|
|
Wellcome to the Artificial Intelligence & Computational Logic Laboratory !
RESEARCH RESULTS
(partial list, starting with the most recent)
|
|
|
No.
|
Domain(s) / Paper(s)
|
Thesis
|
|
1.
|
Artificial Intelligence, Neural Networks, Evolutionary Intelligent Agents, Computational Logic, Computational Intelligence, Iris Recognition,
Exploratory Simulation of an Intelligent Iris Verifier Distributed System
|
©
N. Popescu-Bodorin, February 2011:
- Inconsistent enrollment can change the logic of recognition from a fuzzified 2-valent consistent logic of biometric certitudes to a
fuzzified 3-valent inconsistent possibilistic logic of biometric beliefs justified through experimentally determined
probabilities, or to a fuzzified 8-valent logic which is almost consistent as a biometric theory - this quality being
counterbalanced by an absolutely reasonable loss in the user comfort level.
- The fuzzy 3-valent logical understanding (fuzzy different, fuzzy identical, fuzzy EER interval) of iris recognition is logically inconsistent and will prove anything,
sooner or later (this is the logical mechanism through which the wolves and the lambs appear/enter in a
stationary/non-adaptive biometric system, which in this way exceeds the framework of Consistent Biometry).
- If in an IIVDS the logic of accepts and rejects is the
Propositional Binary Logic (PBL), then the state of corresponding to the EER interval (i.e. PA&NA) is not observable
for IIVDS (or in other words the IIVDS is logically controllable).
- The modal values of truth E (empty set), D (fuzzy different), I (fuzzy identical), and O (fuzzy othewise) are four
elements of a Boolean algebra defined over the congruence classes within Z8 (modulo 8 integers). The
intrinsic 8-valent logic of this Boolean algebra is the 8-vlaent formal logic language of computing with E, D, O,
and I in a logically consistent manner.
- Even when simulating an Intelligent Iris Verifier Distributed System with 1441 terminals allowed to practice random enrollment,
the statistical aspect of recognition is so weak that ensures for the IIVDS outstanding
performance in terms of:
1E-10 pessimistic odds of false accept,
1E-10 pessimistic odds of false reject,
4.12E-4%
undecidable cases (2.7E-4% cases of honest positive claims and 1.42E-4% cases of honest negative claims),
and a safety interval [0.3725 0.55] of width 0.1775 between the maximum reject and minimum accept scores. Hence, the
IIVDS is an almost consistent iris identifier, at least.
|
|
1.
|
Artificial Intelligence, Neural Networks, Evolutionary Intelligent Agents, Computational Logic, Computational Intelligence, Iris Recognition,
Learning Iris Biometric Digital Identities for
Secure Authentication.
A Neural-Evolutionary Perspective Pioneering Intelligent Iris Identification
|
©
N. Popescu-Bodorin, January 2011:
- In Consistent Biometry there is no difference between Iris Verification and Iris Identification.
- An Intelligent Iris Verifier/Identifier stays consistent if and only if it is a non-stationary system enabled to evolve by stepping always through and to a logically consistent state.
- For an Intelligent Iris Verifier/Identifier the time is ticking when a new enrollment occurs.
- An Intelligent Iris Verifier/Identifier is consistent if and only if the histogram of all-to-all comparisons can prove at least a fuzzified and consistent understanding of two words: `genuine` and `imposter`.
- Consistent Iris Recognition is a problem of binary logic or a problem of fuzzified but still consistent binary logic.
- For any inconsistent iris verifier, Monte Carlo Simulations will never be reliable in detecting upper bounds for the False Accept Rate. In Binary Logic the contradiction is explosive.
From a logical point of view, trying to demonstrate an upper bound for the expansion speed of this explosion is a non-sense, and this is mainly because, in the given context, this speed is increasing with time:
in a space saturated with imbricate gravitational clusters (more dense toward the mass center), the process of finding a suitable location for a new cluster to be inserted without colliding it with the other clusters that are already there, only gets harder and harder, and finally impossible.
- For an Intelligent Iris Verifier/Identifier, `evolution` means expanding a vocabulary of digital identities simultaneously with refining a consistent formal biometric theory over this vocabulary.
- The safest way to separate two classes is identifying a third class comfortably situated inbetween them. An Intelligent Iris Verifier/Identifier is an Evolutionary Nonlinear Support Vector Machine.
- Logically Consistent Iris Recognition on a global scale is a problem of computational logic, artificial intelligence, image processing, distributed evolutionary intelligent agents, and supercomputing.
|
|
2.
|
Artificial Intelligence (Code Optimization), Iris Recognition,
Comparing Haar-Hilbert and Log-Gabor based iris encoders on Bath Iris Image Database
|
N. Popescu-Bodorin:
- Iris segmentation is a NP problem. Therefore, it can be optimized either for speed or for accuracy. CFIS2 is a variant of CFIS (Circular Fuzzy Iris Segementation) optimized for speed. Still, it preserves enough accuracy for obtaining good recognition results.
- Haar-Hilbert Encoders are more accurate than Log-Gabor Encoders.
- Multi-enrollment and the use of MDSS (Mean-Deviation Similarity Score) lead to very good separation between the classes of genuine and imposter scores. Hence, multi-enrollment is a step further in defining what a digital identitiy realy is.
- The number FAR(MIS) - False Accept Rate at Maximum Impostoer Score - is a measure of all errors accumulated in the biometric system prior to the matching and prior to the binary encoding of iris texture. Unforced Encoding and Matching techniques are naturally unable to overcome eye image preprocessing errors.
