Tuesday, May 24, 2005

Neuro signal detection chip

Startup that makes chips that detect neural activity and eye movements with starter applications in drowsiness detection for automotive industry (and much more to come I'm sure)... NeuroSky

Friday, May 20, 2005

On Intelligence

Jeff Hawkins book 'On Intelligence': Great insights. Quotes John Searle and his Chinese Room (To Read: The Rediscovery of the Mind)!

Chapter 1 (Artificial Intelligence): Background... Cornell EE -> Intel. But heart with understanding the brain. Declined support to proposals in Intel because AI was 25 yrs behind (and rightly so). Tried MIT but they were blind to the 'understanding brain' approach. Persisted with taking correspondence courses in biology and wrote exams and finally got grad admission into UC Berkeley in biophysics! Summary... AI is the wrong approach, Chinese Room argument sums it up. Understanding how brain works is fundamental. The means not the end is what is important and needs research. So far everything has been based on replicating the 'ends' in a 'behavioral' sense.

Chapter 2 (Neural networks): Neural nets are examples of right approach, although trends were not satisfactory. They are to the brain what transistor based radios are to computers. Handspring not based on neural nets (although still mysterious on what it is based on).


Relevant Courses: Neurobiology, Statistical Learning, Probabilistic Models In AI
Jeff Hawkins's SCPD talk here

Indian software engineers and cost dynamics

Competitive Software Through an India Lens
Very interesting insights on Indian software programming cost effectiveness by Aditi's CEO Pradeep Singh.

Math Instinct

Math Instinct: Keith Devlin's program on in-built mathematics in humans, animals and even plants. Golden ratios in pyramids, leaf rotations in plants for maximal sunlight exposure. How we and dogs curve to catch a ball in order to maintain a straight line for the perceived ball trajectory.

Thursday, May 19, 2005

Striping

Lecture on Fast I/O with multiple disks:
  • Knuth's Gilbraith? principle. Superblock striping.
  • Ripple shuffle.
  • Randomized Striping

Friday, May 06, 2005

Google's Cinco De Mayo '05 open house...

Google's Cinco De Mayo open house. Don McMilan of technicallyfunny.com was a riot. MapReduce being used extensibly in Google for its fault tolerant and parallelness. Keyhole has other UI challenges (Lars Rasmussen, Mark Aubin) (Topics: Javascript, multi resolution rendering), Google Print has mechanical challenges and space and compression challenges (Joseph O' Sullivan) (Topics: Image processing, OCR, compression). Search quality has its own set of daily evolving challenges (Amit Singhal) (topics here relate to Information Retrieval). Larry says there are always only 20% of things that they are working and want to work on the rest as well and restricted by good people resources. We let employees work on their hobbies that make their mothers happy :) Urs Hoelzle the Hardware Platform guy best quip if the big companies don't solve our heating and energy problem we're going to have to solve it ourselves :) Same with the network routers that get prohibitively expensive when you scale. Rob Pike of Log analysis mentioned their language used for analyzing a record (and how this is so as to enable parallelism on multiple records). The log-language-programs emit data that they are programmed to emit while parsing. This feeds into independent aggregators that interpret the emitted data.