3/12/2026
HTM-EAR: Importance-Preserving Tiered Memory with Hybrid Routing under Saturation
HTM-EAR hierarchical memory system for long-running agents combining importance-aware eviction with HNSW-based working memory and archival storage.
3/12/2026
Evaluating Progress in Graph Foundation Models: A Comprehensive Benchmark and New Insights
Benchmark evaluating graph foundation models across topic and format domain shifts to assess knowledge transfer capabilities.
3/12/2026
Targeted Bit-Flip Attacks on LLM-Based Agents
Research paper introducing Flip-Agent, first framework for targeted bit-flip attacks on multi-stage LLM-based agent pipelines with external tools.
3/12/2026
Safety Under Scaffolding: How Evaluation Conditions Shape Measured Safety
Large-scale controlled study (N=62,808) on how agentic scaffolds affect measured safety in LLM deployments across six frontier models and four configurations.
3/12/2026
Gated Adaptation for Continual Learning in Human Activity Recognition
Research paper on continual learning for wearable sensor activity recognition, addressing catastrophic forgetting in IoT human activity recognition systems.
3/12/2026
Toward Epistemic Stability: Engineering Consistent Procedures for Industrial LLM Hallucination Reduction
Research paper comparing five prompt engineering strategies to reduce LLM hallucinations and increase output consistency in industrial high-stakes applications.
3/12/2026
Revisiting Sharpness-Aware Minimization: A More Faithful and Effective Implementation
Research paper on improved implementation of Sharpness-Aware Minimization (SAM), an optimization technique that enhances model generalization by minimizing loss in parameter neighborhoods.
3/12/2026
InFusionLayer: a CFA-based ensemble tool to generate new classifiers for learning and modeling
Python tool implementing Combinatorial Fusion Analysis for ensemble learning classifier generation using rank-score characteristics and cognitive diversity.
3/12/2026
Training Language Models via Neural Cellular Automata
Proposes neural cellular automata for synthetic data generation in LLM pre-training to address natural language limitations and reduce human bias.
3/12/2026
SBOMs into Agentic AIBOMs: Schema Extensions, Agentic Orchestration, and Reproducibility Evaluation
Agentic AIBOMs extends Software Bills of Materials to capture runtime behavior and reproducibility in AI systems, addressing supply-chain security for dynamic execution.
3/12/2026
Tool Receipts, Not Zero-Knowledge Proofs: Practical Hallucination Detection for AI Agents
NabaOS framework for detecting hallucinations in tool-using AI agents via lightweight verification receipts, avoiding expensive zero-knowledge proof overhead.
3/12/2026
Multi-Agent Memory from a Computer Architecture Perspective: Visions and Challenges Ahead
Position paper addressing memory architecture challenges in multi-agent LLM systems, proposing three-layer hierarchy and identifying protocol gaps for cache sharing and memory access control.
3/12/2026
HTMuon: Improving Muon via Heavy-Tailed Spectral Correction
HTMuon optimizer improves upon Muon by preserving heavy-tailed weight spectra for more effective LLM training using spectral correction.
3/12/2026
ADVERSA: Measuring Multi-Turn Guardrail Degradation and Judge Reliability in Large Language Models
ADVERSA framework measuring guardrail degradation in multi-turn LLM interactions using automated red-teaming with 70B attacker model.
3/12/2026
Dissecting Chronos: Sparse Autoencoders Reveal Causal Feature Hierarchies in Time Series Foundation Models
Sparse autoencoders applied to time series foundation model (Chronos) revealing causal feature hierarchies through ablation experiments.
3/12/2026
Why LLMs Fail: A Failure Analysis and Partial Success Measurement for Automated Security Patch Generation
Failure analysis of LLM-generated security patches across 319 examples showing only 24.8% achieve full correctness on security vulnerabilities.
3/12/2026
Amnesia: Adversarial Semantic Layer Specific Activation Steering in Large Language Models
Adversarial semantic layer activation steering technique for red-teaming and suppressing harmful content generation in LLMs.
3/12/2026
KernelSkill: A Multi-Agent Framework for GPU Kernel Optimization
Multi-agent framework using LLMs to optimize GPU kernels with explicit, interpretable optimization strategies replacing opaque heuristics.
3/12/2026
ES-dLLM: Efficient Inference for Diffusion Large Language Models by Early-Skipping
ES-dLLM improves diffusion language model inference efficiency through early-skipping of redundant context processing iterations.
3/12/2026
Multi-Stream Perturbation Attack: Breaking Safety Alignment of Thinking LLMs Through Concurrent Task Interference
Multi-Stream Perturbation Attack exploits vulnerabilities in LLM thinking mode by concurrent task interference to bypass safety alignment.
3/12/2026
Execution Is the New Attack Surface: Survivability-Aware Agentic Crypto Trading with OpenClaw-Style Local Executors
Analysis of safety risks in agentic LLM systems using local executors and tool use, focusing on execution-layer attack surfaces and capability supply chains.
3/12/2026
Equivariant Asynchronous Diffusion: An Adaptive Denoising Schedule for Accelerated Molecular Conformation Generation
Equivariant asynchronous diffusion model with adaptive denoising schedule for 3D molecular conformation generation capturing hierarchical structure.
3/12/2026
Code-Space Response Oracles: Generating Interpretable Multi-Agent Policies with Large Language Models
Code-Space Response Oracles (CSRO) generates interpretable multi-agent policies using LLMs instead of black-box neural networks for game-theoretic equilibria.
