Research map / Vision-language and multimodal / Multimodal LLMs
AURELIA: Test-time Reasoning Distillation in Audio-Visual LLMs
Sanjoy Chowdhury, Hanan Gani, Nishit Anand, Sayan Nag, Ruohan Gao, Mohamed Elhoseiny, Salman Khan, Dinesh Manocha
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add AURELIA reasoning distillation | VITA · AVReasonBench AV-GeoIQ | accuracy: 26.7 -> 39.0 | Improved |
| add AURELIA reasoning distillation | VITA · AVReasonBench AV-Captioning (CIDEr) | CIDEr: 73.8 -> 78.8 | Improved |
| add AURELIA reasoning distillation | VITA · AVReasonBench AV-Meme | accuracy: 44.0 -> 50.0 | Improved |
| add AURELIA reasoning distillation | X-InstructBLIP · AV-GeoIQ | accuracy: 14.5 -> 29.0 | Improved |
| add AURELIA reasoning distillation | VITA · Video-MME overall | accuracy: 80.7 -> 83.73 | Improved |
| add AURELIA reasoning distillation | VITA · Video-MME long videos | accuracy: 73.3 -> 76.98 | Improved |
| add AURELIA reasoning distillation | VITA · AV-Captioning | BLEU@4: 12.9 -> 14.5 | Improved |
| add AURELIA reasoning distillation | PandaGPT · AV-GeoIQ | accuracy: 12.5 -> 25.0 | Improved |
| use GPT-4o as summarizer instead of Gemini for all agents (ablation) | AURELIA (VideoLLaMA-2) · AV-GeoIQ | accuracy: 36.5 -> 38.0 | Improved |
| use GPT-4o as summarizer instead of Gemini for all agents (ablation) | AURELIA (VideoLLaMA-2) · DM-Match | accuracy: 33.0 -> 34.5 | Improved |
| increase feedback iterations from 1 to 5 (ablation) | AURELIA (VideoLLaMA-2) · AV-GeoIQ | accuracy: 27.5 -> 38.0 | Improved |
| increase iterations from 1 to 5 (ablation) | AURELIA (VideoLLaMA-2) · AV-Comp | accuracy: 27.3 -> 31.6 | Mixed |
Closest papers by meaning
- MMAR: A Challenging Benchmark for Deep Reasoning in Speech, Audio, Music, and Their Mix (NeurIPS 2025 · similarity 0.90)
- Mellow: a small audio language model for reasoning (NeurIPS 2025 · similarity 0.89)
- OmniVideoBench: Towards Audio-Visual Understanding Evaluation for Omni MLLMs (ICLR 2026 · similarity 0.89)
- Incentivizing Consistent, Effective and Scalable Reasoning Capability in Audio LLMs via Reasoning Process Rewards (ICLR 2026 · similarity 0.89)
- When Eyes and Ears Disagree: Can MLLMs Discern Audio-Visual Confusion? (AAAI 2026 · similarity 0.89)
- JointAVBench: A Benchmark for Joint Audio-Visual Reasoning Evaluation (ICLR 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
