Online vs. Offline Active Learning: Performance Comparison Across Iterations

Table of Links

Abstract and 1 Introduction

1. Related work

   2.1. Generative Data Augmentation

   2.2. Active Learning and Data Analysis

2. Preliminary

3. Our method

   4.1. Estimation of Contribution in the Ideal Scenario

   4.2. Batched Streaming Generative Active Learning

4. Experiments and 5.1. Offline Setting

   5.2. Online Setting

5. Conclusion, Broader Impact, and References

   \

A. Implementation Details

B. More ablations

C. Discussion

D. Visualization

B. More ablations

Momentum Coefficient β. In Algorithm 3, we introduce a momentum coefficient β to update the grad cache. Here we explore the effect of different β on the model performance. A larger beta signifies a greater focus on global information, while a smaller beta indicates a higher attention to the current test batch Ub. Detailed results are presented in Table 9. Observations suggest that when β is 0.1, the performance is the best, which is also the β we finally adopted.

\

\ Contribution threshold τ. In Algorithm 3, we incorporate a contribution threshold τ , intended for filtering the produced data. Here we investigate the impact of varying values of τ on the model’s performance. The larger τ implies a stricter filtration of the generated data, while the smaller τ signifies a looser filtering of the generated data. The specific results are shown in Table 10. We can see the performance is optimal when τ equals -0.05, which is also the τ we eventually settle on for our final model.

\

\ Online learning vs. Offline learning We compare online learning and offline learning under different iterations. The result is shown in Figure 9.

\

:::info Authors:

(1) Muzhi Zhu, with equal contribution from Zhejiang University, China;

(2) Chengxiang Fan, with equal contribution from Zhejiang University, China;

(3) Hao Chen, Zhejiang University, China (haochen.cad@zju.edu.cn);

(4) Yang Liu, Zhejiang University, China;

(5) Weian Mao, Zhejiang University, China and The University of Adelaide, Australia;

(6) Xiaogang Xu, Zhejiang University, China;

(7) Chunhua Shen, Zhejiang University, China (chunhuashen@zju.edu.cn).

:::

:::info This paper is available on arxiv under CC BY-NC-ND 4.0 Deed (Attribution-Noncommercial-Noderivs 4.0 International) license.

:::

\