Selected work

Representative Papers

FRENCH-YMCA: A FRENCH Corpus meeting the language needs of Youth, froM Children to Adolescents

Apr 07, 2026

Children and adolescents are in a critical period of language development, exhibiting linguistic characteristics that markedly differ from those of adults; however, age-appropriate, large-scale, and high-quality French textual resources remain scarce. To address this gap, this study presents French-YMCA, the first systematically constructed open-access corpus spanning the full developmental range from childhood through adolescence. Comprising 39,200 normative texts totaling over 22.47 million words, the corpus was assembled through multi-source collection, rigorous cleaning, and standardized processing. An accompanying open-access platform ensures broad usability. French-YMCA fills a critical void in French youth language resources and provides essential support for training age-adapted language models and enhancing the comprehensibility and developmental appropriateness of digital interactions.

0 citationsRead paper

Ternarization of Vision Language Models for use on edge devices

Apr 07, 2025

To enable efficient deployment of vision-language models (VLMs) on resource-constrained edge devices, this work proposes a direct ternarization compression paradigm for pre-trained VLMs—bypassing costly from-scratch training. Methodologically: (1) we introduce a k-means–driven weight clustering initialization strategy to accelerate ternarization convergence and improve accuracy; (2) we implement high-precision ternary matrix multiplication and develop custom TensorFlow Lite–native ternary operators. Our key contributions include the first end-to-end ternarization framework for pre-trained VLMs; achieving ~13× memory reduction over full-precision counterparts while maintaining near-original perplexity (increase < 0.5%) and outperforming binary models in token generation latency; and establishing the state-of-the-art trade-off between accuracy and efficiency among ternary VLM compression methods.

0 citationsRead paper

Recent publications

Latest Papers

FRENCH-YMCA: A FRENCH Corpus meeting the language needs of Youth, froM Children to Adolescents

Apr 07, 2026

Children and adolescents are in a critical period of language development, exhibiting linguistic characteristics that markedly differ from those of adults; however, age-appropriate, large-scale, and high-quality French textual resources remain scarce. To address this gap, this study presents French-YMCA, the first systematically constructed open-access corpus spanning the full developmental range from childhood through adolescence. Comprising 39,200 normative texts totaling over 22.47 million words, the corpus was assembled through multi-source collection, rigorous cleaning, and standardized processing. An accompanying open-access platform ensures broad usability. French-YMCA fills a critical void in French youth language resources and provides essential support for training age-adapted language models and enhancing the comprehensibility and developmental appropriateness of digital interactions.

0 citationsRead paper

Ternarization of Vision Language Models for use on edge devices

Apr 07, 2025

To enable efficient deployment of vision-language models (VLMs) on resource-constrained edge devices, this work proposes a direct ternarization compression paradigm for pre-trained VLMs—bypassing costly from-scratch training. Methodologically: (1) we introduce a k-means–driven weight clustering initialization strategy to accelerate ternarization convergence and improve accuracy; (2) we implement high-precision ternary matrix multiplication and develop custom TensorFlow Lite–native ternary operators. Our key contributions include the first end-to-end ternarization framework for pre-trained VLMs; achieving ~13× memory reduction over full-precision counterparts while maintaining near-original perplexity (increase < 0.5%) and outperforming binary models in token generation latency; and establishing the state-of-the-art trade-off between accuracy and efficiency among ternary VLM compression methods.

0 citationsRead paper