From Newsgroup: comp.lang.python


An autoencoder learns two functions: an encoding
function that transforms the input data, and a
decoding function that recreates the input data
from the encoded representation. We approach
autoencoders via our already developed SAT Learning
in the Prolog programming language.

Switching from a marginal maximizer to a conditional
maximizer gives better results but also requires a
more costly and slower optimizer. Maximum entropy
methods were already suggest by Peter Cheeseman in
1987. Mostlikely flawed since there is not yet a
feedback loop from the decoder to the encoder.

Maximum Entropy in SAT Autoencoding https://x.com/dogelogch/status/1890093860782764409

Maximum Entropy in SAT Autoencoding
https://www.facebook.com/groups/dogelog

Mild Shock schrieb:

Dogelog Player is a Prolog system for JavaScript,
Python and Java. It is 100% written in Prolog itself.
We present an enhancement to DCG translation. It uses
unification spilling to reduce the number of needed
unify (=)/2 calls and intermediate variables.

Unification spilling can be readily implemented by
performing unification (=)/2 during DCG translation.
Careful spilling without breaking steadfastness gave
us a 10% — 25% speed increase not only for the calculator
example but also for the Albufeira transpiler.

See also:

DCG Translation with Unification Spilling https://x.com/dogelogch/status/1889270444647182542

DCG Translation with Unification Spilling https://www.facebook.com/groups/dogelog

--- Synchronet 3.20c-Linux NewsLink 1.2

From Newsgroup: comp.lang.python


We made our remark reality that a binary decision
tree can be directly created from the data. Starting
from adaptive trees we built a new aggregate that can
perform the statistics for a Bayes Classifier using
the majority rule. We only use Prolog code!

The adaptive tree can be used like a bitwise trie,
and allows us to compute some statistics in one pass.
From this statistics we can then derive a decision
tree using a majority rule. The entropy of the computed
output will be inside an 1/2 bit interval of the
sample output entropy.

See also:

Bayes Classifier for SAT Learning https://x.com/dogelogch/status/1892517071730135467

Bayes Classifier for SAT Learning
https://www.facebook.com/groups/dogelog

Mild Shock schrieb:

An autoencoder learns two functions: an encoding
function that transforms the input data, and a
decoding function that recreates the input data
from the encoded representation. We approach
autoencoders via our already developed SAT Learning
in the Prolog programming language.

Switching from a marginal maximizer to a conditional
maximizer gives better results but also requires a
more costly and slower optimizer. Maximum entropy
methods were already suggest by Peter Cheeseman in
1987. Mostlikely flawed since there is not yet a
feedback loop from the decoder to the encoder.

Maximum Entropy in SAT Autoencoding https://x.com/dogelogch/status/1890093860782764409

Maximum Entropy in SAT Autoencoding
https://www.facebook.com/groups/dogelog

Mild Shock schrieb:

Dogelog Player is a Prolog system for JavaScript,
Python and Java. It is 100% written in Prolog itself.
We present an enhancement to DCG translation. It uses
unification spilling to reduce the number of needed
unify (=)/2 calls and intermediate variables.

Unification spilling can be readily implemented by
performing unification (=)/2 during DCG translation.
Careful spilling without breaking steadfastness gave
us a 10% — 25% speed increase not only for the calculator
example but also for the Albufeira transpiler.

See also:

DCG Translation with Unification Spilling
https://x.com/dogelogch/status/1889270444647182542

DCG Translation with Unification Spilling
https://www.facebook.com/groups/dogelog

--- Synchronet 3.20c-Linux NewsLink 1.2