These modules form our eXtreme Analytics^TM line of intelligent data analysis components.

eXtreme Sammon^TM
Our adaptation of the well-known Sammon Mapping projection, a powerful tool for visualization of high dimensional data. The Sammon Map everyone else has suffers from convergence problems, is slow to execute and produces a different mapping every time you run it. And if you want to add a point to the map, you have to start all over! We have radically enhanced the Sammon map to make it extremely fast, stable, and reproducible. Do you want to map your 78-dimensional data to 3D? 2D? 1D? No Problem! Do you want to add a point to find out where it will map? No problem! Want to add 20 more points and recalculate the map almost instantly? No problem!

eXtreme SOM^TM
Our very special implementation of Kohonen's Self-Organizing Map, ideally suited to clustering large data sets. You may have given up on the SOM before because it took so long to run the size map you needed. And since the SOM generates a different map, you may have had to run it a few times to get one you liked, making things even worse. Peer Science's SOM is very, very fast, so go ahead and build the map you wanted. And it produces the same optimal result every time, so you only need to run it once.

eXtreme Evolver^TM
Our Genetic Algorithm / Evolutionary System toolkit, capable of solving the most complex optimization problems. Peer Science's technical team has been applying Evolutionary Systems to real-world problems for over a decade. With every project we learned something new and built it into our Evolver toolkit. So you don't need to spend ten years catching up!

eXtreme NeuralNets^TM
Our special implementations of neural networks, useful tools for learning and identifying hidden patterns in your data. Our technical team members were pioneers in neural network development, publishing research papers and putting neural networks to work in government and industry. Over the years we've tried just about every version of neural networks and every trick that has been published, and some that have not. We've weeded out the dross and kept the silver, so you get the fastest, most stable neural network models that will give you the most useful results. For example, our backpropagation neural network model has a built-in fuzzy system that monitors and controls convergence. Internal controls speed convergence and guard against overtraining. Benefit from our experience, and avoid having to re-invent the neural network yourself.

eXtreme FeatureSelector^TM
A special implementation of our eXtreme Evolver^TM has been developed to select prediction feature subsets for pattern recognition systems. The unique (and proprietary) problem-solving techniques used in this product allow the user to select features by ranking of the best performing subsets discovered (without regard to number of features) or to specify the number of features to be used.

eXtreme Rules^TM
Our hybrid inference engine, processing rules that may have both fuzzy and binary components. This inference engine uses our proprietary universal fuzzy set representation, avoiding much of the complication of fuzzy system development and enabling the automatic generation of robust fuzzy systems from data.

eXtreme RuleRefiner^TM
Another special implementation of our eXtreme Evolver^TM has been developed to optimize a set of fuzzy rules for maximum performance and utility. A set of fuzzy rules is evolved from an initial set to improve the following characteristics:

• Small number of rules. A large number of rules can make a rulebase incomprehensible. Minimizing the number of rules without sacrificing significant performance yields a more transparent system with better generalization properties.
• Optimum fuzzy set description. Modifies the positioning of each fuzzy set to produces a low number of distinguishable fuzzy sets for each variable with little similarity. This process produces fuzzy sets that have clear semantic representation.
• Small number of premises in rules. Rules with more than a small number of inputs are difficult to interpret.

eXtreme Classifiers ^TM
Our speedy implementations of the classic K-Nearest Neighbor and Parzen Windows distance-based classifiers are useful for quick classification over moderately sized datasets. Since they learn in a single pass through the data, they can be rapidly implemented, or use them in concert with eXtreme NeuralNets to build more powerful combined classifiers. We have heavily optimized these classics for both speed and flexibility, and support multiple distance measures including Euclidean, Manhattan, L(n), Canberra, Bray-Curtis, and Mahalanobis.

eXtreme Cases^TM
Our Case Based Reasoning product is versatile and ready for web service implementation. This is an ideal product for supercharging your Help Desk, equipping your sales force with rapid, accurate Cost Estimation and Planning tools, or bringing new levels of performance to your Fraud Detection system.

eXtreme Search^TM
Until now, the ability to perform approximate, or "soft" matching on patterns containing heterogeneous data types was just a dream; now that dream has been realized. Matching is performed regardless of whether the data resides in a database, files, or computer memory.