Prefill-Free Cross-Family KV Cache Transfer for Heterogeneous Multi-Agent LLMs

HeteroFold enables prefill-free cross-family KV cache transfer between frozen heterogeneous LLM agents, accelerating 32K context transfer up to 10.7x while matching text-based multi-agent performance.

Vincent-Daniel Yun, Woosang Lim, Haneul Yoo, Sungjoo Yoo and 2 more

Published Sep 26, 2026 · 0 citations · ▲ 95 on Hugging Face · Code ★ 1

Attention Is All You Need

The Transformer replaces recurrence and convolutions with attention, achieving superior translation quality and faster training.

Ashish Vaswani, Noam Shazeer, Niki Jitendra Parmar, Jakob Uszkoreit and 4 more

Published Aug 23, 2025 · 26,828 citations

Reinforcement Learning from Rich Feedback with Distributional DAgger

Rishabh Agrawal, Jacob Fein-Ashley, Paria Rashidinejad

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

A Foundational Model System for Datacenter Machine Repairs

Yuanlin Wen, Elan S Markowitz, Zubo Gu, Sami Abu-El-Haija and 10 more

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

Chain-of-Thought Is Not Explainability

Fazl Barez, Tung-Yu Wu, Iván Arcuschin Moreno, Michael Lan and 12 more

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

TritonTune: LLM-Guided Multi-Agent Optimization of GPU Kernel Configurations

Shukai Duan, Mihai Capotă, Guixiang Ma, Hou-Jen Ko and 4 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

C-GRPO: Conformal Group Relative Policy Optimization

Arya Fayyazi, Seyedarmin Azizi, Mehdi Kamal, Massoud Pedram

Atlanta Poster Session 1, Wed, Dec 9, 10:00 AM–1:00 PM, Hall C1 · Published 2026

Even Sailors Need Calm Seas: Taming the Geometry of VLMs for Fast Adversarial Fine-Tuning

Yuqing Wen, Junhao Dong, Ting Peng, Xudong Zhang and 5 more

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

The Sharp Directions Are Against You: Curvature Analysis of Activation Steering

Seyedarmin Azizi, Arya Fayyazi, Parsa Razmara, Erfan Baghaei Potraghloo and 2 more

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

Say the Same, Act Differently: Text-Orthogonal Action Subspaces in Reasoning Vision-Language-Action Models

Zihao Feng, Qingzhao Zhang, Chunyu Xia, Bo Yu and 2 more

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

Context-Aware Autoregressive Image Generation for Emerging Reasoning Properties

Jixuan Ying, Haoyu Liu, Timing Yang, Tingyu Zhu and 6 more

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

Modality-Depth Routing for Visual Reasoning in VLM Post-Training

Yiming Ren, Yiran Xu, Chufan Shi, Yu Qiao and 2 more

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

The Heel of RLVR: Benchmark Glory Should Not Outpace Honest Measurement

Shuo Yang, Chiyu Ma, Kexin Huang, Jinda Lu and 10 more

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

Generalized Laplacian in Spectral Seriation on Manifold Data

Ruizi Wu, Wanjie Wang, Jinchi Lv

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

Accelerated last-iterate convergence of Extragradient via power-law stepsizes

Yue Wu, Weiqiang Zheng, Yang Cai, Haipeng Luo

Atlanta Poster Session 1, Wed, Dec 9, 10:00 AM–1:00 PM, Hall C1 · Published 2026

DDBench: A Benchmark for Agentic Debugging on Distributed Systems

Yibo Yan, Huijuan Wang, Junzhou He, Yizhuo Liang and 3 more

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026

Preferential dynamic modeling with forward-backward smoothing

Omid G. Sani, Trisha Jha, Mohammad Hosseini, Maryam Shanechi

Atlanta Poster Session 1, Wed, Dec 9, 10:00 AM–1:00 PM, Hall C1 · Published 2026

Precise Debugging Benchmark: Is Your Model Debugging or Regenerating?

Miaosen Chai, Wang Bill Zhu, Shangshang Wang, Yejia Liu and 4 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

PhysGraphNet: Physical-State Scene Graphs via Latent Graph Reasoning and Counterfactual Supervision

Zhengtao Yao, Runhao Li, Yan Wen, Guang Yang and 6 more

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

Not All Slots Are Equal: Non-Co-Progressive Markov Bridge for Bundle Construction

Rongchao Zhang, Haodong Jing, Siheng Wang, Zhengtao Yao and 1 more

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

Align as You Couple: Learning Spatial Resolved Inference from H&E Images with Mollified Flow Matching

Rongchao Zhang, Guangyuan Dong, Siheng Wang, Zhengtao Yao and 2 more

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

Overcoming Attention Distraction: Training-Free Latent Communication for Multi-Agent Systems

zhengjie zhou, Yahao Liu, Ziyue Feng, Tengfei LIU and 1 more

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

Statistical Inference in Causal Partial Identification under Smooth Densities

Sirui Lin, Zijun Gao, Jose Blanchet, Peter W Glynn

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

Adaptive Generate-Rank-Verify: Inference-Time Search with Costly Verification

ADAP adaptively increases response sampling and verification to find verified positives with near-optimal expected cost under monotonic rewards.

