Isolater - Feed

Ax Md Tanvirul Alam, Aritran Piplai, Ionut Cardei, Nidhi Rastogi, Peter J Worth Jr 2/13/2026

Minerva: Reinforcement Learning with Verifiable Rewards for Cyber Threat Intelligence LLMs

Minerva applies reinforcement learning with verifiable rewards to train LLMs for cyber threat intelligence standardization tasks.

Ax Abhijit Gupta 2/13/2026

Cardinality-Preserving Attention Channels for Graph Transformers in Molecular Property Prediction

CardinalGraphFormer applies graph transformers to molecular property prediction with attention augmentation for drug discovery applications.

Ax Elizabeth Bates, Chris Hicks, Vasilios Mavroudis 2/13/2026

Beyond Rewards in Reinforcement Learning for Cyber Defence

Study of reinforcement learning for autonomous cyber defense agents, examining reward function design beyond traditional dense reward approaches.

Ax Songtao Wei, Yi Li, Bohan Zhang, Zhichun Guo, Ying Huang, Yuede Ji, Miao Yin, Guanpeng Li, Bingzhe Li 2/13/2026

CoSA: Compressed Sensing-Based Adaptation of Large Language Models

CoSA proposes compressed sensing-based approach for parameter-efficient fine-tuning of LLMs, addressing expressivity limitations of low-rank decomposition methods like LoRA.

Ax Zhiqi Yu, Zhangquan Chen, Mengting Liu, Heye Zhang, Liangqiong Qu 2/13/2026

Unveiling Implicit Advantage Symmetry: Why GRPO Struggles with Exploration and Difficulty Adaptation

Analysis of GRPO limitations in exploration and difficulty adaptation stemming from implicit advantage symmetry in reward estimation.

Ax Yuntong Hu, Matthew Trager, Yuting Zhang, Yi Zhang, Shuo Yang, Wei Xia, Stefano Soatto 2/13/2026

Evolutionary Generation of Multi-Agent Systems

Evolutionary algorithm automatically generates multi-agent system architectures from LLMs without code generation limitations.

Ax Francesco Cagnetta, Allan Ravent\'os, Surya Ganguli, Matthieu Wyart 2/13/2026

Deriving Neural Scaling Laws from the statistics of natural language

Theory predicting neural scaling law exponents from natural language statistics for data-limited LLM scaling.

Ax Zhendong Mi, Yixiao Chen, Pu Zhao, Xiaodong Yu, Hao Wang, Yanzhi Wang, Shaoyi Huang 2/13/2026

Effective MoE-based LLM Compression by Exploiting Heterogeneous Inter-Group Experts Routing Frequency and Information Density

MoE-based LLM compression using SVD exploiting heterogeneous expert routing frequency and information density for practical deployment.

Ax Aaditya Vikram Prasad, Connor Watts, Jack Merullo, Dhruvil Gala, Owen Lewis, Thomas McGrath, Ekdeep Singh Lubana 2/13/2026

Features as Rewards: Scalable Supervision for Open-Ended Tasks via Interpretability

Uses learned LLM features as reward signals for reinforcement learning to reduce hallucinations and improve open-ended task performance.

Ax Iv\'an Arcuschin, David Chanin, Adri\`a Garriga-Alonso, Oana-Maria Camburu 2/13/2026

Biases in the Blind Spot: Detecting What LLMs Fail to Mention

Automated pipeline detects unverbalized biases hidden in LLM chain-of-thought reasoning without predefined categories.

Ax Tony Feng, Trieu H. Trinh, Garrett Bingham, Dawsen Hwang, Yuri Chervonyi, Junehyuk Jung, Joonkyung Lee, Carlo Pagano, Sang-hyun Kim, Federico Pasqualotto, Sergei Gukov, Jonathan N. Lee, Junsu Kim, Kaiying Hou, Golnaz Ghiasi, Yi Tay, YaGuang Li, Chenkai Kuang, Yuan Liu, Hanzhao Lin, Evan Zheran Liu, Nigamaa Nayakanti, Xiaomeng Yang, Heng-Tze Cheng, Demis Hassabis, Koray Kavukcuoglu, Quoc V. Le, Thang Luong 2/13/2026

Towards Autonomous Mathematics Research

Aletheia is a math research agent using LLMs for autonomous theorem discovery, proof generation, and literature navigation in mathematics research.

Ax Kanghyun Noh, Jinheon Choi, Yulwha Kim 2/13/2026

QTALE: Quantization-Robust Token-Adaptive Layer Execution for LLMs

QTALE combines token-adaptive layer execution with quantization for efficient LLM deployment while handling robustness challenges.

Ax Seonglae Cho, Zekun Wu, Adriano Koshiyama 2/13/2026

Control Reinforcement Learning: Interpretable Token-Level Steering of LLMs via Sparse Autoencoder Features

Control Reinforcement Learning uses sparse autoencoders to identify and steer interpretable features at token-level for LLM steering and intervention.

Ax Jonathan Williams, Esin Tureci 2/13/2026

Prioritize the Process, Not Just the Outcome: Rewarding Latent Thought Trajectories Improves Reasoning in Looped Language Models

LoopLMs improved via reinforcement learning that rewards intermediate reasoning steps rather than final outputs for better multi-step reasoning.

Ax Yifan Zhang, Zunhai Su, Shuhao Hu, Rui Yang, Wei Wu, Yulei Qian, Yuchen Xie, Xunliang Cai 2/13/2026

SnapMLA: Efficient Long-Context MLA Decoding via Hardware-Aware FP8 Quantized Pipelining

SnapMLA optimizes DeepSeek MLA decoding using FP8 quantization and hardware-aware pipelining for efficient long-context inference.

