Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient and Reliable Vector Similarity Search Using Asymmetric Encoding with NAND-Flash for Many-Class Few-Shot Learning

Hao-Wei Chiang, Chi-Tse Huang, Hsiang-Yun Cheng, Po-Hao Tseng, Ming-Hsiu Lee, An-Yeu, Wu

TL;DR

An integrated framework that reduces search iterations by up to 32×, and increases overall accuracy by 1.58% to 6.94%, and optimizes controller training by modeling the hardware characteristics of MCAM, thus enhancing the reliability of the system.

Abstract

While memory-augmented neural networks (MANNs) offer an effective solution for few-shot learning (FSL) by integrating deep neural networks with external memory, the capacity requirements and energy overhead of data movement become enormous due to the large number of support vectors in many-class FSL scenarios. Various in-memory search solutions have emerged to improve the energy efficiency of MANNs. NAND-based multi-bit content addressable memory (MCAM) is a promising option due to its high density and large capacity. Despite its potential, MCAM faces limitations such as a restricted number of word lines, limited quantization levels, and non-ideal effects like varying string currents and bottleneck effects, which lead to significant accuracy drops. To address these issues, we propose several innovative methods. First, the Multi-bit Thermometer Code (MTMC) leverages the extensive capacity of MCAM to enhance vector precision using cumulative encoding rules, thereby mitigating the bottleneck effect. Second, the Asymmetric vector similarity search (AVSS) reduces the precision of the query vector while maintaining that of the support vectors, thereby minimizing the search iterations and improving efficiency in many-class scenarios. Finally, the Hardware-Aware Training (HAT) method optimizes controller training by modeling the hardware characteristics of MCAM, thus enhancing the reliability of the system. Our integrated framework reduces search iterations by up to 32 times, and increases overall accuracy by 1.58% to 6.94%.

Efficient and Reliable Vector Similarity Search Using Asymmetric Encoding with NAND-Flash for Many-Class Few-Shot Learning

TL;DR

An integrated framework that reduces search iterations by up to 32×, and increases overall accuracy by 1.58% to 6.94%, and optimizes controller training by modeling the hardware characteristics of MCAM, thus enhancing the reliability of the system.

Abstract

While memory-augmented neural networks (MANNs) offer an effective solution for few-shot learning (FSL) by integrating deep neural networks with external memory, the capacity requirements and energy overhead of data movement become enormous due to the large number of support vectors in many-class FSL scenarios. Various in-memory search solutions have emerged to improve the energy efficiency of MANNs. NAND-based multi-bit content addressable memory (MCAM) is a promising option due to its high density and large capacity. Despite its potential, MCAM faces limitations such as a restricted number of word lines, limited quantization levels, and non-ideal effects like varying string currents and bottleneck effects, which lead to significant accuracy drops. To address these issues, we propose several innovative methods. First, the Multi-bit Thermometer Code (MTMC) leverages the extensive capacity of MCAM to enhance vector precision using cumulative encoding rules, thereby mitigating the bottleneck effect. Second, the Asymmetric vector similarity search (AVSS) reduces the precision of the query vector while maintaining that of the support vectors, thereby minimizing the search iterations and improving efficiency in many-class scenarios. Finally, the Hardware-Aware Training (HAT) method optimizes controller training by modeling the hardware characteristics of MCAM, thus enhancing the reliability of the system. Our integrated framework reduces search iterations by up to 32 times, and increases overall accuracy by 1.58% to 6.94%.
Paper Structure (14 sections, 2 equations, 9 figures, 2 tables)

This paper contains 14 sections, 2 equations, 9 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background and Motivation
  3. Memory-Augmented Neural Network (MANN) for Few-Shot Learning (FSL)
  4. Mult-bit Content Addressable Memory
  5. Motivation
  6. Proposed Method
  7. Multi-bit Thermometer Code
  8. Asymmetric Vector Similarity Search
  9. Hardware-aware Training
  10. Evaluation and Results
  11. Experimental Setup
  12. Pareto Front of Energy-Accuracy Trade-off
  13. Comparison between SVSS and AVSS
  14. Conclusion

Figures (9)

  • Figure 1: The proposed processing flow using MCAM
  • Figure 2: (a) The Schematic of MCAM. (b) Simulated current distributions of MCAM with various string mismatch level. (c) Simulated current distributions of MCAM with 6-level string mismatch level, but with different maximum mismatch level in each string. The measured data points in (b) and (c) are derived from ref::mxic_mcam.
  • Figure 3: (a) Distribution of each type of mismatch level with B4E (b) occurrence probability of each type of mismatch level under difference distance
  • Figure 4: Illustration of (a) mapping vectors with large dimensions or long code word length in MCAM (b) the matching process in MCAM.
  • Figure 5: (a) Distribution of each type of mismatch level with MTMC (b) occurrence probability of each type of mismatch level under difference distance
  • ...and 4 more figures