A Hidden Markov Model-Inspired Sequence Classification Method for Hyperdimensional Computing

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Sequence classification is crucial for analyzing real-world data.
• Traditional methods like Hidden Markov Models (HMMs) face challenges with variable-length, misaligned, or noisy sequences.
• Hyperdimensional Computing (HDC) offers an alternative paradigm for data processing.

Purpose of the Study:

• Introduce a novel HDC-based method for discrete-sequence classification.
• Address limitations of existing sequence analysis techniques, particularly in handling real-world data complexities.
• Develop a method suitable for efficient hardware implementation.

Main Methods:

• Replaced algebraic operations in HMMs with bitwise operations on hyperdimensional binary vectors (hypervectors).
• Developed a hypervector transformation pipeline mirroring HMM algebraic manipulations.
• Incorporated a procedure to prevent information decay in long sequences.

Main Results:

• Achieved superior classification accuracy compared to existing HDC sequence analysis methods on artificial and real-world datasets.
• Demonstrated robustness against bit flips, a common issue in hardware implementations.
• Showcased efficiency through the use of binary bit-wise operations.

Conclusions:

• The proposed HDC method effectively classifies discrete sequences, overcoming challenges like variable length and noise.
• The approach is highly suitable for hardware implementation, particularly in VLSI devices.
• Offers significant advantages in accuracy and robustness over traditional and other HDC methods for sequence analysis.