Howard Science Limited

dr.daniel.howard AT

+44 7532 395483

 Dear Friends, Valued Clientèle And Potential Customers:

 Our mission is to provide you with useful products, helpful and honest advice. 

 Daniel Howard, Prof. (Dr.)

CEO Howard Science Limited

 Our Products:

Here is an ISR (Intelligence, Surveillance, and Reconnaissance) example from 2002. It is a bespoke product that I built with research into Genetic Programming, detecting thousands of vehicles in IRLS infrared Tornado Reconnaissance imagery, it was validated by RAF (Royal Airforce) Image Analysts (Photographic Interpreters): click.

Analogy to Hinton’s Deep Learning.  Already nearly two decades ago was motivated almost by instinct and necessity to do what Hinton recently advocates that “labels are bad”.  I implemented “delaying the label” by various techniques. For example, in a paper co-authored with John Koza by means of multi-run and subtree encapsulation in Genetic Programming I delayed the label:

Simon C. Roberts and Daniel Howard and John R. Koza. Subtree Encapsulation Versus ADFs in Genetic Programming for the Even-5-Parity Problem. In Erik D. Goodman editor, 2001 Genetic and Evolutionary Computation Conference Late Breaking Papers, pages 359-365, San Francisco, California, USA, 2001.

Simon C. Roberts and Daniel Howard and John R. Koza. Subtree encapsulation versus ADFs in GP for parity problems. In Lee Spector and Erik D. Goodman and Annie Wu and W. B. Langdon and Hans-Michael Voigt and Mitsuo Gen and Sandip Sen and Marco Dorigo and Shahram Pezeshk and Max H. Garzon and Edmund Burke editors, Proceedings of the Genetic and Evolutionary Computation Conference (GECCO-2001), page 186, San Francisco, California, USA, 2001. Morgan Kaufmann.

Simon C. Roberts and Daniel Howard and John R. Koza. Evolving modules in Genetic Programming by subtree encapsulation. In Julian F. Miller and Marco Tomassini and Pier Luca Lanzi and Conor Ryan and Andrea G. B. Tettamanzi and William B. Langdon editors, Genetic Programming, Proceedings of EuroGP'2001, volume 2038, pages 160-175, Lake Como, Italy, 2001. Springer-Verlag.

Daniel Howard. Modularization by Multi-Run Frequency Driven Subtree Encapsulation. In Rick L. Riolo and Bill Worzel editors, Genetic Programming Theory and Practice, chapter 10, pages 155-171. Kluwer, 2003.

And also by being generous (high false alarm rate) and then exploring by means of a computational ant that resembled foveation strategy by the eye I in essence delayed the label. 

Daniel Howard. Innovating with Automatic Programming. Journal of Defence Science, 8(2):76-82, 2003 (received the annual accolade from the journal and its editor the then Chief Scientist of the MoD, Mike Markin OBE).

Daniel Howard and Simon C. Roberts and Conor Ryan. Pragmatic Genetic Programming strategy for the problem of vehicle detection in airborne reconnaissance. Pattern Recognition Letters, 27(11):1275-1288, 2006.

These techniques remain to be compared and contrasted to Deep Learning and their high potential remains largely unexplored. The first (multi-run modularization) has attracted the attention of noted computational economists.  It was cited in a recent address (see paper by Shu-Heng Chen and look for citation to Simon C. Roberts, Daniel Howard, and John Koza) in honour of Herbert Simon, late Nobel Prize winner in Economics:

An example of an Artificial Intelligence Solution based on a special implementation of Genetic Programming for a customer: click

 Medical Image Example: Organizing an image archive.  Here I develop a bespoke SONNET (Grossberg Carpenter) ART2 ANN (Artificial Neural Network) to achieve a self-organized image taxonomy of part of the central Sweden mammography image archive (Prof Lazslo Tabar, Falun Lassarett, Sweden).  It is done by textural statistics because it is texture that top radiologists use to detect the lesion (architectural distortions)  click.   click.

Description of this SME

We carry out bespoke solutions for software developers who wish to embed Artificial Intelligence technology into their web products, server side or back office engine offerings.