Shaddin Dughmi, Mahdi Haghifam, Yusuf H Kalayci

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 1 on Hugging Face · Code ★ 2

Adaptive Calibration in Non-Stationary Environments

Online prediction algorithms achieve calibration error adapting to non-stationarity via $\tilde O(\min\{\sqrt{T}+(TC)^{1/3},\sqrt{KT}\})$ bounds, smoothly interpolating between i.i.d. and adversarial settings.

Junyan Liu, Haipeng Luo, Lillian Ratliff

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026

EVIDENT: Routing MLLM Adaptation through Entity-Grounded Visual Evidence for Cross-Domain Video Temporal Grounding

EVIDENT routes video temporal grounding through explicit visual entity evidence to improve cross-domain robustness via parameter-efficient adapters and distillation.

Geo Ahn, Jiwook Han, Youngrae Kim, Joonseok Lee and 1 more

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

3D-PLOT-LLM: Part-Level Object Tokens for 3D Large Language Models

3D-PLOT-LLM inserts learnable part tokens into frozen point features to enable part-level reasoning in 3D LLMs with under 1M parameters, outperforming prior part-aware models on part-QA and grounded description benchmarks.

Jintang Xue, Xinyu Wang, Yixing Wu, Jingwen Chen and 1 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

Graph-Regularized Sparse Autoencoders for LLM Safety Steering

Graph-Regularized Sparse Autoencoders smooth SAE decoder vectors over a neuron co-activation graph to learn safety-steering directions, improving selective refusal by over 16 points across jailbreak benchmarks while preserving benign performance and generalizing across models.

Jehyeok Yeon, Federico Cinus, Yifan Wu, Luca Luceri

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

PG-LRF: Physiology-Guided Latent Rectified Flow for Electro-Hemodynamic PPG-to-ECG Generation

PG-LRF uses a physiology-guided latent rectified flow with an electro-hemodynamic simulator to generate physiologically plausible ECGs from PPG, improving generation and cardiovascular disease classification.

Xiaoda Wang, Minxiao Wang, Kaiqiao Han, Defu Cao and 9 more

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

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

Learning Fractional-Order Dynamics from a Single Trajectory

FO-GS estimates discrete-time fractional-order linear dynamics from one trajectory with both errors scaling as O(t^{-1/2}).

Xiaole Zhang, Ziyi Zhang, zehao zhao, Stephen Tu and 3 more

Atlanta Poster Session 1, Wed, Dec 9, 10:00 AM–1:00 PM, Hall C1 · Published 2026

Video Models Can Reason with Verifiable Rewards

VideoRLVR applies reinforcement learning with verifiable rewards to video diffusion models, improving rule-consistent visual reasoning and cutting training latency 40% via early-step optimization.

Tinghui Zhu, Sheng Zhang, James Yipeng Huang, Selena Song and 4 more

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026 · ▲ 9 on Hugging Face · Code ★ 28

Ghosted Layers: Unconstrained Activation Alignment for Recovering Layer-Pruned LLMs

Ghosted Layers recovers layer-pruned LLMs via training-free activation alignment with a closed-form linear operator, outperforming constrained baselines.

Vincent-Daniel (Juyoung) Yun, Junhyuk Jo, Sai Praneeth Karimireddy, Sunwoo Lee

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026 · ▲ 1 on Hugging Face

OASES: Outcome-Aligned Search-Evaluation Co-Training for Agentic Search

OASES co-trains a search policy and adaptive evaluator to provide outcome-aligned process rewards, outperforming RL baselines on multi-hop QA benchmarks.

Erhan Zhang, Yiqun Chen, Zechun Niu, Wei Yang and 5 more

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

DiPO: Disentangled Perplexity Policy Optimization for Fine-grained Exploration-Exploitation Trade-Off

DiPO disentangles perplexity into exploration and exploitation subspaces to enable fine-grained trade-offs, improving LLM reasoning and function calling via stable perplexity-guided policy optimization.

