作者位次 #4/5ICLR 2026Poster· Theory / Interpretability and Visualization
Compose and Fuse: Revisiting the Foundational Bottlenecks in Multimodal Reasoning
Yucheng Wang · Yifan Hou · Aydin Javadov · Mubashara Akhtar · Mrinmaya Sachan
首次获取该候选人的完整档案及 StartLux 评估共 20 积分；每位用户单独解锁，之后查看免费。主动刷新 20 积分。
实时查找电子邮箱，费用不超过20积分；未找到邮箱不扣费。
Yucheng Wang · Yifan Hou · Aydin Javadov · Mubashara Akhtar · Mrinmaya Sachan
Yu Fan · Jingwei Ni · Jakob Merane · Yang Tian · Yoan Hermstrüwer · Yinya Huang · Mubashara Akhtar · Etienne Salimbeni · Florian Geering · Oliver Dreyer · Daniel Brunner · Markus Leippold · Mrinmaya Sachan · Alexander Stremitzer · Christoph Engel · Elliott Ash · Joel Niklaus
Jingwei Ni · Ekaterina Fadeeva · Tianyi Wu · Mubashara Akhtar · Jiaheng Zhang · Elliott Ash · Markus Leippold · Timothy Baldwin · See-Kiong Ng · Artem Shelmanov · Mrinmaya Sachan
LLMs can solve complex tasks by generating long, multi-step reasoning chains. Test-time scaling (TTS) can further improve LLM performance by sampling multiple variants of intermediate reasoning steps, verifying their correctness, and strategically choosing the best steps for continuation. However...