Flash-BoN: Instant Drafts for Inference-Time Scaling in Diffusion Models
ECCV 2026, 2026
Test Time Scaling of Diffusion Models Done Right
Publications
Catalog-Native LLM: Speaking Item-ID dialect with Less Entanglement for Recommendation
ICLR 2026, 2026
IDIOMoE decouples item-ID and language processing inside an LLM via token-type MoE, reducing interference and improving recommendation quality at roughly the same compute
Cost-Aware Contrastive Routing for LLMs
NeurIPS 2025 Spotlight, 2025
CSCR embeds both prompts and LLMs into a shared space using fast logit or perplexity fingerprints. A cost‑banded InfoNCE loss trains the space to balance quality against cost. It generalizes to unseen models and out‑of‑distribution prompts.
Bilevel ZOFO: Bridging Parameter-Efficient and Zeroth-Order Techniques for Efficient LLM Fine-Tuning and Meta-Training
NeurIPS 2025, 2025
We propose Bilevel ZOFO, a novel bilevel optimization framework that bridges parameter-efficient and zeroth-order optimization techniques for efficient large language model (LLM) fine-tuning and meta-training.
ARGUS: Hallucination and Omission Evaluation in Video-LLMs
ICCV 2025, 2025
An evaluation framework and benchmark for free-form video-LLMs hallucination and omission.
ToMoE: Converting Dense Large Language Models to Mixture-of-Experts through Dynamic Structural Pruning
TMLR @ICML 2026, 2025
We propose ToMoE, a novel method for converting dense large language models to mixture-of-experts through dynamic structural pruning.
Efficient Fine-Tuning and Concept Suppression for Pruned Diffusion Models
CVPR 2025, 2024
How to train a small diffusion model while suppressing unwanted concepts
Not All Prompts Are Made Equal: Prompt-based Pruning of Text-to-Image Diffusion Models
ICLR 2025, 2024
Dynamic Prompt-based Pruning of Text-to-Image Diffusion Models
From Pixels to Prose: A Large Dataset of Dense Image Captions
ArXiv, 2024
PixelProse is a comprehensive dataset of over 16M (million) synthetically generated captions, leveraging cutting-edge vision-language models for detailed and accurate descriptions.
Deep Prompt Tuning for Graph Transformers
ArXiv, 2023
Prompt tuning for Graph Transformers and Message Passing Graph Neural Networks, improved efficiency and resource utilization of Graph Transformers.
Incomplete Multimodal Learning for Complex Brain Disorders Prediction
ArXiv, 2023
We propose a new incomplete multimodal data integration approach that employs transformers and generative adversarial networks to effectively exploit auxiliary modalities available during training in order to improve the performance of a unimodal model at inference.
Dementia Severity Classification under Small Sample Size and Weak Supervision in Thick Slice MRI
ArXiv, 2021
we propose to classify the disease severity based on the Fazekas scale through the visual biomarkers, namely the Periventricular White Matter (PVWM) and the Deep White Matter (DWM) changes, in the real-world setting of thick-slice MRI