Home
  Journal   Friends   Calendar   User Info   Memories
 

Pankaj Mehra

4th August, 2007. 1:29 am. Fascinating interactions at Google Science Foo Camp!

I really enjoyed the talk by Charles Semonyi sharing his space tourism experience. Saw the $100 laptop PC, which is way cool. Learned about a program between MIT and schools around the world that allowed a bunch of teenagers to genetically engineer "eau d'E. coli." Saw Martha Stewart and Sergei Brin among the attendees.

But the most fascinating discussions I had were with a kindred technical soul, Jeff Jonas, SRD founder and now an IBM Distinguished Engineer. We have such resonating views on complex real-time information processing and event analytics that the discussion could immediately get to the guts of query processing and routing architectures and their relationship to the ingest-side processing.

Saw a couple of issues of Make magazine: this one I am going to become a lifetime subscriber to!

More after tomorrow's day-long sessions.

Current mood: pleased.

Make Notes

13th June, 2007. 9:15 am. Looking forward to the Google/O'Reilly Science Foo camp

http://wiki.oreillynet.com/scifoo07/index.cgi?

I was just going through the attendee list and noticed the excellent set of people I am going to run into for 3 whole days. I haven't prepared a topic yet but maybe we will talk about some work I have been doing in combinatorial methods of interconnect design, a monograph I am working on for the past 6 years in my copious free time.

Current mood: optimistic.

Make Notes

4th April, 2007. 10:30 pm. Great audience at my talk at International Semantic Web Conference in Atlanta




Industry 1: Managing Richly Connected Information

Pankaj Mehra

Current mood: pleased.

Make Notes

16th October, 2006. 5:44 am. Trying out Office 2007 beta

I installed Office 2007 beta and love it. Many good and productivity enhancing features from better menus to slide libraries. The TR2 fixed a few glitches I had experienced with Outlook and it has been very stable since I updated.

Make Notes

25th June, 2005. 10:40 pm. The second coming of AI?

If you want to ponder an evergreen question with an evergreen answer, try: What will we do with computers when CPU cycles and bandwidth are abundant and pervasive? I can think of at least three answers to this question that are evergreen:
1. Wait (a.k.a. latency)
2. Try to cheat reality by buying gobs of memory
3. Do AI

I will entertain #1 and #2 further only if someone disputes their existence, but #3 I find interesting. (watch out for that Yoda effect, doc!)

Except for database people who will beat about the bush and tell you things such as "distributed databases are here" or "we can build really smart Web services," most honest technologists will admit that AI is hear, again! It is bigger, better, more real, and far more pervasive than the expert systems and heuristic search programs of 70s and 80s. In fact, it is found in your cell phone (the word completion software ...), in your Office software (grammar checker ...) and in the new breed of unstructured information management software from the likes of IBM, MIT Media Lab, StoredIQ, StreamSage and MiTi. If you don't believe me, try http://www.campaignsearch.com or go visit Henry Lieberman's homepage at http://web.media.mit.edu/~lieber .

The "once burnt, twice shy" crowd is wondering whether the second coming is producing goods as fragile as the first coming did. Yes, I am refering to the brittleness of AI. The diehards like Doug Lenat never gave up and are finally claiming to have codified commonsense itself. (Remember Eliza?) I think this has been the disturbing side-effect of the much-idolized Turing Test that AI researchers did, and still do, pay too much attention to "looking smart" and less to "being smart."

The commonsense stuff sure looks smart. It is powered by deep look-smart technologies such as Latent Semantic Analysis. This is where my resistance starts to crumble for I am a great believer in the AI that has always worked ... I mean the stuff that Minsky wrote about in his early papers on reinforcement learning, and that the likes of Rich Sutton, Gerry Tesauro enriched during "the Neural 90s."

To me, the aha moment was when I read the JHU connectionists, esp. Paul Smolensky. They finally gave a neurologically plausible explanation of Newell's physical system hypothesis. This I consider a critical link between being smart and looking smart. That is, you can be smart if you compute like neural networks, and you can look smart if you pay attention to the stable states (attractors) of your neural dynamic systems, assigning them symbolic names.

In this sense, stuff that looks smart and is powered underneath by a statistical technique such as LSA feels credible. And Henry's demos will turn less skeptical minds into believers easily. I am watching this stuff for now, and trying to use it, too in my research. The skeptic in me still nags but I guess while computation and bandwidth feel abundant and pervasive, I will be doing AI, again!

Current mood: optimistic.

Make Notes

4th June, 2005. 7:13 pm. Bill Bryson's A Brief History about Everything

Highly readable science. I had no trouble convincing my daughter to pick up this work of non-fiction.

Make Notes

2nd June, 2005. 11:33 am. Congratulations to Professor Darrell Long UC Santa Cruz

This evening we celebrate Darrell Long's appointment as the first Kumar Malavalli chair. This is long overdue and well-deserved honor for Darrell, who keeps the flame of systems research alive. If I had to name a few thankless areas of research in computer science, I would list compilers, databases, filesystems, and operating systems at the top. A good thesis in these areas takes time, real effort, good experimentation technique, and a load of luck. Darrell has generated one good Ph.D. after another and built up the department at UCSC through leadership and example. My hats off to you, Darrell. CONGRATULATIONS!

Current mood: optimistic.

Make Notes