Archive for May, 2011



May 19, 2011

Señor God’s Gift Achiuwa, you asked whether I would play Monopoly with a dashboard or against a computer.  I would play with the help of a dashboard.  Playing Scrabble against Nanaji making occasional use of the dictionary is more fun for me than without, but I don’t think playing against a computer would be nearly as fun.  Much of that is because of Nanaji’s inimitable style.

Not only does he have his own style in playing Scrabble, but in many other aspects of life as well.  No matter what one thinks of him and his choices, one has to agree that he has certainly dared to be different

As you well know, one of his deliberate, oft-used phrases is “क्या बेचता है?” (“What does he sell?”) when asking who someone is.  Having been at IBM and looking at business analytics applications for more than a year, I am more deeply understanding how basic and important buying and selling are to everything.  To help build my intuition regarding sales and marketing, I have been subscribed to the MarketingProfs newsletter for a while now. 

In the newsletter, I have been seeing ‘storytelling’ mentioned more and more as a tactic to sell, including an article this week.  You have some nice ideas for what grand challenge-type things IBM Research can do next after Watson, the Jeopardy! challenge, which you might share here in the future.  One of my ideas is a machine that can create stories or screenplays that make you laugh, think. and cry.  It fits in with the business focus of the company via this new storytelling trend in marketing, is challenging but doable in my estimation (and David Ferrucci’s), and is related to storytelling thoughts I have shared here previously (1, 2, 3). 

Nanaji has phrases related to chess “शाह मौत [shah maut]” and to Jeopardy! “I know, I know, I know, you don’t know.” What ideas do readers have as challenges for IBM to tackle after chess and Jeopardy!?


Krishnan Eswaran

May 18, 2011

This time around, we converse with Krish Eswaran, a software engineer in Video at Google.

Kush: You say that you are not currently active in the research community. What are your thoughts on this blog post about having a research career at Google?

Krish Eswaran: The blog post was interesting to read, and it helped me crystallize better what I think are the differences between being active in the research community and working at Google. To do this, I’m going to make an analogy to musical composition, and let me say from the start that I don’t know anything about composition as a profession, so this analogy is both a product of ignorance and a convenient way to draw a distinction. Don’t take it too seriously.

Let’s say there are a group of people whom I’ll call the composer community, almost all of whom studied music theory for many years. This group is told to compose music and that they will be judged by their influence as composers: how often other composers quote their style and/or passages. The composers might be commissioned to compose pieces for performers, but there’s no guarantee that a particular composition will ultimately be played.

Let’s take another group: a jazz band. Not everyone in the jazz band studied music theory for many years, but they’ve all played and enjoy playing music. The jazz band is judged on how well they can execute a performance: how well they play together. There might be a member or few members who enjoy composing and writing music for the group, and in some sense, one could argue each member is composing on the fly when playing a jazz solo. Members can also choose to transcribe a performance or even individual solos into a composition, but it doesn’t bear any significant weight in how they’ll be judged.

Now, if I were in a jazz band after having studied music theory for many years and wanted to convince my friends who’d also studied music theory to join, I might emphasize the fact that yes, there are opportunities to compose music in the jazz band, and by the way, the jazz band I’m in has a lot of people who’ve also studied music theory for many years, in addition to those who thought their time would be better spent playing in bands. However, I would expect the people I were trying to convince to feel comfortable helping to execute the parts of a performance that have nothing to do with composition, which may not be the case for everyone.

Lav: Interestingly, one of my colleagues at IBM holds a doctorate in music composition, and today he was showing me the results of a music morphing system that he had developed many years ago, not dissimilar to face morphing. Which musicians’ works do you think would produce the most interesting morphs? Ludakrishna and Antonín Dvořák? Paul Simon and Dispatch?

KE: It’s easier to approach this question through individual songs since musicians’ styles can change. I think the most interesting music morphs would provide some type of contrast, and I can immediately think of two categories: very different songs that conjure up similar memories and superficially similar songs with subtle but significant differences. For the former, music morphing Europe’s “Final Countdown”  to Kanye West’s “Diamonds From Sierra Leone” while face morphing Will Arnett into Kanye West would be pretty cool. For the latter, music morphing Fleetwood Mac’s “Landslide” to Faith Hill’s “Breathe” while face morphing Lav into Kush would be really cool, as well.

