Energy-Efficient Transformer Inference: Optimization Strategies for Time Series Classification

Arshia Kermani Department of Computer Science, Texas State University [email protected], Ehsan Zeraatkar Department of Computer Science, Texas State University [email protected], Habib Irani Department of Computer Science, Texas State University [email protected] · 2025

Read Energy-Efficient Transformer Inference: Optimization Strategies for Time Series Classification by Arshia Kermani Department of Computer Science, Texas State University [email protected], Ehsan Zeraatkar Department of Computer Science, Texas State University [email protected], Habib Irani Department of Computer Science, Texas State University [email protected] (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-16627.txt.

Description

Arshia Kermani, Ehsan Zeraatkar, and Habib Irani test two ways to slim transformer models for time series sorting so they sip rather than gulp energy. Their paper appeared in two thousand twenty-five. Transformers classify sequences well but cost heavily at run time. Devices at the edge cannot afford that appetite.

Across three data sets, from fridges to household power traces, they tried pruning and quantization in several settings. Static quantization cut energy use by a wide margin while holding accuracy. Pruning by weight size sped inference sharply with little accuracy loss. The two tricks attack different costs: one trims arithmetic, the other trims the network itself. Their numbers give deployers a menu for trading accuracy against watts. The foundation suits sensors and appliances that must think locally.

Catalog record

CREATORArshia Kermani Department of Computer Science, Texas State University [email protected], Ehsan Zeraatkar Department of Computer Science, Texas State University [email protected], Habib Irani Department of Computer Science, Texas State University [email protected]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-16627

Browse the full library · About the Tower of Babel