Bayesian Pseudoinverse Learners: From Uncertainty to Deterministic Learning.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Pseudo-inverse learners (PILs) are feedforward neural networks utilizing pseudoinverse learning algorithms, originating in 1995.
• PIL offers non-gradient descent learning with advantages in computational cost and speed, relevant for edge computing.

Purpose of the Study:

• To address the limitation of PILs in handling deterministic learning problems by extending their applicability to uncertainty learning.
• To introduce a novel learning framework, Bayesian PIL, within the synergetic learning system (SLS) framework.

Main Methods:

• Developed an approximated synergetic learning scheme to convert uncertainty learning problems into deterministic ones.
• Integrated this scheme into the PIL framework, creating the Bayesian PIL.

Main Results:

• Demonstrated the effectiveness of the Bayesian PIL in transforming uncertainty learning into deterministic learning.
• Showcased the advantages of the Bayesian PIL framework in handling real-world uncertainty scenarios.

Conclusions:

• The proposed Bayesian PIL framework successfully extends the capabilities of PILs to address uncertainty learning.
• This advancement holds significant potential for applications in edge computing and other fields dealing with uncertain data.