인포메이션시스템즈

We study discrete token diffusion for language generation to understand sampling behavior, convergence, and model dynamics. Publications emphasize how discrete diffusion reveals new insights into text modeling. We develop training methods that protect data and reduce harmful outputs. Our work combines differential privacy, robust training, and safety-aligned model design. Large language models must be made efficient for real-world use. We focus on low-bit quantization, compression-aware training, and model scaling that preserves performance. We investigate learning dynamics, optimization landscapes, and theoretical foundations of deep networks. This research helps connect empirical results with principled understanding. Artificial Intelligence and Information Systems Laboratory (AI-ISL) is a deep learning research lab focused on information-theoretic approaches to machine learning problems. We focus on the theoretical understanding of diverse deep learning problems, including generative models, privacy and safety, and network compression.