Learning from Runtime Feedback through Failure-Bank Self-Evolution for Vision-Language-Action Models

FailBank turns runtime shield feedback into persistent VLA policy updates via failure-bank self-evolution, raising success rates up to 25.4 points and cutting policy-induced cost up to 35.6%.

Mingyue Cui, Zheyuan Liu, Yihan Zhu, Zheyuan Zhang and 1 more

Published Sep 30, 2026 · 0 citations · ▲ 15 on Hugging Face · Code ★ 2

Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data

Reinforcement learning on saturated reasoning data causes mode collapse as advantage signals vanish; CUTS sampling and Mixed-CUTS restore diversity, boosting AIME25 accuracy by up to 15.1%.

Zhenwen Liang, Yujun Zhou, Sidi Lu, Xiangliang Zhang and 2 more

Published Apr 20, 2026 · 0 citations

MoCo: A One-Stop Shop for Model Collaboration Research

MoCo unifies 26 model collaboration methods and 25 benchmarks to show collaboration outperforms single models in 61% of settings by up to 25.8%.

Shangbin Feng, Yuyang Bai, Ziyuan Yang, Yike Wang and 16 more

Published Jan 29, 2026 · 0 citations · Code ★ 63

From Personal to Collective: On the Role of Local and Global Knowledge in LLM Personalization

LoGo augments individual user signals with evolving global and community-level behavioral patterns via adaptive mediation, improving LLM personalization and reducing overfitting.

Zehong Wang, Junlin Wu, Zhaoxuan Tan, Bolian Li and 3 more

Published 2026 · 0 citations

Instant Personalized Large Language Model Adaptation via Hypernetwork

A hypernetwork enables instant personalized large language model adaptation by generating user-specific parameters directly from user data.

Zhaoxuan Tan, Zixuan Zhang, Haoyang Wen, Zheng Li and 7 more

Published 2026 · 1 citation

Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data

Standard RL collapses on saturated reasoning data due to vanishing advantage signals, so CUTS sampling and Mixed-CUTS training restore exploration and boost AIME25 Pass@1 by 15.1%.

Zhenwen Liang, Yujun Zhou, Sidi Lu, Xiangliang Zhang and 2 more

Published 2026 · 0 citations

IHEval: Evaluating Language Models on Following the Instruction Hierarchy

IHEval evaluates language models on instruction hierarchy compliance, finding they frequently disregard system-level priority instructions.

Zhihan Zhang, Shiyang Li, Zixuan Zhang, Xin Liu and 10 more

Published 2025 · 4 citations

Modality-Aware Neuron Pruning for Unlearning in Multimodal Large Language Models

Modality-aware neuron pruning removes target knowledge from multimodal LLMs by selectively pruning modality-specific neurons to enable precise unlearning.

Zheyuan Liu, Guangyao Dou, Xiangchi Yuan, Chunhui Zhang and 2 more

Published 2025 · 4 citations

What Does the Bot Say? Opportunities and Risks of Large Language Models in Social Media Bot Detection

Instruction-tuned LLMs with a mixture-of-experts framework outperform bot detectors by 9.1%, but LLM-guided manipulation reduces their performance by up to 29.6%.

Shangbin Feng, Herun Wan, Ningnan Wang, Zhaoxuan Tan and 2 more

Published 2024 · 27 citations

LMBot: Distilling Graph Knowledge into Language Model for Graph-less Deployment in Twitter Bot Detection

LMBot distills graph neural network knowledge into language models for efficient graph-less Twitter bot detection, achieving state-of-the-art results across four benchmarks.

Cai, Zijian, Zhaoxuan Tan, Zhenyu Lei, Zhu, Zifeng and 3 more

Published Jun 30, 2023 · 0 citations

BotPercent: Estimating Bot Populations in Twitter Communities

BotPercent estimates community-specific Twitter bot populations by calibrating detection models across social contexts, revealing heterogeneous spatial-temporal bot distributions and achieving state-of-the-art community-level detection accuracy.

Zhaoxuan Tan, Shangbin Feng, Melanie Sclar, Herun Wan and 3 more

Published 2023 · 16 citations

From Facts to Personas: Interpretable Role Unlearning in LLMs via Mixture-of-Experts

Ruihong Zeng, Puning Yang, Jinghui Zhang, Shen Gao and 3 more

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

ScrapeBench: Evaluating Legal Compliance of AI Agents in Website Scraping

Joseph Marvin Imperial, Daniel Slate, Noam Kolt

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

Scheming Is a Symptom: Alignment Research Should Probe Reflexive Fragility

Nan Zhang, Heng Xu

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

Can Agents Price a Reaction? Evaluating LLMs on Chemical Cost Reasoning

Yuyang Wu, Yue Huang, Shuaike Shen, Xujian Wang and 7 more

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

Weight Space Learning needs to unify benchmarking! A taxonomy of evaluation practices

Tobias Ettling, Damian Falk, Aron Asefaw, Léo Meynent and 6 more

Paris Poster Session 3, Thu, Dec 10, 12:30 PM–2:30 PM, Paris Poster Hall · Published 2026

Factor Augmented High-Dimensional SGD

FSGD is a streaming SGD method that uses latent factor representations for high-dimensional tasks, achieving scalable optimization with theoretical convergence guarantees including factor estimation error.

Shubo Li, Yuefeng Han, Xiufan Yu

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

Bridging Textual Profiles and Latent User Embeddings for Personalization

BLUE aligns interpretable LLM-generated user profiles with embedding-based recommendation objectives via reinforcement learning, outperforming baselines in sequential recommendation and cross-domain transfer.

Zhaoxuan Tan, Xiang Zhai, Yan Zhu, Meng Jiang and 1 more

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

Controllable Molecular Generative Foundation Models

CoMole unifies molecular graph generation via motif-aware diffusion and reinforcement learning, achieving top controllability across nine targets with up to 48.2% lower MAE and over 0.94 validity.

Yihan Zhu, Yuhan Liu, Weijiang Li, Tengfei Luo and 1 more

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

Covariate-Adjusted Deep Causal Learning for Heterogeneous Panel Data Models

CoDEAL integrates neural covariate adjustments with autoencoder factor structures for heterogeneous causal panel data and proves convergence of counterfactual estimates.

Guanhao Zhou, Yuefeng Han, Xiufan Yu

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