An interval type-2 neural fuzzy chip with on-chip incremental learning ability for time-varying data sequence...

Area of Science:

• * Artificial Intelligence
• * Hardware Engineering
• * Fuzzy Logic Systems

Background:

• * Traditional interval type-2 fuzzy systems are computationally intensive for hardware, limiting their use in changing environments.
• * Existing hardware implementations often sacrifice learning performance for efficiency.
• * Online learning capabilities are crucial for systems adapting to dynamic conditions.

Purpose of the Study:

• * To propose a novel Mamdani-type interval type-2 neural fuzzy chip with on-chip incremental learning (IT2NFC-OL).
• * To reduce hardware implementation costs through a simplified type reduction operation without compromising learning performance.
• * To enable real-time fuzzy inference and online parameter learning in a field-programmable gate array (FPGA) chip.

Main Methods:

• * Development of a software-implemented IT2NFC-OL featuring online structure and parameter learning via gradient descent.
• * Implementation of the learned fuzzy model onto an FPGA chip.
• * Design of novel circuits for system output computation and interval consequent value updates.

Main Results:

• * The IT2NFC-OL demonstrates efficient hardware implementation with reduced costs due to a simplified type reduction.
• * The FPGA-implemented system successfully performs fuzzy inference and online parameter learning.
• * Verified learning performance in time-varying data sequence prediction and system control tasks.

Conclusions:

• * The proposed IT2NFC-OL provides an efficient hardware solution for fuzzy systems in changing environments.
• * On-chip incremental learning capability allows for real-time adaptation and parameter updates.
• * The novel circuits and simplified approach offer a cost-effective and high-performance alternative to existing fuzzy systems.