Xiaofan Li, Ming Yang, Zhiyuan Ma, Shichao Ma and 8 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 13 on Hugging Face

Don't Always Pick the Highest-Performing Model: An Information Theoretic View of LLM Ensemble Selection

Formulating ensemble selection as mutual-information maximization reveals an information-theoretic error floor from model correlation and yields a greedy algorithm that outperforms baselines under fixed query budgets.

Yigit Turkmen, Baturalp Buyukates, Melih Bastopcu

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026

WorldMemArena: Evaluating Multimodal Agent Memory Through Action–World Interaction

WorldMemArena evaluates multimodal agent memory through an action-world loop, showing writing and storage improvements do not guarantee performance and harness-based memory remains costly and unreliable.

Chengzhi Liu, Yuzhe YANG, Sophia Xiao Pu, Yepeng Liu and 15 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 13 on Hugging Face · Code ★ 29

RATS! Patches Talk Through Registers: Emergent Parts in Register Attention Transformers

RATS decomposes vision transformers' classification token into learnable register tokens that spontaneously specialize into object parts, improving segmentation by up to 12 mIoU.

Timing Yang, Predrag Neskovic, Jansen Seheult, Wenchao Han and 3 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

Prediction Under Imperfect Compression: A Theory of Approximate MDL

Approximate balanced MDL with additive slack yields finite cumulative squared error for regularization weight lambda at least 1, but multiplicative approximation and lambda under 1 cause failure, establishing additive approximation is essential.

Qian Li, Xinyu Mao, Shang-Hua Teng, Guangxu Yang

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

Computer Use at the Edge of the Statistical Precipice

A 1MB replay script outperforms frontier agents on static benchmarks because of flawed environment design and evaluation; the paper proposes PRISM principles, DigiWorld, and hierarchical bootstrap aggregation to fix both.

Pierluca D Oro, Sneha Silwal, William R Wong, Yuxuan Sun and 5 more

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

Not all uncertainty is alike: volatility, stochasticity, and exploration

Volatility and stochasticity both increase uncertainty but drive optimal exploration in opposite directions; CAUSE captures this asymmetry and improves restless-bandit performance.

Payam Piray

Atlanta Poster Session 3, Thu, Dec 10, 10:00 AM–1:00 PM, Hall C1 · Published 2026

Causal Survival Forests with Negative Controls

NC-CSF integrates negative controls with causal survival forests to estimate heterogeneous treatment effects in survival studies with unmeasured confounding, reducing bias and error versus baselines.

Zijun Gao, Kyoungeui Hong, Leyi Ma, Qianli Wu and 2 more

Atlanta Poster Session 1, Wed, Dec 9, 10:00 AM–1:00 PM, Hall C1 · Published 2026

TraXion: Rethinking Pre-training Frameworks for Mobility and Beyond

TraXion introduces MESES axioms and a pre-training framework for multi-entity spatiotemporal event streams that beats mobility baselines and generalizes to security and health logs.

Shang-Ling Hsu, Mark Tenzer, Cyrus Shahabi, Khurram Shafique

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

Sparse Layers are Critical to Scaling Looped Language Models

Looped-MoE models scale better than standard transformers via routing divergence that recovers expressivity, and loop boundaries enable efficient early exits with minimal quality loss.

Ryan Lee, Jacob Biloki, Edward J Hu, Jonathan May

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

Group Distributionally Robust Optimization with Flexible Sample Queries

A GDRO algorithm via flexible sample queries and a prediction-with-limited-advice game achieves high-probability error O(√(∑ m/r_j)/t) with consistent sample complexity O(m log m/ε²).

Haomin Bai, Dingzhi Yu, Shuai Li, Haipeng Luo and 1 more

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

Categorical Bayes filtering for computational phenotyping in adaptive learning

Categorical Bayes Filter deterministically disentangles environmental volatility from observation stochasticity via differentiable quantile grids, recovering cross-over phenotyping patterns and trial-level ambiguity signals that particle filters miss.

Junxi Chen, Payam Piray

Atlanta Poster Session 3, Thu, Dec 10, 10:00 AM–1:00 PM, Hall C1 · Published 2026

Evolutionary Feature Engineering for Structured Data

Evolutionary Feature Engineering uses LLM-based evolution to discover structured-data preprocessing transformations, reducing forecasting errors up to 19% and improving tabular predictions.

Ege Onur Taga, Yilin Zhuang, Muhammed Emrullah Ildiz, Petros Mol and 3 more

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

When Dynamics Shift, Robust Task Inference Wins: Offline Imitation Learning with Behavior Foundation Models Revisited

Robust minimax task inference for behavior foundation models enables offline imitation learning robust to dynamics shifts using only single-environment data, outperforming standard and robust baselines.

