3/26/2026
Smooth Gate Functions for Soft Advantage Policy Optimization
Smooth gate functions for stabilizing GRPO LLM training. Replaces hard clipping with sigmoid-based gating to improve optimization stability in reasoning tasks.
3/26/2026
Beyond State-Wise Mirror Descent: Offline Policy Optimization with Parametric Policies
Theoretical analysis of offline reinforcement learning with general function approximation and parametric policies, extending beyond finite action spaces.
3/26/2026
OSS-CRS: Liberating AIxCC Cyber Reasoning Systems for Real-World Open-Source Security
Open-source framework for deploying DARPA AIxCC cyber reasoning systems locally. Makes competition CRSs usable outside original infrastructure with improved accessibility.
3/26/2026
MM-tau-p$^2$: Persona-Adaptive Prompting for Robust Multi-Modal Agent Evaluation in Dual-Control Settings
Evaluation framework for persona-adaptive LLM-powered agents in multi-modal settings, addressing user-aware behavior in customer experience management.
3/26/2026
Exploring Collatz Dynamics with Human-LLM Collaboration
Mathematical analysis of Collatz conjecture dynamics using modular arithmetic and combinatorial methods. Pure mathematics research unrelated to AI/ML.
3/26/2026
Red-Teaming Vision-Language-Action Models via Quality Diversity Prompt Generation for Robust Robot Policies
Mathematical analysis of Collatz conjecture dynamics using modular arithmetic and combinatorial methods with LLM assistance.
3/26/2026
AgentDrift: Unsafe Recommendation Drift Under Tool Corruption Hidden by Ranking Metrics in LLM Agents
Red-teaming Vision-Language-Action models through quality diversity prompt generation to improve robot policy robustness.
3/26/2026
Geometry-Guided Camera Motion Understanding in VideoLLMs
AgentDrift: reveals safety risks in LLM agent recommendations when tools are corrupted, hidden by standard metrics.
3/26/2026
Pixel-level Scene Understanding in One Token: Visual States Need What-is-Where Composition
Framework for improving VideoLLM understanding of camera motion through benchmarking, diagnosis, and explicit geometry injection.
3/26/2026
FedPBS: Proximal-Balanced Scaling Federated Learning Model for Robust Personalized Training for Non-IID Data
Visual state representations for robotic agents using what-is-where composition for dynamic scene understanding.
3/26/2026
Sample-Efficient Hypergradient Estimation for Decentralized Bi-Level Reinforcement Learning
FedPBS: federated learning algorithm for personalized training on non-IID data with improved robustness.
3/26/2026
100x Cost & Latency Reduction: Performance Analysis of AI Query Approximation using Lightweight Proxy Models
Sample-efficient hypergradient estimation for decentralized bi-level reinforcement learning in strategic decision-making.
3/26/2026
Mitigating LLM Hallucinations through Domain-Grounded Tiered Retrieval
Proxy models reduce cost and latency of AI queries in SQL databases by 100x through approximation techniques.
3/26/2026
Evolutionarily Stable Stackelberg Equilibrium
Domain-grounded tiered retrieval architecture to reduce LLM hallucinations through retrieval-based verification.
3/26/2026
Ontology-Guided Diffusion for Zero-Shot Visual Sim2Real Transfer
Evolutionarily Stable Stackelberg Equilibrium: game theory solution concept for asymmetric leader-follower games.
3/26/2026
Agent Control Protocol: Admission Control for Agent Actions
Ontology-Guided Diffusion for zero-shot sim2real transfer using neuro-symbolic approach to bridge simulation-reality gap.
3/26/2026
LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels
Agent Control Protocol: formal specification for admission control governance of autonomous agents with cryptographic identity and policy compliance.
3/26/2026
An Agentic Multi-Agent Architecture for Cybersecurity Risk Management
Multi-agent AI system with six specialized agents for automated NIST CSF-aligned cybersecurity risk assessments for small organizations.
3/26/2026
Alignment Whack-a-Mole : Finetuning Activates Verbatim Recall of Copyrighted Books in Large Language Models
Study showing finetuning bypasses LLM safety mechanisms and triggers verbatim recall of copyrighted training data.
3/26/2026
DeepXplain: XAI-Guided Autonomous Defense Against Multi-Stage APT Campaigns
Explainable DRL framework for autonomous APT defense using provenance-based graphs and stage-aware modeling.
3/26/2026
LLM-Powered Workflow Optimization for Multidisciplinary Software Development: An Automotive Industry Case Study
LLM-based workflow system for multidisciplinary software development coordinating domain experts and developers in automotive.
3/26/2026
PRISM: Breaking the O(n) Memory Wall in Long-Context LLM Inference via O(1) Photonic Block Selection
PRISM photonic accelerator approach reducing KV cache memory bandwidth from O(n) to O(1) for long-context LLM inference.
3/26/2026
mSFT: Addressing Dataset Mixtures Overfitting Heterogeneously in Multi-task SFT
mSFT algorithm for optimizing heterogeneous multi-task SFT data mixtures by dynamically adjusting compute per sub-dataset.
3/26/2026
Extending Precipitation Nowcasting Horizons via Spectral Fusion of Radar Observations and Foundation Model Priors
Weather prediction combining radar observations with foundation model priors for extended nowcasting horizons.
