 |
|
|
 |
 What's New?
On July 27, 2003 VAI celebrated its five-year anniversary. From our humble beginnings in 1998 to the fully staffed small business we have become, Visual Analytics has revolutionized the analytical industry with its state-of-the-art technologies. VAI will continue to provide the industry with the best and most robust analytical capabilities for many years to come.
Support Site/Login
Visual Analytics recently made some enhancements to our Support Site. These changes may affect your log in. Should you encounter a problem logging in to the Support Site, reset your password on the log in page using the "I forgot my password, please reset it for me" option. Your password will be reset to a random value and sent to you via e-mail. You can log in with this new password, but be sure to access your profile and change the password at your earliest convenience.
|
Since our last installment, Visual Analytics staff members attended several conferences including the Fraud Conference, hosted by the Association of Certified Fraud Examiners (AFCE) in Chicago and INTELINK in San Francisco. They also gave an important demonstration to key IRS executive staff members.
ACFE Fraud Conference
VisuaLinks was well received by the attendees at the ACFE Fraud Conference. Due to the "commercial" nature of this conference, many of the attendees had no previous knowledge of, or experience with, VisuaLinks -- although everyone was impressed with VisuaLinks and its capabilities. Their overall reaction was that VisuaLinks is a powerful and valuable tool for finding "fraudsters."
VisuaLinks was so well received that one of the presenters, after seeing a VisuaLinks demo, made some impromptu changes to his speech to add a discussion of VisuaLinks to his presentation. In addition to the interest displayed by the attendees, we were approached by some of the other vendors to discuss using VisuaLinks as a visual "front-end" to their own tools.
INTELINK Conference
VAI attended the INTELINK Conference, August 3-6, in San Francisco, CA demonstrating the leading-edge functionality of both VisuaLinks and Digital Information Gateway (DIG) to the US Department of Defense, US Intelligence Community, Military Commands, FBI, Canadian National Defense, System Integrators and valued partners. In addition, the Defense Intelligence Agency (DIA) Joint Intelligence Virtual Architecture (JIVA) program was demonstrated to illustrate how the robust capabilities of VisuaLinks address critical intelligence challenges including counterterrorism, counter-narcotics, and counterintelligence.
IRS Executive Staff Demonstration
On Wednesday, August 13, VAI staff helped deliver a demonstration of the IRS' Reveal system to members of the IRS executive team. Reveal is a new system, currently being piloted by the IRS' Criminal Investigations (CI) unit as well as their Philadelphia Lead Development Center (LDC), which features VisuaLinks as the core analysis tool for this important new system. In attendance were Mr. Bob Wenzel, Deputy Commissioner of the IRS (who will be retiring soon after 40 years of distinguished service to the IRS) and Ms. Dale Hart, Commissioner of SBSE (Small Business/Self-Employment), along with members of their staffs. Both Mr. Wenzel and Ms. Hart were impressed with the capabilities of VisuaLinks. Ms. Hart observed that, even though the Reveal system is intended for use by CI, it would be equally useful for SBSE as well. Reveal will be deployed for use nationwide by the IRS CI and LDCs in the next few weeks.
|
|
|
 |
continued from the cover
At the "10,000 foot level," classification is a neural network technique for creating output that separates data into categories or clusters that are characterized by distinct features. In much the same way that VisuaLinks presents explicit relationships through graphical displays, neural networks can also imply relationships in the data based on computations. An important difference is that neural networks are not able to justify how their results were established - which is why they are often referred to as "black boxes."
Neural networks are capable of producing clusters of data that can find specific breaks or segmentations within the data set, which may imply some type of relationship, but without the means of describing how the data are related. A neural network has no theoretical basis in the underlying application; it merely fits to a sample of data. Basically, the operator will ask the neural network for say, 10 clusters of data, without regard to how those clusters are actually created. Thus, when using neural networks there is no such thing as a "correct" output which means there are also no "incorrect" outputs.
Often, you will hear references to "training" a neural network. Basically, this is providing a number of positive and negative cases to reinforce the learning of the network. These cases represent situations that are already known, such as fraud reported on a credit card or a person exhibiting certain medical conditions. Once the system has learned the correct responses to the set of training inputs, the neural network can be used to detect and classify those patterns among new data. Thus, the training cases already represent "known" patterns and situations and the goal of the neural network is to determine whether "new" data fits one of these patterns.
The neural network does not "discover" the patterns; rather, it operates as an automated pattern classification system and does not learn any new patterns once training is completed. Needless to say, this type of offering is invaluable for well-defined industries that have fairly fixed or ridged data standards. Unfortunately, you cannot expect to obtain good models from inconsistent or incomplete data, which is exactly the type of data contained in datasets related to terrorism, money laundering, and even fraud.
Additionally, after a neural network has been trained, it must be tested to show that it properly reflects the intended pattern classification. This is often very difficult in complex environments as referenced above because there are no well-established patterns from which to train the neural networks in the first place. What does a terrorist look like? What does a money-launderer look like? These are hard questions to answer.
Neural networks also require that data be presented in the form of continuous values such as numbers (e.g., dollar amounts, call durations, etc.) as well as very limited, discrete values (e.g., hot/warm/cold, red/green/blue, etc.). Unfortunately, neural networks do not perform well on data that cannot be classified nicely into these categories (or bins). As such, databases solely containing the names or people, accounts, addresses, telephone numbers, identifications, facilities, events, and so forth are not good candidates for neural network systems.