As an aside, a cool way to test a music morphing algorithm might be to divide up a song into smaller segments and see how it would morph the different pieces together.

K: Thanks for the music videos on YouTube.  How did you get into bush-hogging? Do you bush-hog in any of your work these days?

KE: My financial aid package in college was tied to working on a research project, but I was so overwhelmed with my course load that I took a break from the project I started my first year. When I felt ready to start again my second year, my interests had shifted, and I went looking for something new.

During the beginning of the fall semester of that year, I sent an email to Toby about the projects in his video lab, all of which sounded really cool to me. I didn’t hear back so I continued talking to other professors about a pie-in-the-sky idea I had at the time. I was almost set to work with one of them near the end of that semester when Toby responded. We met and discussed the projects in his lab. He assumed by then that I might be talking to other professors and continued to encourage that. It was fairly noncommittal.

During the spring semester, I sent him another email saying I was interested in joining one of the projects we had discussed, but there wasn’t an immediate response again. This time, I decided to go directly to his lab and talk with his students. In doing so, I found a starting project, sent an email to Toby letting him know what I was working on, and eventually worked directly with Toby on a signal processing problem that had more to do with neuroscience; however, the initial project led me to the work of other professors when applying to grad school, one of which was Kannan’s.

Fast-forwarding to the present, issues of rate and distortion continue to play a role in my life, and I’ve also been gaining an appreciation about the need for speed.

K: How do you levitate?

KE: In the Fall of 2001, I walked into the dorm room of two of my friends and mentioned in passing that I could float like a character from Street Fighter. They laughed and asked me to demonstrate, and while I had never floated before, felt the need to prove myself. After the demonstration, they stopped laughing and got out a camera.

Here’s one of the photos they took. Your readers should be able to combine it with the anecdote they’ve just read to figure out the secret.


Peter B. Jones

May 9, 2011

Our second interview is with Peter Jones, who completed his Ph.D. thesis last month at MIT.

Kush: You live pretty close to where Ralph Waldo Emerson used to live. What are your thoughts on his statement: “A foolish consistency is the hobgoblin of little minds …. Speak what you think now in hard words, and tomorrow speak what tomorrow thinks in hard words again, though it contradict every thing you said today.”

Peter Jones: Foolish consistency is, of course, foolish, but knowing when consistency is foolish and when it is wise is something even Emerson never really mastered (see, for example, his hesitant conversion to abolitionism). In St. Paul’s epistle to the Ephesians he explains the purpose of the Christian church organization is so that we “be no more children, tossed to and fro, and carried about with every wind of doctrine.” I think the modern world (at least, the Western world), in its revolt against institutions and authority, has veered too sharply toward Emerson and away from Paul (who, let us not forget, knew first hand about overnight conversions). This leaves society open to thrashing as we follow the fickle majority from one pendulum peak to the other. Personally, I feel ardor and zeal are excellent traits (only) when sufficiently tempered by introspection and foresightfulness.

Lav: Your doctoral work looks at coherently combining disparate beliefs not only in static settings but also when the world is changing dynamically. Indeed you have provided theorems on best ways to combine incoherent beliefs. Do you think institutions like churches, governments, or community organizations are adept at this? What could be better?

PJ: There is certainly aggregative benefit in a healthy civil society, including the institutions you mention. However, I think the key benefit of such institutions is less the ability to aggregate opinion and more the impact on community members, as they form connections with those who are ideologically, or demographically other. If, as Robert Putnam has claimed, civil society in America is in retreat then I think the primary loss is not of institutions and institutional knowledge (although that is significant). The greater loss is the trust, understanding, and tolerance of those with different backgrounds, beliefs and ideologies that arises from vibrant organizations.

L: Sometimes there are fundamental differences in subjective beliefs, e.g. about the beauty of a woman or the luckiness of a number. Do you know of any good ways of performing either beauty arbitrage or luckiness arbitrage?

