Monthly Archives: November 2007

Streaming Semidefinite Embedding

I’m posting this just for the purpose of timestamping. Today I proposed an idea to stream kernel learning techniques such as semidefinite embedding. The trick is to pass minors one row and column at a time (actually, the matrix is symmetric, so just one row) and update using incremental kernel PCA. This results in an algorithm that only needs to store N elements in memory at a time rather than N^2.

Finer-tuning the Strong Anthropic Principle

The strong anthropic principle states that any viable universe must have the capacity for observation; that is, life must evolve in it. If the multi-worlds interpretation is correct, however, we may not all necessarily be in the same universe (if “quantum immortality” is correct, it gets even weirder: people might be in the same universe at one moment in time and may forever diverge at a future branching point as their own survival takes them along different paths).

I wonder whether we can propose a “strongest anthropic principle” of some sort that roughly states that the particular universe each person (or lifeform in general) inhabits evolved specifically for that person/organism rather than for the general existence of life as a whole. The existence of multiple universes would permit it.

We could even take it further, actually, and permit free will under the assumption of determinism (disclaimer: I am not a determinist), though this treads dangerous philosophical and theological ground because it would essentially argue that we are God: if the initial state of the universe is organized for a specific organism, it may be organized for the organism’s free will, or even by the organism’s free will.

Not that I believe this, but the ideas are intriguing. It’s the ultimate philosophy of egocentrism πŸ™‚

Dissertation – Week 3

I’m running somewhat behind this week. It really can’t be helped – I had to travel to Philadelphia three times this week (once I’m there, I have a very difficult time concentrating on my work, and the three hour commute drains me for more or less the rest of the day), I’m trying to perform experiments for the first time since I began writing, and my machine learning workload is transitioning from “unreasonably heavy” to “sadistic”. Despite this, I did manage to finish off about five pages thus far, mostly dealing with the nuances of the tensor/outer product and with things like PARAFAC. The methodology section is probably going to need to wait until next week, since I don’t know how quickly I can get experimental results with all of the other things going on.

Fortunately, I planned a timeline that has me finishing approximately 6 months before I need to defend – I have plenty of time to spare πŸ™‚

If my ML load was a bit lighter this week, I could even meet my weekly goal by tomorrow. Unfortunately, not only is it heavy, but the professor decided to make it a competition. I abhor competitive research almost as much as competitive programming (I only competed in one programming contest, when I was 15, which I won to the detriment of some very talented programmers who really should have been lauded for their efforts and talent) – it fosters a spirit of treachery and ensures that the total number of people performing well remains very small. There’s already enough adversity in research; I fail to see why we need to add more.

In any case, I am now bound by the Red Queen hypothesis. I can’t simply submit functional homework; now I have to keep improving it to the extent that it is more accurate than the other students’ submissions.

I’m probably safe if I can get my cross-validated accuracy above 90%.

Kronecker's Delta

δij = 1 if i=j, 0 otherwise.

If you’re going to declare such a trivial operator part of the standard notation, at least don’t name it after someone as if it were a groundbreaking novelty!

Also see Iverson brackets, a similar operator named after the creator of APL who received a Turing award in part “for his contributions to mathematical notation” (hopefully more because he created APL; most creators of popular programming languages have won Turing awards).

Where Maslow Becomes Dabrowski – The Emergence of the Fourth Factor

To quote Maslow:

“I have recently found it more and more useful to differentiate between two kinds of self-actualizing people, those who were clearly healthy, but with little or no experiences of transcendence, and those in whom transcendent experiencing was important and even central… It is unfortunate that I can no longer be theoretically neat at this level. I find not only self-actualizing persons who transcend, but also non-healthy people, non-self-actualizers who have important transcendent experiences. It seems to me that I have found some degree of transcendence in many people other than self-actualizing ones as I have defined this term…”

This is precisely the place in which Maslow’s theory can be extended to Dabrowski’s. These “transcendent experiences” likely correspond to Dabrowski’s crises. The pre-actualization crises are already explained by Dabrowski’s levels II-IV of disintegration. However, if we synthesize the two concepts, Dabrowski claims the final crisis propels an individual to self-actualization, which is characterized by inner harmony, but Maslow claims that further transcendent experiences exist!

So which is right? Well, probably both, in different senses. Dabrowski’s “level 5” is a state in which one’s behavior is completely self-constructed (based on “the third factor”, which is an individual’s drive towards growth and autonomy), which should bring about an internal peace. However, it is the nature of humans (especially those characterized by a “drive towards growth”) to continuously strive for better situations, and thus such value systems will change with time, to be replaced by value systems that the individual considers “higher” as his or her perspective, knowledge, and self-expectations shift! (It is worth noting that we avoid infinite recursion because the “meta-values” are responses to internal or external circumstances embedded within the value system itself; there is no such thing as a separate “meta-value”, which would require a “meta-meta-value”, and that a “meta-meta-meta value”, …).

