Learning to Read the Contextual Tokens in Diffusion Transformers

A framework maps diffusion transformer contextual tokens through a frozen LLM to reveal they encode global emerging scene semantics early, inspiring contextual alignment that improves generation quality.

Omer Dahary, Etai Sella, Hadar Averbuch-Elor, Daniel Cohen-Or and 1 more

Published Oct 5, 2026 · ▲ 5 on Hugging Face

Auditing Routing Entropy as an Uncertainty Signal in Attention-Residual Transformers

Routing entropy in attention-residual transformers fails as a reliability signal beyond model confidence, with no robust calibration gains and low sensitivity to injected effects.

Wenhao Liang, Lin Yue, Wei Emma Zhang, Mingyu Guo and 2 more

Published Oct 1, 2026 · 0 citations

Monitoring the Internal Monologue: Probe Trajectories Reveal Reasoning Dynamics

Maciej Chrabaszcz, Aleksander Szymczyk, Marcin Sendera, Tomasz Trzcinski and 1 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

From Pixels to Concepts: Do Segmentation Models Understand What They Segment?

Shuang Liang, Zeqing Wang, Yuxian LI, Xihui Liu and 1 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · Code ★ 5

Hierarchical Concept Geometry in Language Representations Emerges from Word Co-occurrence

Andres Nava, Matthieu Wyart

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

Structural Blindness in Latent Data Assimilation: Representation Geometry Misleads Sensor Design

Yongquan Qu, Pierre Gentine

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

Learning Interpretable Switching Dynamics in Shared Neural-Behavioral Latent Space

Yongxu Zhang, Josue Ortega Caro, Rachel L Oren, Michael J Higley and 3 more

Atlanta Poster Session 6, Fri, Dec 11, 4:30 PM–7:30 PM, Hall C1 · Published 2026

$\mathrm{Bounded\mbox{-}RS}$: A Bounded Evaluation Protocol and Testbed for Representational Systematicity

Boyu Cao, Wenshuo Wang

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

Is a Linear Probe Evidence of a Linear Representation?

Nadir Kutluozen, Christian Internò, David Klindt

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

Boundary Mass in Feature Space as a Label-Budget Diagnostic for Representations

Siming Zhang, Zhehui Shen, Shijie Chen, Yansen Yu and 1 more

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

What Does the Brain See? Multiview Neural Representation to Demystify the Brain-Visual Alignment

Salini Yadav, Taveena Lotey, Pravendra Singh, Partha P Roy

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

Early Detection of Backbone Divergence via Neighborhood Structure in Learned Embeddings

Yiming Zhang, Reza Hassanpour, Georgi Gaydadjiev

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

Tracing Psychometric Inference in Large Language Models

Chih-Hao Hsu, Feng-Chun B Chou, Pin-Hao A Chen

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

From Representation to Intervention: Using Emotion Vectors to Monitor and Guide Language Models

Georgia Dimaki, Martin Villanueva Dimakis

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

Probing Persona-Dependent Preferences in Language Models

Linear probes on LLM residual streams identify a shared preference vector tracking pairwise choices across personas, with cross-persona transfer and causal steering.

Oscar Gilg, Pierre Beckmann, Daniel Paleka, Patrick Butlin

Paris Poster Session 5, Fri, Dec 11, 11:30 AM–1:30 PM, Paris Poster Hall · Published 2026

RAPTOR: Ridge-Adaptive Logistic Probes

RAPTOR is a ridge-regularized logistic probe that uses validation-tuned L2 penalties to extract accurate, stable concept vectors with low training cost for activation steering. It matches or exceeds baseline accuracy and stability across instruction-tuned LLMs, with theoretical analysis via CGMT exp

Ziqi Gao, Yaotian Zhu, Qingcheng Zeng, Xu Zhao and 3 more

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

Geometric Signatures of Reasoning: A Spectral Perspective on Task Hardness

Reasoning trajectories are discrete hidden-state curves whose spectral flatness measures task hardness and whose kinematics predict correctness early.

Aria Masoomi, Mahsa Bazzaz, Adel Javanmard, Vahab Mirrokni

Atlanta Poster Session 6, Fri, Dec 11, 4:30 PM–7:30 PM, Hall C1 · Published 2026

Rep2Text: Decoding Full Text from a Single LLM Token Representation

Rep2Text recovers roughly half of tokens from single LLM representations via adapter-based decoding, showing sequence-length bottlenecks preserve semantics but reduce token recovery.

Haiyan Zhao, Zirui He, Yiming Tang, Fan Yang and 3 more

Atlanta Poster Session 4, Thu, Dec 10, 4:30 PM–7:30 PM, Hall C1 · Published 2026

The Geometry of Forgetting: Temporal Knowledge Drift as an Independent Axis in LLM Representations

Temporal knowledge drift is geometrically orthogonal to correctness and uncertainty in LLM residual streams, making drift undetectable via standard signals despite linear probes reaching 0.83, 0.95 AUROC.

Rania Elbadry, Ahmed Heakl, Fan Zhang, Dani Bouch and 3 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

The Invisible Hand of Physics: When Video Diffusion Models Know More Than They Show

Video diffusion transformers encode physical plausibility linearly in latent states around 81% accuracy despite lacking predictive training, emerging inside the denoiser rather than the VAE.

Parsa Esmati, Somjit Nath, Katja Hofmann, Derek Nowrouzezahrai and 2 more

Atlanta Poster Session 4, Thu, Dec 10, 4:30 PM–7:30 PM, Hall C1 · Published 2026 · ▲ 1 on Hugging Face