Institution profile

Nazarbayev University

Academic institutionasia · kz
Official website
Research library20linked papers
Opportunities0open roles
Selected work

Representative Papers

AI Generated Text Detection

Jan 07, 2026arXiv.org

This study addresses the growing concern of academic integrity violations due to students’ misuse of large language models (LLMs) for text generation by establishing a unified benchmark to systematically evaluate various AI-generated text detection methods. To mitigate information leakage caused by topic memorization and enhance generalization to unseen domains, the work introduces an innovative topic-based data partitioning strategy. The evaluation encompasses both traditional approaches—such as TF-IDF combined with logistic regression—and deep learning architectures, including BiLSTM and DistilBERT, with plans to incorporate parameter-efficient fine-tuning techniques like LoRA. Experimental results demonstrate that DistilBERT achieves the best performance with 88.11% accuracy and a 0.96 ROC-AUC score, while BiLSTM also attains 88.86% accuracy, significantly outperforming baseline methods and underscoring the advantage of semantic modeling in AI-generated text detection.

2 citationsRead paper

Benchmarking Dexterity of Multifingered Robot Hands: A Review and Perspective

Sep 04, 2026

Robot hands are a key interface between AI and the physical world, making advances in robotic dexterity essential to realizing the vision of physical AI. While impressive dexterity has been demonstrated with simple grippers, multifingered hands offer the potential for substantially greater versatility, precision, and adaptability in manipulation. In this review, we survey the state of the art in benchmarking the dexterity of multifingered robot hands. Recognizing dexterity as a complex and multifaceted concept, we present the perspective of the U.S. National Science Foundation HAND Engineering Research Center, with a particular focus on fine in-hand manipulation. We introduce a framework consisting of three benchmark levels that correspond to increasing system complexity, review representative benchmarks at each level, and propose new benchmarks and metrics to address limitations in the literature. More information can be found at https://hand-erc.github.io/benchmarking/.

0 citationsRead paper
Recent publications

Latest Papers

Benchmarking Dexterity of Multifingered Robot Hands: A Review and Perspective

Sep 04, 2026

Robot hands are a key interface between AI and the physical world, making advances in robotic dexterity essential to realizing the vision of physical AI. While impressive dexterity has been demonstrated with simple grippers, multifingered hands offer the potential for substantially greater versatility, precision, and adaptability in manipulation. In this review, we survey the state of the art in benchmarking the dexterity of multifingered robot hands. Recognizing dexterity as a complex and multifaceted concept, we present the perspective of the U.S. National Science Foundation HAND Engineering Research Center, with a particular focus on fine in-hand manipulation. We introduce a framework consisting of three benchmark levels that correspond to increasing system complexity, review representative benchmarks at each level, and propose new benchmarks and metrics to address limitations in the literature. More information can be found at https://hand-erc.github.io/benchmarking/.

0 citationsRead paper

Energy- and Memory-Efficient PEFT Methods for Personalized On-Device SLMs on Consumer GPUs

Aug 05, 2026

Deploying personalized large language models on resource-constrained edge devices faces significant challenges in GPU memory consumption, energy usage, and computational latency. This work systematically evaluates five parameter-efficient fine-tuning (PEFT) methods—LoRA, LoRA+, QLoRA, BitFit, and full fine-tuning—on small language models (SLMs) based on both Transformer and Mamba architectures across general and personalized tasks. The study introduces NetScore-E/M, a unified evaluation framework that jointly accounts for energy and memory constraints, and establishes an energy-aware PEFT selection strategy. Experimental results demonstrate that LoRA+ achieves the best overall performance, while QLoRA substantially reduces memory footprint. Among evaluated models, TinyLlama-1.1B consistently excels across multiple benchmarks, confirming the feasibility of deploying SLMs with PEFT for efficient on-device personalization.

0 citationsRead paper