Our core technology is the Autonomous Adaptive Algorithm (AAA). Autonomous Adaptive Algorithm™ (AAA) synergistically matches state-of- the-art machine learning algorithms such as neural computation, fuzzy computation, and evolutionary computation to form a hybrid algorithm. AAA combines the best features from all three methodologies to produce neural-fuzzy, fuzzy-evolutionary, and/or evolutionary-neural models.
Neural network-based systems are good at discovering relationships and patterns from data and are well-suited for clustering and classifications tasks. Fuzzy logic-based systems are capable of performing high-level, human-like reasoning using linguistic variables and are well-suited for knowledge representation and decision support problems. Genetic-based systems are able to conduct random, multi-modal searches and are well- suited for optimization problems.
The AAA forms a hybrid neural-fuzzy model that is able to achieve accurate estimates of the posterior probabilities as required for performing risk- weighted classification and prediction in accordance with Bayes' Theorem. It has been demonstrated to be capable of learning autonomously and, at the same time, refining its knowledge based perpetually in response to an arbitrary sequence of input patterns in non-stationary (time-varying) as well as stationary (time-invariant) environments. AAA is also able to extract knowledge in the IF -THEN rule format that is comprehensible to users to justify its predictions.
The learning dynamics of the AAA is based on the Adaptive Resonance Theory (ART) neural network family which gives AAA the adaptive ability in response to significant events, and yet remains stable in response to irrelevant events. In addition, AAA can absorb new information without corrupting or forgetting previously learned information.
AAA is able to acquire knowledge in a noisy environment, with missing or incomplete data. In practice, it is not uncommon to encounter situations where data samples collected from the real environment contain missing features/values, e.g. object occlusions in machine vision, sensor failures in control systems or non-responses in survey questionnaires. The efficacy of machine learning systems is often compromised by the occurrence of incomplete data samples. Some systems will simply discard incomplete samples and learn using complete ones, causing a great loss in valuable information. Thus, the ability of a data-based learning system to learn from samples with missing features such as AAA is crucial.
Features
- Utilizes state-of-the-art machine learning algorithms such as neural computation, fuzzy computation, and evolutionary computation
- A hybrid neural-fuzzy model that is able to achieve accurate estimates of the posterior probabilities as required for performing risk-weighted classification and prediction in accordance with Bayes' Theorem
- Capable of learning autonomously
- Refine its knowledge based perpetually in response to an arbitrary sequence of input patterns in non-stationary (time-varying) as well as stationary (time-invariant) environments
- Extract knowledge in the IF-THEN rule format that is comprehensible to users to justify its predictions
- The learning dynamics of the AAA is based on the Adaptive Resonance Theory (ART) neural network family which gives AAA the adaptive ability in response to significant events, and yet remains stable in response to irrelevant events
- Absorb new information without corrupting or forgetting previously learned information
- Able to acquire knowledge in a noisy environment, with missing or incomplete data
- Proven operational with superior performance in various applications, including credit rating, complex manufacturing process monitoring and medical diagnostics
top
