4/23/2026
Explicit Dropout: Deterministic Regularization for Transformer Architectures
Deterministic dropout formulation expressing regularization as additive loss terms for Transformer architectures with explicit optimization objectives.
4/23/2026
CHASM: Unveiling Covert Advertisements on Chinese Social Media
CHASM dataset and benchmark for evaluating multimodal LLMs on detecting covert advertisements in Chinese social media moderation tasks.
4/23/2026
Differentially Private Clustered Federated Learning with Privacy-Preserving Initialization and Normality-Driven Aggregation
Federated learning framework with differential privacy and privacy-preserving initialization for heterogeneous cross-device deployments.
4/23/2026
Too Sharp, Too Sure: When Calibration Follows Curvature
Study of relationship between model calibration and loss surface curvature in neural networks, showing calibration emerges during training on vision tasks.
4/23/2026
Occupancy Reward Shaping: Improving Credit Assignment for Offline Goal-Conditioned Reinforcement Learning
Occupancy reward shaping method extracting temporal information from generative world models to improve credit assignment in offline goal-conditioned reinforcement learning.
4/23/2026
GRPO-VPS: Enhancing Group Relative Policy Optimization with Verifiable Process Supervision for Effective Reasoning
GRPO-VPS extends Group Relative Policy Optimization with verifiable process supervision to improve credit assignment for LLM reasoning tasks.
4/23/2026
Variance Is Not Importance: Structural Analysis of Transformer Compressibility Across Model Scales
Systematic empirical study of transformer compression across 40+ experiments on GPT-2 and Mistral 7B identifying five structural properties relevant to model compression.
4/23/2026
MGDA-Decoupled: Geometry-Aware Multi-Objective Optimisation for DPO-based LLM Alignment
MGDA-Decoupled proposes geometry-aware multi-objective optimization for DPO-based LLM alignment balancing helpfulness, truthfulness, and harmlessness objectives.
4/23/2026
COMPASS: COntinual Multilingual PEFT with Adaptive Semantic Sampling
COMPASS framework using parameter-efficient fine-tuning (PEFT) and adaptive semantic sampling to improve multilingual LLM performance and reduce cross-lingual interference.
4/23/2026
Tokenised Flow Matching for Hierarchical Simulation Based Inference
Tokenised flow matching technique for hierarchical simulation-based inference to reduce simulator evaluation costs through likelihood factorization.
4/23/2026
Supplement Generation Training for Enhancing Agentic Task Performance
Proposes Supplement Generation Training (SGT), a method to train smaller LLMs to generate supplemental text for improving agentic task performance without expensive post-training.
4/23/2026
Near-Future Policy Optimization
Research on reinforcement learning with verifiable rewards (RLVR) exploring mixed-policy methods to improve convergence by combining off-policy and on-policy trajectories.
4/23/2026
Fast Bayesian equipment condition monitoring via simulation based inference: applications to heat exchanger health
Simulation-based Bayesian inference framework for real-time industrial equipment condition monitoring that overcomes MCMC computational bottlenecks.
4/23/2026
F\textsuperscript{2}LP-AP: Fast \& Flexible Label Propagation with Adaptive Propagation Kernel
F²LP-AP combines label propagation with adaptive kernels for fast semi-supervised node classification on heterophilous graphs without expensive iterative training.
4/23/2026
Lifecycle-Aware Federated Continual Learning in Mobile Autonomous Systems
Federated continual learning approach for autonomous vehicle fleets that adapts to evolving terrain while preventing catastrophic forgetting across network layers.
4/23/2026
ParetoSlider: Diffusion Models Post-Training for Continuous Reward Control
ParetoSlider framework extends diffusion model post-training with multi-objective reward control, enabling inference-time trade-off adjustment between conflicting goals without early scalarization.
4/23/2026
Stream-CQSA: Avoiding Out-of-Memory in Attention Computation via Flexible Workload Scheduling
Stream-CQSA algorithm enables efficient long-context LLM attention computation by removing full tensor fitting requirement, reducing memory complexity from quadratic to near-linear.
4/23/2026
FedSIR: Spectral Client Identification and Relabeling for Federated Learning with Noisy Labels
FedSIR multi-stage federated learning framework using spectral client identification and relabeling for handling noisy labels in distributed training.
4/23/2026
Algorithm Selection with Zero Domain Knowledge via Text Embeddings
ZeroFolio feature-free algorithm selection method using pretrained text embeddings and k-nearest neighbors without domain knowledge.
4/23/2026
Exploring Data Augmentation and Resampling Strategies for Transformer-Based Models to Address Class Imbalance in AI Scoring of Scientific Explanations in NGSS Classroom
Study on data augmentation and transformer-based models addressing class imbalance in automated scoring of student scientific explanations.
4/23/2026
Explainable AML Triage with LLMs: Evidence Retrieval and Counterfactual Checks
LLM approach for anti-money laundering transaction triage with explainable evidence retrieval and counterfactual verification for regulated workflows.
4/23/2026
ThermoQA: A Three-Tier Benchmark for Evaluating Thermodynamic Reasoning in Large Language Models
ThermoQA benchmark of 293 thermodynamics problems across three tiers evaluating reasoning capabilities in six frontier LLMs.
4/23/2026
EvoForest: A Novel Machine-Learning Paradigm via Open-Ended Evolution of Computational Graphs
EvoForest machine learning paradigm using open-ended evolution of computational graphs for discovering transformations in structured prediction.
4/23/2026
TTKV: Temporal-Tiered KV Cache for Long-Context LLM Inference
TTKV temporal-tiered KV cache system reducing memory footprint for long-context LLM inference using hierarchical precision approach.
