IBM Watson Deep Learning: A Great Leap Forward
August 16, 2017
I read in the IBM marketing publication Fortune Magazine. Oh, sorry, I meant the independent real business news outfit Fortune, the following article: “IBM Claims Big Breakthrough in Deep Learning.” (I know the write up is objective because the headline includes the word “claims.”)
The main point is that the IBM Watson super game winning thing can now do certain computational tasks more quickly is mildly interesting. I noticed that one of our local tire discounters has a sale on a brand called Primewell. That struck me as more interesting than this IBM claim.
First, what’s the great leap forward the article touts? I highlighted this passage:
IBM says it has come up with software that can divvy those tasks among 64 servers running up to 256 processors total, and still reap huge benefits in speed. The company is making that technology available to customers using IBM Power System servers and to other techies who want to test it.
How many IBM Power 8 servers does it take to speed up Watson’s indexing? I learned:
IBM used 64 of its own Power 8 servers—each of which links both general-purpose Intel microprocessors with Nvidia graphical processors with a fast NVLink interconnection to facilitate fast data flow between the two types of chips
A couple of questions:
- How much does it cost to outfit 64 IBM Power 8 servers to perform this magic?
- How many Nvidia GPUs are needed?
- How many Intel CPUs are needed?
- How much RAM is required in each server?
- How much time does it require to configure, tune, and deploy the set up referenced in the article?
My hunch is that this set up is slightly more costly than buying a Chrome book or signing on for some Amazon cloud computing cycles. These questions, not surprisingly, are not of interest to the “real” business magazine Fortune. That’s okay. I understand that one can get only so much information from a news release, a PowerPoint deck, or a lunch? No problem.
The other thought that crossed my mind as I read the story, “Does Fortune think that IBM is the only outfit using GPUs to speed up certain types of content processing?” Ah, well, IBM is probably so sophisticated that it is working on engineering problems that other companies cannot conceive let alone tackle.
Now the second point: Content processing to generate a Watson index is a bottleneck. However, the processing is what I call a downstream bottleneck. The really big hurdle for IBM Watson is the manual work required to set up the rules which the Watson system has to follow. Compared to the data crunching, training and rule making are the giant black holes of time and complexity. Fancy Dan servers don’t get to strut their stuff until the days, weeks, months, and years of setting up the rules is completed, tuned, and updated.
Fortune Magazine obviously considers this bottleneck of zero interest. My hunch is that IBM did not explain this characteristic of IBM Watson or the Achilles’ heel of figuring out the rules. Who wants to sit in a room with subject matter experts and three or four IBM engineers talking about what’s important, what questions are asked, and what data are required.
AskJeeves demonstrated decades ago that human crafted rules are Black Diamond ski runs. IBM Watson’s approach is interesting. But what’s fascinating is the uncritical acceptance of IBM’s assertions and the lack of interest in tackling substantive questions. Maybe lunch was cut short?
Stephen E Arnold, August 16, 2017
Comments
4 Responses to “IBM Watson Deep Learning: A Great Leap Forward”
-
- Subscribe to Beyond Search
Feature archive
News archive
-
I ju?t could not depart your site prior to suggesting that
I e?tremely loved the usual info a person supply on your visitors?
Is gonna be back often to chec? out new posts
You were quite interesting.. But sadly I didnrrrt trust them much :/ Although I might disagree I still give you support as how confident you are on your writing lol
Hi there friends, its great article on the topic of tutoringand completely explained, keep it up all the time.
I visited a lot of website but I conceive this one has something special in it.