Ax Jie Jiang, Ke Cheng, Xin Xu, Mengyang Pang, Tianhao Lu, Jiaheng Li, Yue Liu, Yuan Wang, Jun Zhang, Huan Yu, Zhouchen Lin 2/13/2026

PRISM: Parallel Residual Iterative Sequence Model

PRISM architecture balances transformer expressivity with linear model efficiency for sequence generation through parallel residual iterations.

Ax Matthew L. Daggitt, Wen Kokke, Robert Atkey 2/13/2026

Compiling High-Level Neural Network Specifications into VNN-LIB Queries

Framework for compiling high-level neural network specifications into VNN-LIB format for formal verification of neural networks.

Ax Florian Valade 2/13/2026

Accelerating Large Language Model Inference with Self-Supervised Early Exits

Technique to accelerate LLM inference by adding self-supervised early exit heads at intermediate transformer layers using confidence thresholds.

Ax Marc Goerigk, Michael Hartisch, Sebastian Merten, Kartikey Sharma 2/13/2026

Feature-Based Interpretable Surrogates for Optimization

Framework using decision trees and optimization rules for interpretable surrogate models in optimization.

Ax Yujun Zhou, Jingdong Yang, Yue Huang, Kehan Guo, Zoe Emory, Bikram Ghosh, Amita Bedar, Sujay Shekar, Zhenwen Liang, Pin-Yu Chen, Tian Gao, Werner Geyer, Nuno Moniz, Nitesh V Chawla, Xiangliang Zhang 2/13/2026

LabSafety Bench: Benchmarking LLMs on Safety Issues in Scientific Labs

Benchmark evaluating safety of LLMs and vision language models on laboratory safety issues and procedural guidance tasks.

Ax Alexander Spangher, Michael Lu, Sriya Jeslyn Kalyan, Hyundong Justin Cho, Weiyan Shi, Jonathan May 2/13/2026

NewsInterview: a Dataset and a Playground to Evaluate LLMs' Ground Gap via Informational Interviews

Dataset of 40,000 journalistic interviews to evaluate LLM grounding and strategic dialogue capabilities in informational interviews.

Ax Xinyi Yang, Liang Zeng, Heng Dong, Chao Yu, Xiaoran Wu, Huazhong Yang, Yu Wang, Milind Tambe, Tonghan Wang 2/13/2026

Translate Policy to Language: Flow Matching Generated Rewards for LLM Explanations

Framework training LLMs to generate policy explanations via reinforcement learning using generative normalizing flows for reward generation.

Ax Shiping Yang, Jie Wu, Wenbiao Ding, Ning Wu, Shining Liang, Ming Gong, Hongzhi Li, Hengyuan Zhang, Angel X. Chang, Dongmei Zhang 2/13/2026

Quantifying and Improving the Robustness of Retrieval-Augmented Language Models Against Spurious Features in Grounding Data

Research on robustness of retrieval-augmented generation systems, identifying and mitigating spurious features in grounding data that affect RAG performance.

Ax Feijiang Han, Jiaming Zhang, Chuyi Deng, Jianheng Tang, Yunhuai Liu 2/13/2026

Can LLMs Handle WebShell Detection? Overcoming Detection Challenges with Behavioral Function-Aware Framework

Framework for using LLMs to detect WebShell attacks with behavioral function-aware analysis, addressing data scarcity and concept drift.

Ax Cong Qi, Hanzhang Fang, Siqi Jiang, Xun Song, Tianxing Hu, Wei Zhi 2/13/2026

Bidirectional Mamba for Single-Cell Data: Efficient Context Learning with Biological Fidelity

GeneMamba: bidirectional Mamba architecture for single-cell RNA sequencing analysis with efficient context learning.

Ax Moein Khajehnejad, Forough Habibollahi, Devon Stoliker, Adeel Razi 2/13/2026

BrainSymphony: A parameter-efficient multimodal foundation model for brain dynamics with limited data

BrainSymphony: parameter-efficient multimodal foundation model for brain dynamics combining fMRI and structural connectivity data.

Ax Itai Allouche, Itay Asael, Rotem Rousso, Vered Dassa, Ann Bradlow, Seung-Eun Kim, Matthew Goldrick, Joseph Keshet 2/13/2026

How Does a Deep Neural Network Look at Lexical Stress in English Words?

Interpretability study examining how CNNs predict lexical stress in English words from speech data using neural network analysis.

Ax Ana-Cristina Rogoz, Radu Tudor Ionescu, Alexandra-Valentina Anghel, Ionut-Lucian Antone-Iordache, Simona Coniac, Andreea Iuliana Ionescu 2/13/2026

A Large-Scale Benchmark for Evaluating Large Language Models on Medical Question Answering in Romanian

MedQARo benchmark: 105,880 medical QA pairs in Romanian for evaluating LLM performance on medical question answering requiring reasoning.

Ax Mohsen Fayyaz, Ali Modarressi, Hanieh Deilamsalehy, Franck Dernoncourt, Ryan Rossi, Trung Bui, Hinrich Sch\"utze, Nanyun Peng 2/13/2026

Steering MoE LLMs via Expert (De)Activation

SteerMoE framework for controlling Mixture-of-Experts LLMs by detecting and activating/deactivating behavior-associated expert networks.

Ax Anvit Garg, Sohom Bhattacharya, Pragya Sur 2/13/2026

Preventing Model Collapse Under Overparametrization: Optimal Mixing Ratios for Interpolation Learning and Ridge Regression

Analysis of model collapse in generative models trained on their own synthetic outputs, deriving generalization error formulae for overparameterized linear regression.

Ax Zhen Bi, Zhenlin Hu, Xueshu Chen, Mingyang Chen, Cheng Deng, Yida Xue, Zhen Wang, Qing Shen, Ningyu Zhang, Jungang Lou 2/13/2026