4/23/2026
Development and Preliminary Evaluation of a Domain-Specific Large Language Model for Tuberculosis Care in South Africa
Domain-specific LLM for tuberculosis care in South Africa developed to assist patients and healthcare providers.
4/23/2026
Hidden Reliability Risks in Large Language Models: Systematic Identification of Precision-Induced Output Disagreements
Systematic study identifying precision-induced output disagreements in LLMs across floating-point and quantized formats affecting reliability.
4/23/2026
SkillGraph: Graph Foundation Priors for LLM Agent Tool Sequence Recommendation
SkillGraph system for LLM agent tool sequence recommendation using execution-transition graphs mined from 49,831 successful workflows.
4/23/2026
Improving Molecular Force Fields with Minimal Temporal Information
Neural network approach for predicting molecular energies and forces incorporating temporal information from molecular dynamics simulations.
4/23/2026
MIRROR: A Hierarchical Benchmark for Metacognitive Calibration in Large Language Models
MIRROR benchmark with 250,000 evaluation instances across 16 LLMs measuring metacognitive calibration and self-knowledge in decision-making.
4/23/2026
Co-Located Tests, Better AI Code: How Test Syntax Structure Affects Foundation Model Code Generation
Large-scale empirical study of 830+ files across 12 LLM models examining how test code structure affects AI code generation quality.
4/23/2026
ChipCraftBrain: Validation-First RTL Generation via Multi-Agent Orchestration
Multi-agent LLM framework for RTL code generation achieving 95.9% functional correctness with validation-first approach and synthesis awareness.
4/23/2026
From Signal Degradation to Computation Collapse: Uncovering the Two Failure Modes of LLM Quantization
Systematic mechanistic analysis of LLM quantization failure modes, distinguishing between signal degradation and computation collapse in 2-bit vs 4-bit quantization.
4/23/2026
DistortBench: Benchmarking Vision Language Models on Image Distortion Identification
DistortBench: Diagnostic benchmark with 13,500 questions evaluating vision-language models on image distortion perception across 27 types.
4/23/2026
Separable Pathways for Causal Reasoning: How Architectural Scaffolding Enables Hypothesis-Space Restructuring in LLM Agents
Extends blicket detector paradigm to test AI agents' capacity for causal reasoning and hypothesis space restructuring through experimentation.
4/23/2026
Bootstrapping Post-training Signals for Open-ended Tasks via Rubric-based Self-play on Pre-training Text
Self-play paradigm for LLM training using rubric-based reward models on pre-training text with minimal supervision for open-ended tasks.
4/23/2026
SkillLearnBench: Benchmarking Continual Learning Methods for Agent Skill Generation on Real-World Tasks
SkillLearnBench: First benchmark for evaluating continual skill learning in LLM agents across 20 real-world tasks in 15 domains.
4/23/2026
HiPO: Hierarchical Preference Optimization for Adaptive Reasoning in LLMs
HiPO extends Direct Preference Optimization for LLMs with hierarchical feedback on reasoning steps, addressing limitations in complex task alignment.
4/23/2026
Pre-Execution Query Slot-Time Prediction in Cloud Data Warehouses: A Feature-Scoped Machine Learning Approach
Feature-scoped ML approach predicts query slot-time in cloud data warehouses before execution for cost estimation.
4/23/2026
Robust Out-of-Distribution Stochastic Optimization
Framework for robust stochastic optimization under distribution shift using relevant source distributions for decision-making.
4/23/2026
Meta-Tool: Efficient Few-Shot Tool Adaptation for Small Language Models
Meta-Tool empirically compares hypernetwork LoRA adaptation vs few-shot prompting for tool-use in small language models.
4/23/2026
Toward Safe Autonomous Robotic Endovascular Interventions using World Models
World model-based reinforcement learning approach for safe autonomous robotic endovascular intervention with variable anatomies.
4/23/2026
WildFireVQA: A Large-Scale Radiometric Thermal VQA Benchmark for Aerial Wildfire Monitoring
WildFireVQA benchmark combines RGB and thermal imagery for visual question answering on aerial wildfire monitoring.
4/23/2026
Mol-Debate: Multi-Agent Debate Improves Structural Reasoning in Molecular Design
Mol-Debate uses multi-agent debate to improve structural reasoning for text-guided molecular design in drug discovery.
4/23/2026
RADS: Reinforcement Learning-Based Sample Selection Improves Transfer Learning in Low-resource and Imbalanced Clinical Settings
Reinforcement learning-based sample selection improves transfer learning in low-resource, imbalanced clinical settings.
4/23/2026
AROMA: Augmented Reasoning Over a Multimodal Architecture for Virtual Cell Genetic Perturbation Modeling
AROMA system uses augmented reasoning over multimodal architecture for virtual cell genetic perturbation prediction.
4/23/2026
Surrogate modeling for interpreting black-box LLMs in medical predictions
Surrogate modeling framework to interpret black-box LLM knowledge and explain medical predictions quantitatively.
4/23/2026
AI models of unstable flow exhibit hallucination
Systematic study of hallucination in AI models of fluid dynamics, showing physically implausible but visually realistic predictions.
4/23/2026
WebGen-R1: Incentivizing Large Language Models to Generate Functional and Aesthetic Websites with Reinforcement Learning
WebGen-R1 uses reinforcement learning to train small LLMs to generate functional, multi-page websites end-to-end.
4/23/2026
DialToM: A Theory of Mind Benchmark for Forecasting State-Driven Dialogue Trajectories
DialToM benchmark evaluates theory of mind reasoning in LLMs through human-verified dialogue trajectory forecasting tasks.
4/23/2026
VTouch++: A Multimodal Dataset with Vision-Based Tactile Enhancement for Bimanual Manipulation
VTouch++ dataset for bimanual robotic manipulation with vision-based tactile sensing for contact-rich tasks.