Yiheng Wu

Bernini Team, Chenchen Liu, Junyi Chen et al.

Multimodal large language models (MLLMs) and diffusion models have each reached remarkable maturity: MLLMs excel at reasoning over heterogeneous multimodal inputs with strong semantic grounding, while diffusion models synthesize images and videos with photorealistic fidelity. We argue that these two families can be unified through a simple division of labor: MLLMs perform semantic planning, while diffusion models render pixels from high-level semantic guidance and low-level visual features. Building on this idea, we propose Bernini, a unified framework for video generation and editing. An MLLM-based planner predicts the target semantic representation directly in the ViT embedding space, and a DiT-based renderer synthesizes pixels conditioned on this plan, augmented by text features and, for editing, source VAE features for detail preservation. Because semantics serve as the interface, the planner and renderer can be trained separately and only lightly co-trained, preserving the pretrained strengths of both components while keeping training efficient. To better handle multiple visual inputs, we introduce Segment-Aware 3D Rotary Positional Embedding (SA-3D RoPE), and further incorporate chain-of-thought reasoning in the planner to better transfer understanding into generation. Bernini achieves state-of-the-art performance across a wide range of video generation and editing benchmarks, with the MLLM's pretrained understanding translating into strong generalization on challenging editing tasks.

Yiheng Wu, Junhui Li, Muhua Zhu

Previous approaches to the task of implicit discourse relation recognition (IDRR) generally view it as a classification task. Even with pre-trained language models, like BERT and RoBERTa, IDRR still relies on complicated neural networks with multiple intermediate layers to proper capture the interaction between two discourse units. As a result, the outputs of these intermediate layers may have different capability in discriminating instances of different classes. To this end, we propose to adapt a supervised contrastive learning (CL) method, label- and instance-centered CL, to enhance representation learning. Moreover, we propose a novel constrained multi-layer CL approach to properly impose a constraint that the contrastive loss of higher layers should be smaller than that of lower layers. Experimental results on PDTB 2.0 and PDTB 3.0 show that our approach can significantly improve the performance on both multi-class classification and binary classification.

Yiheng Wu, Roman Yangarber, Xian Mao

The remarkable capabilities of Large Language Models (LLMs) in text comprehension and generation have revolutionized Information Extraction (IE). One such advancement is in Document-level Relation Triplet Extraction (DocRTE), a critical task in information systems that aims to extract entities and their semantic relationships from documents. However, existing methods are primarily designed for Sentence level Relation Triplet Extraction (SentRTE), which typically handles a limited set of relations and triplet facts within a single sentence. Additionally, some approaches treat relations as candidate choices integrated into prompt templates, resulting in inefficient processing and suboptimal performance when determining the relation elements in triplets. To address these limitations, we introduce a Discriminative and Voice Aware Paradigm DiVA. DiVA involves only two steps: performing document-level relation extraction (DocRE) and then identifying the subject object entities based on the relation. No additional processing is required simply input the document to directly obtain the triplets. This streamlined process more accurately reflects real-world scenarios for triplet extraction. Our innovation lies in transforming DocRE into a discriminative task, where the model pays attention to each relation and to the often overlooked issue of active vs. passive voice within the triplet. Our experiments on the Re-DocRED and DocRED datasets demonstrate state-of-the-art results for the DocRTE task.