Adaptive Filter Pruning via One-Shot Ratio Search and In-Training Sparsity Scheduling 


Vol. 15,  No. 4, pp. 349-358, Apr.  2026
https://doi.org/10.3745/TKIPS.2026.15.4.349


PDF
  Abstract

One-shot pruning is computationally efficient, yet selecting architecture-specific layer sparsity ratios in advance is difficult and can cause severe accuracy collapse. This paper proposes ASE(Adaptive Sensitivity Estimation)-OneShot Pruning, which combines filter-sensitivity probing with performance feedback to automatically derive layer-wise pruning ratios. We further present Gradient-Adaptive EMA Pruning, which uses the exponential moving average (EMA) of training gradients to schedule sparsification and mitigate pruning shock in efficiency-optimized architectures. Across diverse model families, the proposed methods adaptively balance layer-wise protection and sparsification, preserving accuracy under high compression. In particular, selective protection stabilizes deep models, while EMA-guided scheduling promotes elastic recovery after pruning in efficiency-optimized networks, suggesting a robust, architecture-agnostic solution for reliable layer-wise pruning.

  Statistics
  Cite this article

[IEEE Style]

H. Lee and Y. Moon, "Adaptive Filter Pruning via One-Shot Ratio Search and In-Training Sparsity Scheduling," The Transactions of the Korea Information Processing Society, vol. 15, no. 4, pp. 349-358, 2026. DOI: https://doi.org/10.3745/TKIPS.2026.15.4.349.

[ACM Style]

Hyunsu Lee and Yong-Hyuk Moon. 2026. Adaptive Filter Pruning via One-Shot Ratio Search and In-Training Sparsity Scheduling. The Transactions of the Korea Information Processing Society, 15, 4, (2026), 349-358. DOI: https://doi.org/10.3745/TKIPS.2026.15.4.349.

Forms

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)

Old Journals

Indexing

All publications of TKIPS is indexed in DOI, EBSCO, Google Scholar, Crossref, and CrossCheck.

TKIPS is also selected as the Journal for Accreditation by NRF (National Research Foundation of Korea).

Related Journals