Cerrisian Blog

So, I’m going to try a little experiment called sleep free-running. As much as possible, I’m going to let my body dictate my sleep cycle. I’m going to go to bed when I sense a rapid increase in drowsiness, if I wake up, I’ll get up and learn something. The eventual goal will be to use my sun alarm clock as a zeitgeber.

I really do enjoy learning, and my StrengthsFinder results are consistent with that. I think I’m going to spin up mnemosyne again and start learning the definitions of names. I’m going to put more emphasis on creating good “first chapters” and fundamental concepts to learn, rather than just cloze deletions. This may be a bit labor intensive, but I think it’ll be worth it.

I’ve been using “don’t break the chain” a lot recently. It’s been very effective at breaking habits, but not so good at instilling new ones. Maybe I try to instill too many new habits at once.

Anyways, we’ll see how this goes. I think that having a more natural sleep cycle will be helpful in a lot of things.

The ideas is simple, create a program/library to allow for literate programming in RST and Sphinx. Basically take labeled code blocks and combine them into files.

Use something like #< ># as insertion blocks

May need a special block to allow for cross references between code blocks.

It should also allow for pieces of code to be maintained externally.

I just took the StrengthFinder test and here are my top five strengths:

• Learner
• Maximizer
• Input
• Achiever
• Intellection

In this post, I’m going to explore how they can connect to each other and support each other. The first example I’m going to use is my interest in spaced repetition learning. This is definitely a “maximizer” type activity, but is useful to the two other themes of learner (by memorizing faster I can learn more information) and input (more memorization means more knowledge).

Similarly, code katas feed my maximizer need and give a good framework for achievement. They’re also definitely a form of learning.

And this, posts like these are intellection.

http://www.eswc2011.org/content/cfp

15 pages LNCS format

December 6

Prior Research done by Oct. 27

Final paper by Nov. 15

Literage Programs is a wiki that has literate explanations of various programs and algorithms. It’s very small at this point, and needs to be expanded.

A fun outcome of code katas would be algorithms or programs to post on here, or code that starts from here.

They could also be used for porting via “boot strapping”. First, take the implemented program and use it as an oracle for fuzz testing, then implement the program in hte host language of choice.

Tonight, I got results better than the state of the art for Ontology alignment. Here’s the approach I took.

• Create a cross-validating set of test cases, as described in the Spambayes documentation.
• Analyze the current algorithm and existing code. Pull hair out in frustration, make no progress.
• Decide to try things to see if they work. Have great success.
• Repeat previous step until fresh ideas are gone
• Take a shower, come up with new ideas
• Read previous work out loud until additional features are found
• Implement those features, reach point of parity and above with previous work

In short, DEFOCUS DEFOCUS DEFOCUS! Have a strong spine to hold onto (in this case, f-measure and automated tests) coupled with extreme freedom to play around and experiment. Fiddle factor for the win!

Really cool tool for sketching on the computer: Alchemy

Bribe “inputs”, behaviors, not results.

Act as though you don’t have impulse control, then you won’t be tempted and thus won’t be impulsive.

To communicate properly, cover the following things when introducing a new technology

• Here’s how it works now (share context, rider)
  • Problems with old way (elephant)
• Here’s how the new technology is different and what it enables
  • Here’s what’s different (share context, rider)
  • Here’s what it enables (guide the elephant)
• Here is what we have to do to get from old way to new way
  • (engage the rider)

Epistemology

• Study of knowledge
• The probabilities associated with specific causal paths
• What we believe — a subset of what we know to be possible

Epistemic errors:

• Making errors of “maximum likelihood” ignoring what we know is possible for what we believe will happen.
• Pearl harbor — believing that the only practical attack from Japan was terrorism, not an airstrike by carriers.
• 9-11 — believing that a terrorist attack using airplanes was unpractical — not feeding it in to the rest of the system, not arresting the terrorists.

Ontology

• How the world is modeled
• All possibilities
• Our causality graph
• Should be stable [Pearl]
• Can be learned, just as epistemic structure can be.

Ontological Errors:

• Missnig nodes or arrows in our causal graph
• European encounter with the Americas
• Can these be mitigated by “unknown nods” and other mechanistic things?

Black Swans

• Errors in epistemology (possible gray swans)
• Errors in ontology (’true’ black swans, we may know they exist, but we don’t know how hard they’ll hit or what they’ll affect)

Bayesian networks (and machine learning/artificial intelligence) that clearly denotes where black swans are possible and are robust against them.