Thus, we have a level 6 state, Meta-Integration, in which the value system itself becomes subject to an individual’s scrutiny. This state is likely the final resting point of the fully actualized psyche, but only because it is iterative: it represents a “punctuated equilibrium” of peaceful periods followed by intense and quite deliberately guided revisions (which are, in a sense, rapidly occurring re-disintegrations) due to rapid changes in one’s underlying values brought about by what Maslow calls “transcendent experiences”.

We can theoretically call self-scrutiny the “fourth factor”, but it’s more like an inwardly-turned version of the third. Still, the step from being certain in one’s value systems (though a healthier condition than relying upon society or self-benefit to justify one’s behaviors) to devising value systems that are internally consistent and stable, yet flexible as the individual gains new knowledge and experience is clearly a healthy one: we can never acquire the sum total of the world’s knowledge or experience, so absolute rigidity is pathological. Dabrowski himself was the one who stated that healthy people must accept the world as it is, and it is not rigid.

The step from Level 5 to Level 6 is huge, however, perhaps even to the extent of the step from Level 1 to Level 2, as it requires abandoning stability. However, it is necessary to fully achieve one’s potential, rather than to simply act as the image of one particular basis of values, even one that was self-chosen, because values are fluid. It represents the extension of one’s moral reasoning from synchronic to diachronic, as one can now envision a direction or change in a value system, and thus an internally driven future expectation. Though (barely) expressible in Maslow’s theory through his addition of “transcendent events”, this is impossible to describe in Dabrowski’s theory as Dabrowski stated it.

So how can we summarize this?

  • Secondary Integration represents a point of stability, but people operating at Level 5, though “self-actualized”, do not have the ability to effectively question their own value systems as new internal or external circumstances compel them to.
  • The initial crisis that forces a person to adopt a new set of values does not represent negative adjustment unless it indicates a regression to social or self-driven values (the first or second factors). If the revision to one’s value system is conscious and directed, it represents a higher level of self-actualization rather than a lower one, which we call Level 6 – “Meta-Integration”.
  • Paradoxically, this state is not as stable as Level 5, as it undergoes rapid periods of change coincident with Maslow’s “Transcendent Events”. When not undergoing these changes, it is at least as stable as Level 5, as one’s behavior is not only consistent with one’s value system, but one’s value system is consistent with one’s circumstances and expectations.
  • Because this can happen many times, it likely represents the final state of the psyche. Thus, even a self-actualized person must undergo crises from time-to-time; the highest state of consciousness is still directed by the presence of distress (depressing? Well, the result is a more personally-optimal value-system, so the hardship is greatly offset by the newfound knowledge; it can be thought of as learning).
  • The inward expression of the third factor to the end of self-examination can be considered a “fourth-factor” that has not yet emerged at level 5. It’s not really distinct from the third, however.
  • It’s a big jump, and many self-actualized people do not successfully make it. Such people likely remain at Level 5 rather than negatively adjusting, as they are already convinced of the rectitude of their value systems. These are the artists or scientists who ply their craft in a manner that they are convinced is correct due to their internal values, but are unable to abstract themselves away from the situation and ask “is this really correct anymore?”

With this addition, Dabrowski’s theory falls neatly into place with Maslow’s and my own. I think it’s quite an elegant and powerful idea that unifies two major developmental theories. Ignore it at your own risk.

I also made some less elegant distinctions on Dabrowski’s first level at my Temple page.

Bug in Google Maps

Click on a “more info”/”reviews” link. It will launch a large dialog with detailed information on the location you’re considering.

Now, while this dialog is up, position the cursor outside of it and scroll the mouse wheel. The map zooms. The dialog goes off-screen. Chaos ensues if you zoom out to get it back (though panning generally works). You can even get part of the map to extend outside of the mapping window if you keep zooming.

Tested in Firefox and IE. It seems worse in Firefox.

It’s just cosmetic, so it isn’t that serious.

Downright Obvious Idea for Google Scholar

“Sort by date”. It’s one of those features that you browse to a site and wonder “how did they miss that?”, because it just makes doing a real literature search with Google Scholar a pain.

Unless it’s just buried in some undocumented syntax? It isn’t an option in the advanced search page.

There’s also the idea of a general media similarity search that I mentioned earlier, but I’ve proposed this both during the first Summer of Code and when I interviewed at the NYC office (and I even gave them my BACH paper the second time to give them an idea of how to do it for music!), and neither time was it acted upon, so I’m going to shut up about that one. I did, however, suggest it to Yahoo!, who seems closer to having that sort of search than Google does at the moment.