Sangmin Bae

Composer: A Search Framework for Hybrid Neural Architecture Design

Temporal Alignment Guidance: On-manifold Sampling in Diffusion Models

Hybrid Architectures for Language Models: Systematic Analysis and Design Insights

Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation

Multi-Task Corrupted Prediction for Learning Robust Audio-Visual Speech Representation

Relaxed Recursive Transformers: Effective Parameter Sharing with Layer-wise LoRA

Automated Filtering of Human Feedback Data for Aligning Text-to-Image Diffusion Models

MoHAVE: Mixture of Hierarchical Audio-Visual Experts for Robust Speech Recognition

Fine-tuned In-Context Learning Transformers are Excellent Tabular Data Classifiers.