Boost Multi-Token Prediction with Self-Distillation
Enhance large language model efficiency with MTP-D, a self-distillation method that improves multi-token prediction and speeds up inference by 220%.
High-Fidelity Diffusion Inversion for Real-World Image Editing
Enhance real-world image reconstruction and manipulation with latent bias alignment techniques for high-fidelity diffusion inversion.
KDC-Net: Advanced Dual Context Network for Video Retrieval
Discover KDC-Net, a dual context-aware network enhancing partially relevant video retrieval with semantic aggregation and dynamic temporal attention.
SM-Net: Continuous Spectral Manifold from Stellar Libraries
Discover SM-Net, a fast ML model generating stellar spectra from multiple high-res libraries using key stellar parameters for astrophysics research.
Human-AI Epistemic Partnership in Generative AI UX
Explore the Human-AI Epistemic Partnership Theory to enhance user experience with Generative AI in education, balancing trust, agency, and accountability.