Do Yeong Kang

Authors

Do Yeong Kang, Yeong Hwan Oh, Chanwook Hwang, Jinhee Kim, Kang Eun Jeon*, Jong Hwan Ko*

Abstract

The implementation of Hyperdimensional Computing (HDC) on In-Memory Computing (IMC) architectures faces significant challenges due to the mismatch between high-dimensional vectors and IMC array sizes, leading to inefficient memory utilization and increased computation cycles. This paper presents MEMHD, a Memory-Efficient Multi-centroid HDC framework designed to address these challenges. MEMHD introduces a clustering-based initialization method and quantization-aware iterative learning for multi-centroid associative memory. Through these approaches and its overall architecture, MEMHD achieves a significant reduction in memory requirements while maintaining or improving classification accuracy. Our approach achieves full utilization of IMC arrays and enables one-shot (or few-shot) associative search. Experimental results demonstrate that MEMHD outperforms state-of-the-art binary HDC models, achieving up to 13.69% higher accuracy with the same memory usage, or 13.25x more memory efficiency at the same accuracy level. Moreover, MEMHD reduces computation cycles by up to 80x and array usage by up to 71x compared to baseline IMC mapping methods when mapped to 128x128 IMC arrays, while significantly improving energy and computation cycle efficiency.

Keywords

Hyperdimensional Computing, In-Memory Computing, Multi-Centroid

Links

Paper | Poster

Project Overview

Participated in an autonomous driving competition using children's electric cars equipped with 2D lidar, cameras, and ultrasonic sensors to complete three missions:

• Mission 1: Autonomous driving on a track for 2 laps
• Mission 2: Obstacle avoidance and traffic light detection
• Mission 3: Parking between two randomly parked vehicles

My Contributions

• Vehicle Steering Angle Exploration: Experimentally determined the relationship between motor speed and steering, converting analog angles to digital values for use as parameters
• 2D Lidar Sensor Implementation: Used lidar sensors for obstacle detection and triangulation to identify parking spaces without collisions (achieved highest score among participating teams!)

Results

Secured 3rd place in the competition with particularly strong performance in the parking mission.

Links

Competition Video (4:11:00~4:17:00)

Project Overview

Conducted collaborative research with students from various disciplines including Mathematics, Electronics and Electrical Engineering, and Mechanical Engineering.

Research Topic

Optimization of Training for Once-for-All-Networks:

• Shared a single BatchNorm layer across all bit precision levels to maintain accuracy while reducing training time
• Analyzed the impact of Quantization Aware Training (QAT) on weights, confirming different output distributions before and after quantization
• Implemented Weight Bias Correction to maintain consistent output distribution
• Achieved 20% reduction in overall training time while maintaining accuracy

My Contributions

• Provided mathematical background for convolution operations between distributions and bias correction

Links

SKKU News Article