To summarize some of the limitations of neural networks:
- Requires clean data to produce reasonable models for classification which is not readily available in the fraud, terrorism, and money laundering domains
- Not able to process large quantities of highly discrete-valued data such as names, addresses, accounts, etc.
- Unable to justify the basis of their outputs and results
- Often configured as black-boxes to monitor data for known or pre-existing patterns
- Patterns are defined based on known events (fraud on a credit-card, patient-x died) and are not truly "discovered"
- Requires a training set to be provided to reinforce learning
Visual data mining systems like VisuaLinks do provide the most flexibility for those situations in which users may not have a clear understanding of the scope of information contained within a data set, and therefore may not have specific hypotheses from which to begin an analysis. We fully expect that once well-qualified patterns are detected, confirmed, and consistently encountered using a system like VisuaLinks, we could comfortably pass them over to a neural network for additional value-added detection. Perhaps a future version of VisuaLinks will include a coupling to a neural network system.
Now, before the entire neural network Ph.D. community gets up-in-arms to provide counter arguments to the topics discussed, we want to re-emphasize, there is a time-and-a-place for using neural networks. In the world of counter-terrorism, money laundering, financial crimes, and various complex frauds - the use of neural networks do not represent the best technology for the resources it takes to reap a return on investment.
In today's post 9/11 marketplace, many companies have jumped on the "bandwagon" by stating that their products are ideal for detecting patterns of fraud, money laundering, terrorism, crime, and so forth. Detecting patterns of illicit types of activities is a difficult problem and not easily solved. Thus, there is a lot of confusion in the marketplace about the best approach and technologies for dealing with these kinds of problems. With large marketing budgets, smooth talking sales people, and intimidating loss statistics, some of the larger companies are hyping their software as being the miracle cure for all aliments. "Let the buyer beware" that not all data mining, visualization, link analysis systems are created equally.
Many company and government executives are under extreme pressure to make the best decision without fully knowing how the technologies operate or what the true value (ROI) is in what they can "realistically" provide to solve the underlying problems. As with any technology or tool, there is a proper time and place to use them; the key is knowing when.
|
|
 |
August 19-20, 2003
FBI ITEC Conference in Reno, Nevada
VAI will host a booth at the FBI ITEC Conference in Reno, Nevada, August 19-20. VisuaLinks and DIG enable the Bureau to significantly increase its intelligence discovery and dissemination capabilities and provides its agents with state-of-the art capabilities to combat terrorism, narcotics crimes, financial crimes, fraud, and organized and violent crime. Visit us in Booth 27 (mention "The LinkLetter" and receive a special gift).
August 24th-27th, 2003
ACM SIGKDD 2003, Hilton Washington Towers
VAI will host a booth at the Association for Computing Machinery’s Special Interest Group for Knowledge Discover and Data Mining
(ACM SIGKDD) Conference.
This conference is an influential event in the knowledge discovery and data mining industry.
Running concurrently with the SIGKDD conference will be
ICML-2003, the Twentieth International Conference on Machine
Learning, and COLT-2003, the Sixteenth Annual Conference on
Computational Learning Theory. VAI is looking forward to attending these highly technical and informational conferences.
September 7th-14th, 2003
21st Cambridge International Symposium, Cambridge, UK
VAI will be participating in the 21st International Symposium on Economic Crime in Cambridge,
England sponsored by the Jesus College, University of Cambridge and the Institute for Advanced Legal Studies,
University of London. This year's theme is Financial Crime, Terror and Subversion: The Control of Risk in a
Destabilized World Economy. VAI is presenting on September 11th on a plenary workshop on Drug Trafficking,
Financial Crime and Terrorism. For more details, visit www.crimesymposium.org.
|
|
 |
Testing Technician - one (1) position is available
This position will be expected to conduct product testing for the VAI tool suite, to include both VisuaLinks and DIG. The successful candidate will execute test scripts and scenarios, record results and report discrepancies to the development staff. The candidate may be required to write new test scenarios.
The successful candidate will demonstrate a background in structured testing with demonstrable experience in other testing environments.
Requirements: Two years of product testing experience.
Field Support Engineer - two (2) positions available
These on-site technical support engineers will support one or more DOD customers, including analytical and modeling support. Technical support responsibilities will require on-site support for VisuaLinks in the following areas: troubleshooting, model building, installation and configuration, performance analysis and tuning, training and to lend assistance to the database administration staff on matters concerning VisuaLinks.
Requirements: A BA/BS in Computer Science or 5 years of equivalent technical experience with large relational databases (Oracle, MS SQL Server, Sybase, Informix) Oracle experience is a plus. Scripting and network experience, especially troubleshooting, required. Data mining or visualization experience is helpful. Training experience is highly desirable. Current TS/SCI security clearance required.
Trainer and Technical Writer - one (1) position is available
This position will be expected to conduct on-site training for the VAI tool suite, to include both VisuaLinks and DIG. Additionally, the successful candidate will assist our technical writing staff in maintaining and extending our documentation program.
The successful candidate will demonstrate a background in technology training to include the creation of training objectives, lesson plans, presentation materials and end-user exercises. The candidate will also be required to demonstrate strong technical writing skills.
Requirements: A BA in English or a similar course of study, or 5 years of equivalent technical training and writing experience. The candidate must show at least two (2) years of professional experience as a technical trainer and technical writer. A portfolio of sample work would be helpful. Current TS/SCI security clearance is a plus.
All interested candiates should submit a resume to: info@visualanalytics.com
|
|
|
 |
|