Rishabh Agrawal, Rahul Jain, Ashutosh Nayyar

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026 · ▲ 1 on Hugging Face

A Theory of Time-Sensitive Language Generation: Sparse Hallucination Beats Mode Collapse

Eventually consistent generators cannot generate high-ranked strings before deadlines, but vanishing hallucination rates enable timely superlinear-deadline coverage, which is impossible under linear deadlines.

Atul Ganju, Travis McVoy, Shaddin Dughmi, Shang-Hua Teng

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

Lie Generator Networks for Nonlinear Partial Differential Equations

Lie Generator Network-Koopman lifts nonlinear PDE dynamics into linear latent spaces via stable, interpretable Koopman generators that recover turbulence spectra and enable cross-regime transfer without physics supervision.

Shafayeth Jamil, Rehan Kapadia

Atlanta Poster Session 3, Thu, Dec 10, 10:00 AM–1:00 PM, Hall C1 · Published 2026

Default Feature Representations of the Cognitive Map

Default Feature Representations parameterize predictive cognitive maps via a fixed feature basis and an adaptive environment operator, enabling rapid sample-based replanning and local grid-cell remapping.

Armin Bazarjani, Payam Piray

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

Smart Picks in the Dark: Towards Efficient RLVR for Reasoning via Tracing Metacognitive Pivots

PivotTrace traces metacognitive pivots via attention dynamics to select unlabeled reasoning samples, surpassing fully supervised models with 29.3% annotations and 2.75x faster convergence.

Guangcheng Zhu, Shenzhi Yang, Haobo Wang, Xing Zheng and 6 more

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

Robust Policy Optimization to Prevent Catastrophic Forgetting

FRPO robustifies RLHF via a max-min objective that maximizes reward across downstream-adaptation neighborhoods, substantially reducing catastrophic forgetting while preserving task performance.

Mahdi Sabbaghi, George J. Pappas, Adel Javanmard, Hamed Hassani

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026 · Code ★ 6

Discovering What You Can Control: Interventional Boundary Discovery for Reinforcement Learning

IBD treats an RL agent's actions as randomized interventions and uses per-dimension two-sample tests with FDR correction to identify controllable observation dimensions, matching oracle returns across 12 continuous-control tasks with up to 100 distractors.

Jiaxin Liu, Anzhe Cheng, Paul Bogdan

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

Memory Retrieval for Changing Preferences

A Bayes-factor utility framework selects memory turns by evidence of latent preference changes and regulates access accordingly. It outperforms embedding retrieval on preference-intensive long-context dialogue tasks.

Yuehan Qin, Li Li, Linxin Song, Jiate Li and 3 more

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026

A Unified Approach for Computing Wasserstein Barycenters of Discrete and Continuous Measures

A primal mirror descent algorithm computes exact Wasserstein barycenters for discrete and continuous measures in Fisher-Rao geometry with convergence guarantees.

Peng Xu, Changbo Zhu, Xiaohui Chen

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

Survive or Collapse: The Asymmetric Roles of Data Gating and Reward Grounding in Self-Play RL

Self-play RL stability depends mainly on a strict data gate over proposer tasks, not reward design; ground-truth access accelerates collapse via a self-consistent attractor.

Sophia Xiao Pu, Zhaotian Weng, Chengzhi Liu, Jayanth Srinivasa and 3 more

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

Geometry over Density: Few-Shot Cross-Domain OOD Detection

UFCOD uses diffusion score geometry to enable cross-domain OOD detection with ~100 unlabeled ID samples and no retraining, achieving 93.7% AUROC across 12 benchmarks with ~500x sample efficiency gains.

Li Li, You Qin, Jiate Li, Charith Peris and 3 more

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

TrajWiki: Source-Grounded Memory Trajectories for Long-Horizon Dialogue Agents

TrajWiki represents memory as source-grounded evolution trajectories with claim-level updates and a wiki layer to improve long-horizon dialogue performance and interpretability.

Jingyu Sun, Yuyang Xue, Mingyang Li, Zhengtao Yao and 8 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

Attention Alignment Between Humans and Vision-Language Models

Vision-language model decoder architecture dominates human attention alignment, with LSTM decoders reaching 85, 87% of the human noise ceiling but remaining diffuse, while transformer decoders show sharper task differentiation despite lower alignment; encoder effects are secondary and neural predict

Isaac Christian, Udith Haputhanthri, Declan Campbell, Samuel Nastase and 2 more

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