3/12/2026
CLIPO: Contrastive Learning in Policy Optimization Generalizes RLVR
CLIPO framework improving LLM reasoning by using contrastive learning in policy optimization to evaluate intermediate reasoning step correctness.
3/12/2026
Lost in the Middle at Birth: An Exact Theory of Transformer Position Bias
Analysis showing 'Lost in the Middle' performance degradation in LLMs is geometrically inherent at transformer initialization, predating training.
3/12/2026
AR-VLA: True Autoregressive Action Expert for Vision-Language-Action Models
Autoregressive Action Expert for Vision-Language-Action models maintaining long-lived memory context for robot/agent control tasks.
3/12/2026
Mashup Learning: Faster Finetuning by Remixing Past Checkpoints
Method for faster LLM finetuning by reusing and remixing existing model checkpoints instead of training from scratch on each new task.
3/12/2026
Compatibility at a Cost: Systematic Discovery and Exploitation of MCP Clause-Compliance Vulnerabilities
Identifies and exploits MCP clause-compliance vulnerabilities in Model Context Protocol standard for agent-tool integration.
3/12/2026
MCP-in-SoS: Risk assessment framework for open-source MCP servers
Develops risk assessment framework for open-source MCP servers enabling LLM agents to access external tools securely.
3/12/2026
Adaptive Activation Cancellation for Hallucination Mitigation in Large Language Models
Presents Adaptive Activation Cancellation framework treating hallucinations as interference in transformer residual streams for real-time mitigation.
3/12/2026
Delta-K: Boosting Multi-Instance Generation via Cross-Attention Augmentation
Proposes Delta-K, inference framework addressing concept omission in multi-instance diffusion image generation.
3/12/2026
A Diffusion Analysis of Policy Gradient for Stochastic Bandits
Analyzes policy gradient for k-armed stochastic bandits using continuous-time diffusion approximation with regret bounds.
3/12/2026
Rethinking the Harmonic Loss via Non-Euclidean Distance Layers
Studies harmonic loss with non-Euclidean distance layers as alternative to cross-entropy for neural network training.
3/12/2026
DUCTILE: Agentic LLM Orchestration of Engineering Analysis in Product Development Practice
Presents DUCTILE, agentic LLM orchestration system for automating engineering analysis in product development workflows.
3/12/2026
Conversational AI-Enhanced Exploration System to Query Large-Scale Digitised Collections of Natural History Museums
Designs conversational AI system for querying 1.7M digitized natural history museum specimens using semantic search.
3/12/2026
Simulation-in-the-Reasoning (SiR): A Conceptual Framework for Empirically Grounded AI in Autonomous Transportation
Introduces Simulation-in-the-Reasoning framework embedding domain simulators into LLM reasoning loops for autonomous transportation.
3/12/2026
Is this Idea Novel? An Automated Benchmark for Judgment of Research Ideas
Develops automated benchmark for evaluating novelty of research ideas using LLMs to address literature review scalability.
3/12/2026
Overcoming Visual Clutter in Vision Language Action Models via Concept-Gated Visual Distillation
Vision-Language-Action model improvement via concept-gated visual distillation for robotic manipulation in cluttered scenes.
3/12/2026
Federated Active Learning Under Extreme Non-IID and Global Class Imbalance
Federated active learning study addressing class imbalance and non-IID data in privacy-preserving annotation.
3/12/2026
Mitigating Translationese Bias in Multilingual LLM-as-a-Judge via Disentangled Information Bottleneck
Research on translationese bias in multilingual LLM evaluators; proposes information bottleneck solution for low-resource languages.
3/12/2026
Utility Function is All You Need: LLM-based Congestion Control
LLM-based congestion control protocol using utility functions for network applications to optimize sending rates in distributed settings.
3/12/2026
Dynamic Knowledge Fusion for Multi-Domain Dialogue State Tracking
Dynamic knowledge fusion approach for multi-domain dialogue state tracking in task-oriented dialogue systems with limited training data.
3/12/2026
Beyond Interleaving: Causal Attention Reformulations for Generative Recommender Systems
Attention reformulation for generative recommender systems addressing structural inefficiencies of interleaving tokens in sequence generation.
3/12/2026
Few-Shot Adaptation to Non-Stationary Environments via Latent Trend Embedding for Robotics
Few-shot adaptation framework for robots in non-stationary environments using latent trend embeddings to handle concept shift.
3/12/2026
Reactive Writers: How Co-Writing with AI Changes How We Engage with Ideas
Mixed-methods study examining how AI co-writing tools change user engagement with ideas and opinion formation through behavioral analysis.
3/12/2026
Causal Concept Graphs in LLM Latent Space for Stepwise Reasoning
Causal Concept Graphs method using sparse autoencoders and differentiable structure learning to interpret multi-step reasoning in LLMs.
3/12/2026
Optimal Expert-Attention Allocation in Mixture-of-Experts: A Scalable Law for Dynamic Model Design
Neural scaling laws for Mixture-of-Experts models defining optimal compute allocation between expert and attention sub-layers.
3/12/2026
Safe Probabilistic Planning for Human-Robot Interaction using Conformal Risk Control
Safety framework for human-robot interaction combining control barrier functions with conformal risk control for formal guarantees.
3/12/2026
On the Learning Dynamics of Two-layer Linear Networks with Label Noise SGD
Theoretical analysis of SGD learning dynamics in two-layer linear networks trained with label noise, studying implicit bias mechanisms.
3/12/2026
Designing Service Systems from Textual Evidence
Framework using LLMs to design service systems by analyzing textual evidence from customer support and compliance data to optimize configurations.