Institution profile

Singapore University of Technology and Design

Academic institutionasia · sg
Official website
Research library596linked papers
Opportunities0open roles
Selected work

Representative Papers

A Survey on Generative Modeling with Limited Data, Few Shots, and Zero Shot

Jul 26, 2023arXiv.org

To address the limitation of conventional generative models (e.g., GANs, diffusion models) — their reliance on large-scale labeled data — in data-scarce domains such as medical imaging and remote sensing, this paper proposes a unified framework termed “Generative Modeling under Data Constraints” (GM-DC). We systematically establish a two-dimensional taxonomy: (i) task dimension—encompassing low-data, few-shot, and zero-shot settings; and (ii) methodological dimension—integrating transfer learning, meta-learning, prompt engineering, and multi-paradigm fusion. This work is the first to uncover cross-paradigm adaptation principles and synergistic mechanisms under data constraints. The survey comprehensively analyzes lightweight designs and knowledge transfer strategies for mainstream architectures—including VAEs, GANs, and diffusion models—and identifies critical research gaps while charting emerging trends. As the inaugural holistic GM-DC survey, it is accompanied by an open-source platform for continuous resource updates, providing both theoretical foundations and practical guidance for data-efficient generative modeling.

28 citationsRead paper

SpectrumFM: A Foundation Model for Intelligent Spectrum Management

May 02, 2025arXiv.org

To address the low recognition accuracy, slow convergence, and poor generalization of existing small-scale models in dynamic spectrum environments, this paper proposes SpectrumFM—a spectral foundation model. Methodologically, SpectrumFM integrates CNNs with multi-head self-attention to enhance IQ-signal representation learning; introduces the first foundation-model paradigm for spectrum analysis, featuring dual self-supervised pretraining tasks—masked signal reconstruction and next-time-step signal prediction; and employs parameter-efficient fine-tuning (e.g., LoRA) for cross-task transfer. Experiments demonstrate significant improvements: 12.1% higher accuracy in automatic modulation classification (AMC), 9.3% gain in wireless technology classification (WTC), an AUC of 0.97 for spectrum sensing at −4 dB SNR, over 10% improvement in anomaly detection performance, faster convergence, and markedly enhanced few-shot adaptation capability.

3 citations1 influentialRead paper

TransMode-LLM: Feature-Informed Natural Language Modeling with Domain-Enhanced Prompting for Travel Behavior Modeling

Jan 20, 2026

This study addresses the challenge of accurately predicting individual travel mode choice to support transportation planning and policy-making. It proposes a novel three-stage approach that integrates statistical feature analysis with large language models (LLMs): first identifying key trip characteristics, then converting structured data into natural language descriptions, and finally leveraging zero-shot or few-shot learning enhanced with domain-informed prompts for prediction. The method innovatively combines feature-driven natural language modeling with injected domain knowledge. Comprehensive evaluations assess the performance of GPT-4o, o3-mini, and o4-mini under few-shot and domain-enhanced prompting settings. Results show that LLM-based approaches achieve accuracy comparable to state-of-the-art classifiers, with o3-mini yielding up to a 42.9% improvement using only five examples. Domain-enhanced prompts significantly boost general-purpose models (e.g., GPT-4o gains of 2.27%–12.50%), though effects vary across reasoning-oriented architectures.

2 citationsRead paper

Directional Sparsity Based Statistical Channel Estimation for 6D Movable Antenna Communications

May 21, 2025

For six-degree-of-freedom movable antennas (6DMA), acquiring accurate statistical channel state information (CSI) incurs prohibitively high overhead and suffers from low estimation accuracy. This work first reveals a pronounced directional sparsity of the 6DMA channel in the joint position–orientation domain. Leveraging this insight, we propose a covariance-driven statistical CSI estimation framework that reconstructs the full-region average channel power distribution with high fidelity using only a small number of position–orientation samples. Our method jointly estimates multipath average power and directions of arrival (DOAs), circumventing explicit geometric propagation modeling. Experimental results demonstrate that, compared to baseline schemes, the proposed approach reduces pilot overhead by over 80% and decreases the mean squared error of channel power estimation by up to 45%, validating both the effectiveness of directional sparsity modeling and its practical feasibility for 6DMA systems.

2 citationsRead paper

Instance Temperature Knowledge Distillation

Jun 27, 2024arXiv.org

Existing knowledge distillation methods lack a long-term perspective in tuning the temperature parameter, optimizing only the immediate loss at each stage and thus failing to adapt to the dynamically varying difficulty across learning phases. This work formulates instance-level temperature selection as a sequential decision-making problem with delayed rewards and solves it via proximal policy optimization (PPO). We propose three key innovations: (1) an instance-aware dynamic state representation encoding per-sample difficulty and confidence; (2) a reward calibration mechanism to mitigate bias induced by delayed feedback; and (3) an efficient exploration strategy ensuring training stability. The resulting plug-and-play framework consistently improves performance across diverse distillation methods—e.g., KD, RKD, CRD—on both image classification and object detection benchmarks. It exhibits strong generalizability, modular design for easy integration, and requires no architectural modifications to teacher or student models.

2 citationsRead paper
Recent publications

Latest Papers

SinoGlyphBench: A Diagnostic Benchmark for Chinese Glyph-Level Obfuscation in Language-Model Moderation

Sep 05, 2026

Glyph-level obfuscation can leave harmful Chinese content readable to humans while degrading automated moderation. We introduce SinoGlyphBench, a diagnostic benchmark that identifies label-critical semantic anchors and creates matched original and glyph-obfuscated inputs in text and image modalities. By perturbing anchors, background context, or both, this design distinguishes corruption of moderation-relevant evidence from general surface variation. Across 176,916 paired evaluations of 12 LLMs and MLLMs, obfuscation increases harmful false-negative and false-positive rates by 6.1 and 4.7 percentage points, respectively, and reduces four-way accuracy by 5.0 points. Models retain 75.7% of the decisions that were correct on the matched original inputs. Full-scope perturbations cause the largest degradation, anchor-only perturbations are more damaging than background-only perturbations, and cross-script substitution is particularly difficult in the text modality. Analysis of structured outputs identifies observable mismatches in visible-form reading, intended-message recovery, and final safety judgment. The evaluated models, therefore, remain brittle to Chinese content written with non-canonical glyphs. Resources are available at https://github.com/fengshun124/SinoGlyphBench.

0 citationsRead paper