general. The learning systems present users with a series of interactive challenges designed to encourage sorting and comparison of many different images. In one exercise, users are shown works by two photographers but not told which photos were created by whom. The task is to choose the photographers based on a visual analysis of other examples of their work. Another challenge is to identify whether a photo was originally made with fine art in mind or some other purpose. Yet another challenge is to decide which of three photographs does not belong with the other two.
Jones and her colleagues study the responses, often asking the visitors to describe their thought process while researchers take notes. The researchers will update the system every few weeks to study how the changes affect the human learning process. "Ultimately," says Jones, "we're looking to understand the space where people, learning and computer systems come together -- not in the traditional school context, but throughout life and in any environment."
-- Gary Taubes
Fleet management made simple
It's hard to imagine a bigger headache than the one posed by the "traveling salesman problem" -- the challenge of finding the shortest route to link many far-flung sites. But the job of a transport manager is actually more migraine-inducing: the salesman problem must be multiplied by a fleet of delivery trucks, taking into account the location of several warehouses and the time constraints of different customers for accepting deliveries. Transport managers should therefore be grateful to Kazuyoshi Hidaka, manager of the Optimization Simulation Project at the Tokyo Research Laboratory (TRL), who, along with Hiroyuki Okano and other researchers, has developed just the tools for the job.
IBM has begun marketing two TRL logistics applications, both based on heuristic computation (essentially, trial and error): Warehouse Location Planner (WLP) and Vehicle Routing Planner (VRP). Hidaka describes WLP as a "strategic decision-support tool for solving geographically distributed business tasks, such as delivering goods from several warehouses to a large number of customers." The tool proved itself when a Japanese manufacturer applied WLP to help decide where to relocate several warehouses. "They found a solution that achieved a 10 percent cost reduction in what they had initially decided," says Hidaka.
VRP is for helping geographically dispersed businesses make everyday operational decisions about, for example, how to dispatch a fleet of trucks to deliver goods from one or more depots to customers. "We can use VRP to find the optimal routes for trucks on the road network," says Hidaka. "We can minimize total travel time, taking into account constraints like vehicle capacity, number of vehicles, distance, time and customer demand."
Hidaka notes that the logistics of transport is especially difficult to control because it includes many variables that change from day to day. "But we can now make such problems an easily controllable task."
-- John Boyd
Rememberance: Josef Raviv, 1934 - 1999
The death of Josef Raviv and his wife, Joanna, in a car accident in New Zealand last October 13 was deeply felt by the IBM Research community. Raviv, a pioneer of Israel's high-tech industry, had served as director of IBM's Haifa Research Laboratory and had recently been appointed director of the newly formed Haifa Development Center.
Raviv was born in Poland in 1934 and moved to Israel at the age of six. After finishing his studies at the Herzliyah Gymnasium in Tel Aviv, he studied electrical engineering at Stanford University in California, and later earned a doctorate at the University of California at Berkeley.
Raviv joined IBM in 1964, as a researcher at the Thomas J. Watson Research Center, and went on to hold several management posts. In 1972, he returned to Israel to set up the IBM Scientific Center, in Haifa. "Through his leadership," says Paul Horn, senior vice president and director of Research, "he helped establish the concept of customer-oriented research." The center began with three researchers, and in 1982 became the Haifa Research Laboratory -- the largest IBM lab outside the United States. Today, the lab boasts 300 employees and 100 students. James McGroddy, former director of IBM Research, attributes the lab's success to Raviv, whom he calls "the guiding vision and the firm but gentle hand on the rudder of IBM's research lab in Haifa."
Among Raviv's significant technical accomplishments was a 1974 paper, "Optimal Decoding of Linear Codes for Minimizing Symbol Error Rate." The work provided the basis for what is known as the "backward/forward" algorithm, which has been widely applied in natural-language processing, language translation and speech synthesis. The IEEE Information Theory Society selected the publication for its Golden Jubilee Paper Award in 1998.
But it is Raviv's warmth that many IBMers will remember most. "Far more than his technical and management prowess, it was his personal care and affection for people that won all our hearts," recalls Randy Isaac, vice president for systems, technology and science at Watson. "He cared deeply about each individual at the Haifa Research Lab, as well as all those with whom he worked in IBM and in the industry."