- In certain conditions, the number POFA(mGS) - Pessimistic Odds of False Accept at minimum Genuine Score - is a performance measure for multi-enrollment systems.
|
|
3.
|
Artificial Intelligence, Iris Recognition,
AI Challenges in Iris Recognition. Processing Tools
for Bath Iris Image Database
|
N. Popescu-Bodorin:
- Iris Recognition is and should be considered as a challenge in Artificial Intelligence.
|
|
4.
|
Iris Recognition, Time Series Analysis
Automatic Detection of Common Long-Term Monetary Policies on Global Exchange Market Using Gabor Analytic Phase Binary Encoder
|
N. Popescu-Bodorin:
- Binary Iris Encoders may lead to logical inconsistencies very easily: the global exchange market provide us with an example in which the binary
encodings of two very different curves are too similar for comfort (the ideea is extended in Prop.1 / pp.8 in
this paper).
|
|
5.
|
Computational Logic, Artificial Intelligence
From Cognitive Binary Logic to Cognitive Intelligent Agents
|
©
N. Popescu-Bodorin, January 2011:
- The Cognitive Dialect is a formal logic language whose use enable a Cognitive Intelligent Agent to know its environment,
to comunicate and to use its knowledge for others and for itself.
-
A Cognitive Intelligent Agent able to speak
the Cognitive Dialect is very close to self-awareness because the dialect inherits the native self-reference ambiguity of
deductive discourse written in CCBL (Computational formalization of Cognitive Binary Logic).
- The self-reference ambiguity in CCBL reffers the following situation: the truth does not depend on who is talking, and therefore, `p`
is a simbol used by us when we talk about a given propositional variable, is a simbol used by an artificial intelligent agent
when it `talks` about a given propositional variable, or is a simbol used by a propositional variable when talking about itself, all at once.
This looks a little bit strange at first sight, but it comes very naturally: the most rudimentary intelligent agent is a bit storing the truth value
of propositional variable `p`, and the next simple intelligent agent is a logical circuit: `1-->p`, telling that `p is true`, and obviously, p <--> (1 --> p).
This is the beginning of the self-awareness: `p` is equivalent to `p is true`. Hence, who could say that `p is true` ? Me, you, all of us, and even `p`, and obviously, the truth value of `p` does not depend on who is talking about `p`.
If we now cease to exist, the propositional variable will continue to talk about itself (in a silent non-contradictory auto-referential deductive discourse) waiting to be heard, waiting to be discovered.
And this is the essence of CCBL: a self-reference formal deductive discourse (theory) written with and about the propositional variables of binary logic.
Therefore, I say that self-reference sentences are native and non-paradoxical in CCBL. Of course, human understanding about self-reference sentences
formulated in semantically closed languages is a different thing.
|
|
6.
|
Computational Logic, Artificial Intelligence
Cognitive Binary Logic - The
Natural Unified Formal Theory of Propositional Binary Logic
|
©
N. Popescu-Bodorin, January 2011:
- CCBL (Computational Cognitive Binary Logic) is a monoaxiomatic, complete, consistent and semantically closed formalization of Binary Logic.
- CCBL is a non-paradoxical theory (in CCBL any paradox is an illegal syntax / a logical non-sense).
- The Liar Paradox (LP) is decostructed in CCBL: it is not well-formed in CCBL, hence it is not well-formed in Binary Logic (a wrong common belief is that LP would be an well-formed formula of propositional calculus that `paradoxically` does not have a truth value).
- The only way of entering in CCBL as a paradox is through the empty subset of its vocabulary (there is no non-empty support for paradoxical sentences in the vocabulary of CCBL).
- In CCBL, V=FORM. Any formula (any product) of CCBL theory is a propositional variable.
- CCBL gives a dual description of propositional calculus: as a theory of 2-valued propositional variables, and as a meta-theory of 3-valued modal states of truth: contradiction (impossible truth), contextual (possible) truth, tautology (necesary truth).
|
|
7.
|
Iris Recognition, Artificial Intelligence, 3-valent Crisp/Fuzzy Logic of Iris Segmentation
A Fuzzy View on k-Means Based Signal
Quantization with Application in Iris Segmentation
Exploring New Directions in Iris Recognition
|
N. Popescu-Bodorin:
- CFIS (Circular Iris Fuzzy Segmentation) and GAITBE (Gabor Analytic Iris Texture Binary Encoder) are both reliable tools for experimenting iris recognition on Bath Iris Database.
- Improving iris recognition is a matter of understanding why the statistical aspect is dominant at the intersection of imposter and genuine distributions. Minimizing the chances of False Accept depends on knowing what is happening there.
In such cases (in which the statistical aspect is dominant at the intersection of imposter and genuine distributions), iris verification proves to be logically inconsistent because there exist at least one comparison which matches equal chances to be or not to be a genuine or an imposter comparison.
- If a False Accept occurs, it proves that within the vocabulary of binary iris codes enrolled in the system, there is a non-empty support for the following contradiction: "I'm not a genuine comparison AND I am a genuine comparison". Hence the internal logic of the biometric system is no longer consistent.
Studying why is this happening is mainly a problem of logic.
|
|
|
|
|
|
|
|
Page maintained by Nicolaie Popescu-Bodorin,
Contact (e-mail): nb.popescu.mi # spiruharet.ro
Last update: February 17, 2011
|
