Machines can learn from examples and make predictions
in similar but unknown situations. Machine learning addresses statistics
problems
of regression, interpolation, density estimation and pattern
recognition . The applications overlap with KDD
applications
and include financial predictions, medical diagnosis, process control,
fraud detection, handwriting and speech recognition, and information retrieval
in text databases.
How can a machine learn?
Learning machines usually refer to computer programs that learn from examples.
Example of learning machines include artificial neural networks, decision
trees, and Support Vector Machines. Such learning machines have tunable
parameters that are adjusted using training examples to achieve a particular
objective (e.g. classify correctly objects into a number of classes). After
training, the learning machine is ready to make predictions on new unseen
examples.
The right learning machine for the right application
Experience is needed to preprocess data to facilitate the learning task
and select an appropriate learning machine. We have more than 15 years
of experience in designing learning machine architectures and algorithms,
particularly neural networks and kernel methods. We are co-inventor of
the widely used support vector machine technique. We work closely with
our customers to incorporate domain knowledge about the task in the architectural
design and constraints on the parameters. We provide visualization aids
and detailed reports to help understand the predictions made on new data.
Data is almost everything
Having good data is essential. We help our customers with their design
of experiments to optimally train and test your learning machine and
make them benefit from our experience in data collection, benchmarking,
data management and data cleaning.