3/26/2026
Sim-to-Real of Humanoid Locomotion Policies via Joint Torque Space Perturbation Injection
Sim-to-real transfer for humanoid robot control using state-dependent joint torque perturbations instead of domain randomization.
3/26/2026
Tiny Inference-Time Scaling with Latent Verifiers
Inference-time scaling with lightweight latent verifiers instead of MLLMs to reduce computational cost in verification.
3/26/2026
Language Models Can Explain Visual Features via Steering
Method using causal interventions and Vision-Language Models to explain sparse autoencoder features in vision models.
3/26/2026
Beyond Accuracy: Introducing a Symbolic-Mechanistic Approach to Interpretable Evaluation
Interpretable evaluation combining symbolic rules with mechanistic interpretability to detect memorization vs genuine generalization.
3/26/2026
Implicit Turn-Wise Policy Optimization for Proactive User-LLM Interaction
ITPO framework for optimizing multi-turn human-LLM interactions via RL despite sparse rewards and user stochasticity.
3/26/2026
Upper Entropy for 2-Monotone Lower Probabilities
Theoretical analysis of upper entropy computation for credal sets and uncertainty quantification. Pure mathematics focus.
3/26/2026
Synthetic Mixed Training: Scaling Parametric Knowledge Acquisition Beyond RAG
Training method combining synthetic QA and document generation to improve LLM knowledge beyond RAG performance ceiling.
3/26/2026
Safe Reinforcement Learning with Preference-based Constraint Inference
Safe reinforcement learning framework inferring constraints from user preferences with minimal expert demonstrations.
3/26/2026
AscendOptimizer: Episodic Agent for Ascend NPU Operator Optimization
RL agent optimizing operator kernels on Huawei Ascend NPUs. Addresses knowledge gap in alternative hardware ecosystem.
3/26/2026
Causal Reconstruction of Sentiment Signals from Sparse News Data
Causal signal reconstruction approach for converting sparse news sentiment into reliable time series for financial/tech analysis.
3/26/2026
StateLinFormer: Stateful Training Enhancing Long-term Memory in Navigation
StateLinFormer model using linear attention for navigation agents with long-term memory. Addresses context window limitations in Transformers.
3/26/2026
Dual-Criterion Curriculum Learning: Application to Temporal Data
Research on curriculum learning with dual criteria for temporal data. Proposes improved difficulty-based training scheduling.
3/26/2026
PoiCGAN: A Targeted Poisoning Based on Feature-Label Joint Perturbation in Federated Learning
PoiCGAN: Poisoning attack method against federated learning systems using feature-label joint perturbation.
3/26/2026
APreQEL: Adaptive Mixed Precision Quantization For Edge LLMs
APreQEL: Adaptive mixed precision quantization technique for deploying large language models on edge devices with reduced memory and computational requirements.
3/26/2026
The Geometric Price of Discrete Logic: Context-driven Manifold Dynamics of Number Representations
Research on how LLMs form discrete decision boundaries within continuous semantic spaces through context-driven topological distortion of number representations.
3/26/2026
Residual Attention Physics-Informed Neural Networks for Robust Multiphysics Simulation of Steady-State Electrothermal Energy Systems
Physics-informed neural networks using residual attention for steady-state electrothermal multiphysics simulation in energy systems.
3/26/2026
MetaKube: An Experience-Aware LLM Framework for Kubernetes Failure Diagnosis
MetaKube: LLM framework for Kubernetes failure diagnosis with Episodic Pattern Memory Network that learns from operational history to improve diagnostic accuracy over time.
3/26/2026
AI Generalisation Gap In Comorbid Sleep Disorder Staging
Deep learning model for automated sleep disorder staging from EEG with analysis of generalization gaps in clinical populations.
3/26/2026
A Theory of LLM Information Susceptibility
Theoretical framework analyzing fundamental performance limits when deploying fixed LLMs as optimization modules in agentic systems.
3/26/2026
Steering Code LLMs with Activation Directions for Language and Library Control
Method for steering code LLMs via activation space manipulation to control programming language and library preferences at inference time.
3/26/2026
Boost Like a (Var)Pro: Trust-Region Gradient Boosting via Variable Projection
VPBoost applies variable projection to gradient boosting for improved training of smooth parametric learners like neural networks.
3/26/2026
CDMT-EHR: A Continuous-Time Diffusion Framework for Generating Mixed-Type Time-Series Electronic Health Records
Continuous-time diffusion model for generating synthetic electronic health records with mixed numerical and categorical features.
3/26/2026
BXRL: Behavior-Explainable Reinforcement Learning
Framework for generating interpretable explanations of learned behaviors in RL agents with formal behavior definition.
3/26/2026
Self Paced Gaussian Contextual Reinforcement Learning
Curriculum learning approach for contextual RL using closed-form updates for self-paced task sequencing.
3/26/2026
Lightweight Fairness for LLM-Based Recommendations via Kernelized Projection and Gated Adapters
Lightweight fairness method for LLM-based recommenders using kernelized projection and adapters without fine-tuning.
3/26/2026
Probabilistic Geometric Alignment via Bayesian Latent Transport for Domain-Adaptive Foundation Models
Domain adaptation framework for foundation models using probabilistic geometric alignment and Bayesian transport.