Top 10 Research Papers: December 2, 2025

Stay up-to-date with the latest advancements in research! This article summarizes the top 10 research papers published recently, covering exciting topics in multimodal learning, audio-visual processing, and more. For a better reading experience and more papers, be sure to check out the Github page.

Omni

This section highlights research papers focusing on omnimodal learning, which aims to create models that can understand and process information from various modalities (e.g., text, images, audio, video) in a unified way. These papers explore innovative approaches to building versatile and powerful AI systems.

1. OralGPT-Omni: A Versatile Dental Multimodal Large Language Model

Published on November 27, 2025, this paper introduces OralGPT-Omni, a multimodal large language model specifically designed for dental applications. This cutting-edge research explores how AI can be used to analyze and understand complex dental data, potentially revolutionizing dental diagnostics and treatment planning. The 47-page paper, featuring 42 figures and 13 tables, delves into the architecture, training, and performance of OralGPT-Omni. The development of such a specialized model highlights the growing trend of applying AI to niche domains, promising more accurate and efficient solutions tailored to specific needs. OralGPT-Omni's ability to process diverse dental data, including images, text, and potentially even audio, positions it as a versatile tool for dental professionals. This research contributes significantly to the field by demonstrating the potential of multimodal AI in healthcare and setting a precedent for future developments in the area.

2. Uni-MoE-2.0-Omni: Scaling Language-Centric Omnimodal Large Model with Advanced MoE, Training and Data

This paper, published on November 23, 2025, presents Uni-MoE-2.0-Omni, an advanced omnimodal large model that scales language-centric learning using Mixture-of-Experts (MoE). With 47 pages and 10 figures, this research details the model's architecture, training methodologies, and data utilization strategies. The project website (https://idealistxy.github.io/Uni-MoE-v2.github.io/) and code repository (https://github.com/HITsz-TMG/Uni-MoE) provide further resources for those interested in replicating or extending this work. Uni-MoE-2.0-Omni's innovative approach to scaling omnimodal models addresses the critical challenge of efficiently processing vast amounts of multimodal data. The use of MoE allows the model to selectively activate different parts of its network based on the input, improving both performance and computational efficiency. This research represents a significant step forward in the development of large-scale AI systems capable of handling the complexity of real-world multimodal information.

3. T-Rex-Omni: Integrating Negative Visual Prompt in Generic Object Detection

Accepted by AAAI 2026, this paper introduces T-Rex-Omni, a novel approach to generic object detection that integrates negative visual prompts. The main paper spans 7 pages with 4 figures, while the appendix adds 8 pages with 7 figures. T-Rex-Omni's integration of negative visual prompts represents a significant advancement in object detection, allowing the model to better differentiate between objects and background. This approach can lead to more accurate and robust object detection systems, which are crucial for applications such as autonomous driving, robotics, and image analysis. The acceptance of this paper by AAAI 2026 underscores its importance and potential impact on the field of computer vision.

4. Omni-AVSR: Towards Unified Multimodal Speech Recognition with Large Language Models

Published on November 10, 2025, this paper explores Omni-AVSR, a unified multimodal speech recognition system leveraging large language models. The project website (https://umbertocappellazzo.github.io/Omni-AVSR/) provides additional details and resources. Omni-AVSR's use of large language models for multimodal speech recognition is a promising direction, as these models have demonstrated remarkable capabilities in natural language processing. By combining audio and visual information, Omni-AVSR can potentially achieve higher accuracy and robustness compared to traditional speech recognition systems. This research contributes to the ongoing effort to develop more human-like AI systems that can understand and interact with the world through multiple senses.

5. Omni-View: Unlocking How Generation Facilitates Understanding in Unified 3D Model based on Multiview images

This paper, also published on November 10, 2025, and currently under review, investigates Omni-View, a method for understanding unified 3D models from multiview images by leveraging generative techniques. This research explores how AI can generate and interpret 3D models from 2D images, a crucial capability for applications in computer vision, robotics, and virtual reality. Omni-View's approach to 3D model understanding through generation represents a novel and potentially impactful direction in the field. The ability to accurately reconstruct 3D models from multiple viewpoints is essential for enabling AI systems to perceive and interact with the physical world.

Audio Visual

This section showcases research papers focused on audio-visual processing, a field that combines the analysis of both audio and visual data. This area is critical for developing AI systems that can understand and interact with the world in a more natural and human-like way, as humans often rely on both sight and sound to perceive their surroundings.

1. Audio-Visual World Models: Towards Multisensory Imagination in Sight and Sound

Published on November 30, 2025, this paper delves into the creation of audio-visual world models, aiming to develop AI systems capable of multisensory imagination through sight and sound. This ambitious research seeks to build AI models that can not only perceive the world through audio and visual inputs but also imagine and predict future states based on these sensory inputs. The development of such world models is a crucial step towards creating truly intelligent systems that can reason and act in complex environments. The potential applications of audio-visual world models are vast, ranging from robotics and autonomous navigation to virtual reality and entertainment.

2. CACARA: Cross-Modal Alignment Leveraging a Text-Centric Approach for Cost-Effective Multimodal and Multilingual Learning

This paper, published on November 29, 2025, introduces CACARA, a cross-modal alignment method that leverages a text-centric approach for cost-effective multimodal and multilingual learning. Spanning 25 pages with 12 tables and 5 figures, this research details the methodology and performance of CACARA. CACARA's text-centric approach to cross-modal alignment offers a promising solution for the challenges of multimodal and multilingual learning. By using text as a bridge between different modalities and languages, CACARA can effectively transfer knowledge and improve performance in resource-scarce scenarios. This research is particularly relevant in the context of global AI development, where the ability to process diverse languages and modalities is essential.

3. MVAD : A Comprehensive Multimodal Video-Audio Dataset for AIGC Detection

Published on November 29, 2025, this paper presents MVAD, a comprehensive multimodal video-audio dataset designed for AIGC (AI-Generated Content) detection. The 7-page paper includes 2 figures and provides a detailed description of the dataset. MVAD fills a critical need in the AI community by providing a high-quality dataset for training and evaluating AIGC detection models. The increasing prevalence of AI-generated content necessitates the development of robust detection methods to combat misinformation and ensure the authenticity of online media. This research contributes significantly to this effort by providing a valuable resource for researchers and practitioners working in this area.

4. Design and Evaluation of a Multi-Agent Perception System for Autonomous Flying Networks

This paper, also published on November 29, 2025, focuses on the design and evaluation of a multi-agent perception system for autonomous flying networks. This research explores the challenges and opportunities of using multiple agents (e.g., drones) to collaboratively perceive and understand their environment. Multi-agent perception systems are crucial for enabling autonomous flying networks to operate safely and efficiently in complex environments. The potential applications of this research include aerial surveillance, delivery services, and search and rescue operations.

5. OmniFusion: Simultaneous Multilingual Multimodal Translations via Modular Fusion

This paper, published on November 28, 2025, and submitted as a preprint for ACL 2026, introduces OmniFusion, a method for simultaneous multilingual multimodal translations via modular fusion. OmniFusion's modular fusion approach offers a flexible and efficient way to handle the complexities of multilingual multimodal translation. By breaking down the translation process into modular components, OmniFusion can effectively leverage the strengths of different modalities and languages. This research represents a significant step forward in the development of AI systems capable of seamlessly translating information across multiple languages and modalities.

Conclusion

The research papers highlighted in this article represent the cutting edge of AI research in omnimodal and audio-visual learning. These advancements pave the way for more versatile, intelligent, and human-like AI systems. Stay tuned for more updates on the latest research in these exciting fields. For further reading on related topics, consider exploring the AI Journal.