PJ: Luckiness arbitrage shouldn’t be hard. Suppose someone (P1) has a “lucky” number. Step One: get P1 to go long on their lucky number winning, say, a randomized ball draw at odds higher than uniform (you take the short side). Step Two: get someone rational (P2) to go short on the same ball draw at uniform odds. Step Three: Profit! Since you’ve got a long position at the uniform odds and a short position at higher odds, you have a guaranteed win. If the “lucky” ball comes up, you pay P1 some amount, but receive a larger sum from P2 and vice versa if the “lucky” ball doesn’t come up.

I’ll sidestep the question of beauty arbitrage; everything I’ve tried to write about it sounds a bit too much like human trafficking for my comfort.

K: When did you first get the desire to work on problems of national security?

PJ: I’m not sure when I first got interested in national security.  Certainly from about the time I graduated from high school I’ve been interested in politics and political issues, and have felt compelled to make, in some way, a contribution to preserving the values and institutions we all rely on.  My parents were both politically active on a local level, and my mom ran at least twice for state legislature (but was never actually elected). I see in national security research a nexus of my technical abilities, my intellectual interests, and my obligation to “give something back.”

K: Is Jimmer Fredette the reincarnation of Pistol Pete Maravich?

PJ: Fredette-mania was pretty out of control in March.  It seemed like half the posts on my Facebook feed from friends in Utah were about Jimmer.  If Jimmer does as well in the NBA as Maravich did I’ll be both gladdened and a little surprised.  He seems like a good guy and I hope he succeeds.

As a side note, my “Junior Jazz” coach used to call me “Pistol Pete.”  I figure it was mostly an attempt to inspire my play.  Unfortunately, as far as I can tell, it had little impact on my shooting accuracy.


Fun and Games

May 7, 2011

As you describe, the growth of analytics within sports and the growth of fantasy sports in society have grown hand in hand.  Of course this is not a surprise, since both are very much enabled by information technology.  Although it is definitely possible to run fantasy competitions on paper, it surely must be true that the only sports prediction competition most people competed in before the proliferation of the internet were NCAA Tournament pools.  Similarly, box scores have been available in newspapers for generations, but the use of statistics in game strategy is a new phenomenon.  Indeed, interesting new metrics have sprung up that better capture the so-called “business model” of sports teams.  I look forward to reading your magazine article about business analytics, once it is fully formatted.

At some point, perhaps you can also write a magazine article about the strategy of cricket and cricmetrics, since I don’t really get it.  Though of course, I haven’t really ever watched cricket as serious sports fan, either.

As previously mentioned, we had taken a family vacation to the Poconos at the end of last year, and there was some game-playing that happened: notably Connect Four.  The trip was very relaxing but as some people say, information workers are always stressed because they can’t turn off the tap of work like material workers.  That is to say, I kept thinking about an optimal strategy for Connect Four. 

It is self-evident to anyone that Tic Tac Toe has a pure dominant strategy, that always leads to a tie, but what about Connect Four?  It is pretty much the same game, just bigger and with a gravitational constraint on strategies.  Is there a dominant strategy for Connect Four?  Does the first player have an advantage?  As one might expect, a formal game-theoretic analysis is possible.  As it turns out, under optimal play on a standard 6 by 7 board, the first player should always be able to win.  In practice, though, bounded rationality leads to all kinds of outcomes.

As I’ve been learning recently, this game theory stuff is rather interesting, with uncertainty not just due to noise but also what competitors are going to do.  In ongoing work with Joong Bum Rhim, Vivek Goyal, and you, I’ve been looking at quantizer design games and a game-theoretic view of decision-making under limited information of various kinds.  I feel like there is good potential for this line of research, but we’ll see.

Something else I’ve been rather intrigued by is the use of games as a means of doing work, e.g. using the techniques that Luis von Ahn has mastered, but I need to learn more about it, especially in relation to crowdsourcing.  I wonder if games like Monopoly that have a positive probability of never ending are perceived as more fun or less fun, and whether a dashboard makes it too much like work.  Would you play Monopoly with a dashboard or against a computer?