ChatSOP: An SOP-Guided MCTS Planning Framework for Controllable LLM Dialogue Agents

ChatSOP introduces an SOP-guided MCTS framework that improves LLM dialogue agents' controllability and task success by following structured operational procedures.

Zhigen Li, Jianxiang Peng, Yanmeng Wang, Yong Cao and 12 more

Published 2025 · 3 citations

Evaluating Neural Data Tokenizers: A Framework for Assessing Learned Representations of Spiking Activity

Federico D'Agostino, Alex Gilbert, Susanne Keller, Jaivardhan Kapoor and 16 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

When Trackers Fail: VLM-Guided Verification and Recovery for Robust Video Object Segmentation

Valay Mahesh Bundele, Susmit Agrawal, Mehran Hosseinzadeh, The Nam Nguyen and 1 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

Foveated BagNet: Inherent Interpretability Does Not Exclude Global Context

Holger Heidrich, Sarah Müller, Andreas Schilling

Paris Poster Session 4, Thu, Dec 10, 5:30 PM–7:30 PM, Paris Poster Hall · Published 2026

CRISP: Compositional Reasoning over Images via Stackable Programs for VLMs

Arnas Uselis, Yujin Jeong, Yanpeng Zhao, Alexander Rubinstein and 3 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

The Kernel Reality Check: Benchmarking and Distilling Efficient Attention at Scale

Firat Oncel, Cem Subakan, Mirco Ravanelli, Çağatay Yıldız

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

MedKIT: Evaluating Knowledge Integration and Generalization in Large Language Models

MedKIT evaluates medical LLM knowledge integration via clinical updates, revealing strong recall but limited relational, compositional, and operational generalization across 12 strategies.

Lukas Thede, Yash Kumar, David Chen, Danielle Bitterman and 3 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

The Dynamic-Probabilistic Consistency Gap in Chaotic Surrogate Modeling

Finite-horizon probabilistic training creates a consistency gap that decouples uncertainty from local dynamics in chaotic surrogates; a Kalman-aware framework evaluating local innovations while transporting covariance through learned Jacobians closes it.

Andre Herz, Matthijs Pals, Daniel Durstewitz, Georgia Koppe

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

Unlearning That Lasts: Utility-Preserving, Robust, and Almost Irreversible Forgetting in LLMs

JensUn uses Jensen-Shannon divergence to achieve stable, robust LLM unlearning with preserved utility and strong resistance to relearning.

Naman Deep Singh, Maximilian Mueller, Amit Peleg, Francesco Croce and 1 more

Paris Poster Session 1, Wed, Dec 9, 12:30 PM–2:30 PM, Paris Poster Hall · Published 2026

Affine Tracing: A New Paradigm for Probabilistic Linear Solvers

Affine tracing unifies probabilistic linear solvers by showing Bayesian methods are non-stationary affine iterative methods that are calibrated, and automatically generates probabilistic multigrid solvers via symbolic computation graphs.

Disha Hegde, Marvin Pförtner, Jon Cockayne

Paris Poster Session 4, Thu, Dec 10, 5:30 PM–7:30 PM, Paris Poster Hall · Published 2026

Half-Truths Break Similarity-Based Retrieval

CLIP-style dual encoders often prefer half-true image descriptions with incorrect added details over correct shorter ones due to weak part-level supervision; CS-CLIP improves half-truth accuracy to 69.3% through component-level contrastive fine-tuning.

Bora Kargi, Arnas Uselis, Seong Joon Oh

Paris Poster Session 4, Thu, Dec 10, 5:30 PM–7:30 PM, Paris Poster Hall · Published 2026 · ▲ 6 on Hugging Face · Code ★ 15

How to Scale Mixture-of-Experts: From muP to the Maximally Scale-Stable Parameterization

Researchers derive maximally scale-stable parameterizations for Mixture-of-Experts via dynamical mean-field theory, yielding robust learning-rate transfer and monotonic scaling gains across regimes.

Leena Chennuru Vankadara, Moritz Haas, Luke Hayward, Sebastian Bordt and 1 more

Paris Poster Session 6, Fri, Dec 11, 2:30 PM–4:30 PM, Paris Poster Hall · Published 2026 · ▲ 1 on Hugging Face · Code ★ 4

Closed-Form Last Layer Optimization

Closed-form last-layer optimization treats final weights as backbone-dependent functions, yielding convergence guarantees and outperforming SGD and Adam on regression tasks.

Alexandre Galashov, Nathaël Da Costa, Liyuan Xu, Philipp Hennig and 1 more

Paris Poster Session 6, Fri, Dec 11, 2:30 PM–4:30 PM, Paris Poster Hall · Published 2026