A recent NY Times article highlights the issue of data overload. It points out some of the approaches gathering momentum to keep up with the pace of data given the enormous computational requirements. Modern algorithms are not just enormously complex to design they are ravenously processor hungry and require some clever juggling to compute within a reasonable cost. However approaches such as Hadoop (as pointed out in the article), HBase, Hypertable etc. offer a quantum leap in capabilities.
At Cognika we have been leveraging Hadoop extensively for many of our machine-learning activities. Lately we have developed some image-processing and feature detection algorithms which normally would require significant investments in hardware. However, w have managed to achieve comparable performance using commodity hardware and squeeze out some impressive results. Please post your experiences and/or lessons learned and we are happy to share ours.
Showing posts with label algorithms. Show all posts
Showing posts with label algorithms. Show all posts
Tuesday, October 13, 2009
Friday, January 16, 2009
"Biomimetics"? - why not just mimetics?
The field of Biomimetics or Biomimicry applies generated by nature through evolution, to solve scientific challenges. While inventions such as Velcro, Swarm algorithms etc. are some famous examples of technology inspired by nature, why not extend the idea to be inspired from multiple domains.
An example is the idea of TRIZ developed by a scientist in the former Soviet Union Genrich Altschuller. Altschuller postulated that solutions that exist in one domain (example: automobiles) could help solve a challenge in a very different domain (example: medicine). He found such cross-domain solutions using the patent corpus as a source. So biomimetics seems to be a specialized case of TRIZ that uses biological organisms an information source.
AskNature is a website that seems to be a great start at collecting such data and making biomimetics usable. It would be a good next step to release the data into public domain or expose as an API, so it can be integrated into other applications.
Overall we feel that the concept of mimetics should be extended across corpora - biological, patents, journals etc. Only then will the true power of this approach be realized.
An example is the idea of TRIZ developed by a scientist in the former Soviet Union Genrich Altschuller. Altschuller postulated that solutions that exist in one domain (example: automobiles) could help solve a challenge in a very different domain (example: medicine). He found such cross-domain solutions using the patent corpus as a source. So biomimetics seems to be a specialized case of TRIZ that uses biological organisms an information source.
AskNature is a website that seems to be a great start at collecting such data and making biomimetics usable. It would be a good next step to release the data into public domain or expose as an API, so it can be integrated into other applications.
Overall we feel that the concept of mimetics should be extended across corpora - biological, patents, journals etc. Only then will the true power of this approach be realized.
Subscribe to:
Posts (Atom)