Prépublication
· 2026 arXiv
EN
Relations compose through the entity they share - Alice is Bob's mother and Bob is Carol's father, so Alice is Carol's grandmother - and the order matters, since a father's mother is not a mother's father; verifying a composed relation means closing the loop....
Aakash Sarkar, Ananya Mukherjee et al.
#cs.LG
Prépublication
· 2026 arXiv
EN
Atmospheric chemical mechanisms are commonly distributed in the language of the Kinetic PreProcessor (KPP), which generates efficient Fortran or C code but does not fit into the differentiable programming frameworks in which machine-learning models are develop...
Wenqing Peng, Michael Boy et al.
#cs.AI
Prépublication
· 2026 arXiv
EN
Large Language Models (LLMs) are increasingly used in multi-turn agentic workloads that require sustained interaction with complex environments. Existing reflective prompt optimizers such as GEPA learn from failed rollouts, but agentic failures can also come f...
Zack Yu, Nicholas Lee et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Scientific discovery often begins with an unexpected observation that prompts a new research question. In a world with altered physical laws, can large language models (LLMs) notice unexpected behavior, such as heavier objects falling more slowly, and discover...
Zhen Yang, Wen Wang et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Weight decay and explicit weight-norm constraints improve optimizer performance in language model pretraining, yet a unified understanding of why such control helps and how much is needed remains incomplete. We analyze norm control in Muon over the entire trai...
Athanasios Glentis, Dawei Li et al.
#cs.LG
Prépublication
· 2026 arXiv
EN
Self-attention plays a key role in long-sequence modeling, but its quadratic time complexity fundamentally limits the scaling of Large Language Models. Recently, Multi-State Linear Attention (MSLA) mechanisms have been proposed to make sequence modeling more e...
Hee Seung Hwang, Sanghyuk Chun et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Diffusion language models offer a promising path to efficient agents through parallel token generation. However, efficient interaction requires adapting both where computation is spent and how much text is generated. We introduce a metacognitive controller tha...
Mingyu Zhao, Shuhang Lin et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Language models pretrained on web corpora collected over multiple years learn both general and domain-specific knowledge. How these different kinds of knowledge are learned as pretraining progresses remains unclear. Recent point-in-time models trained on web s...
Chang Su, Bin Wu et al.
#cs.AI
Prépublication
· 2026 arXiv
EN
When an LLM group's discussion is compressed into a short memory for a final decision maker, what should the memory keep? Two natural heuristics keep the rarest claims or the claims relevant to the decision criteria. On hidden-profile tasks both look attractiv...
Anouar Imel, Imad Charradi et al.
#cs.AI
Prépublication
· 2026 arXiv
EN
Chain-of-thought reasoning has substantially advanced video understanding in multimodal large language models (MLLMs). However, language-based reasoning traces do not provide a reliable mechanism for models to revisit relevant visual evidence, leading to visua...
Shilong Li, Xingyuan Bu et al.
#cs.CV
Prépublication
· 2026 arXiv
EN
As executable environments for language agents scale, generating and verifying their rollouts becomes increasingly expensive. Under finite rollout budgets, this raises an important question: can data management at the environment level make agentic reinforceme...
Sihan Yang, Weichen Bai et al.
#cs.LG
Prépublication
· 2026 arXiv
EN
For audio-language models, an answer change after a local sound-event edit can reflect a response to the event's class, sensitivity to waveform rewriting, or both. We compare matched same-class and different-class replacements of a queried event. Both use the...
Yunuo Wang, Maoyuan Chen et al.
#cs.SD