3/17/2026
Ghosts of Softmax: Complex Singularities That Limit Safe Step Sizes in Cross-Entropy
Analysis of complex singularities in cross-entropy loss that limit safe optimization step sizes during LLM training.
3/17/2026
Semantic Aware Feature Extraction for Enhanced 3D Reconstruction
Semantic-aware feature extraction framework combining geometric and semantic information for improved 3D reconstruction and feature matching.
3/17/2026
An Empirical Investigation of Pre-Trained Deep Learning Model Reuse in the Scientific Process
Empirical study on how scientists reuse pre-trained deep learning models to reduce computational costs and training complexity.
3/17/2026
Volumetric Radar Echo Motion Estimation Using Physics-Informed Deep Learning: A Case Study Over Slovakia
Physics-informed CNN for volumetric radar motion estimation in precipitation nowcasting with altitude-dependent motion fields.
3/17/2026
NCCL EP: Towards a Unified Expert Parallel Communication API for NCCL
NCCL EP: unified expert parallel communication API for Mixture-of-Experts LLM training using GPU-initiated RDMA operations.
3/17/2026
Locatability-Guided Adaptive Reasoning for Image Geo-Localization with Vision-Language Models
Vision-language model approach for image geo-localization using locatability-guided reasoning to reduce hallucinations and improve accuracy.
3/17/2026
Widespread Gender and Pronoun Bias in Moral Judgments Across LLMs
Study demonstrating gender and pronoun bias in LLM moral judgments using controlled sentence-level counterfactual analysis.
3/17/2026
Benchmarking Large Language Models on Reference Extraction and Parsing in the Social Sciences and Humanities
Benchmark evaluating LLM performance on bibliographic reference extraction in multilingual SSH texts with complex citation formats.
3/17/2026
Unsupervised Adaptation from FDG to PSMA PET/CT for 3D Lesion Detection under Label Shift
Unsupervised domain adaptation framework for 3D lesion detection in medical imaging with label shift compensation between PET tracers.
3/17/2026
SHAMISA: SHAped Modeling of Implicit Structural Associations for Self-supervised No-Reference Image Quality Assessment
SHAMISA: self-supervised framework for no-reference image quality assessment using structured relational supervision on unlabeled data.
3/17/2026
Locally Linear Continual Learning for Time Series based on VC-Theoretical Generalization Bounds
SyMPLER: explainable piecewise linear regression model for nonstationary time series forecasting with VC-theoretical generalization bounds.
3/17/2026
Preconditioned Test-Time Adaptation for Out-of-Distribution Debiasing in Narrative Generation
CAP-TTA test-time adaptation framework using context-aware LoRA updates to address out-of-distribution bias in narrative generation when bias risk is detected.
3/17/2026
$\tau$-Voice: Benchmarking Full-Duplex Voice Agents on Real-World Domains
τ-Voice benchmark for evaluating full-duplex voice agents on real-world tasks requiring multi-turn conversations, policy adherence, and environment interaction.
3/17/2026
QuarkMedBench: A Real-World Scenario Driven Benchmark for Evaluating Large Language Models
QuarkMedBench realistic medical benchmark for evaluating LLMs on unstructured, ambiguous real-world healthcare queries beyond standardized exam questions.
3/17/2026
Routing Channel-Patch Dependencies in Time Series Forecasting with Graph Spectral Decomposition
Graph spectral decomposition approach balancing channel-independent and channel-dependent strategies for improved time series forecasting generalization.
3/17/2026
REFINE-DP: Diffusion Policy Fine-tuning for Humanoid Loco-manipulation via Reinforcement Learning
REFINE-DP method fine-tuning diffusion policies via reinforcement learning for humanoid loco-manipulation with improved tracking and long-horizon task execution.
3/17/2026
R3-REC: Reasoning-Driven Recommendation via Retrieval-Augmented LLMs over Multi-Granular Interest Signals
R3-REC framework using retrieval-augmented LLMs with multi-level user intent reasoning for sequential recommendation with sparse and noisy data.
3/17/2026
Implicit Maximum Likelihood Estimation for Real-time Generative Model Predictive Control
Implicit maximum likelihood estimation approach accelerating diffusion-based trajectory planning for real-time model predictive control applications.
3/17/2026
Level Up: Defining and Exploiting Transitional Problems for Curriculum Learning
Curriculum learning approach using dynamic difficulty estimation based on gradient information to improve training efficiency and task performance.
3/17/2026
Knowledge Distillation for Large Language Models
Knowledge distillation framework compressing Qwen 3B to 0.5B using chain-of-thought reinforcement learning across English, Spanish, and code datasets.
3/17/2026
Retrieval-Feedback-Driven Distillation and Preference Alignment for Efficient LLM-based Query Expansion
Retrieval-feedback-driven distillation framework transferring query expansion behavior from teacher to student LLM for efficient deployment in retrieval systems.
3/17/2026
Generate Then Correct: Single Shot Global Correction for Aspect Sentiment Quad Prediction
Generate-then-correct approach for aspect sentiment quad prediction extracting aspect terms, categories, opinions, and sentiment polarity from text.
3/17/2026
AD-Copilot: A Vision-Language Assistant for Industrial Anomaly Detection via Visual In-context Comparison
AD-Copilot system enhancing multimodal LLMs for industrial anomaly detection through visual in-context comparison and domain-specific adaptation.
3/17/2026
IGU-LoRA: Adaptive Rank Allocation via Integrated Gradients and Uncertainty-Aware Scoring
IGU-LoRA method for adaptive rank allocation in LoRA fine-tuning using integrated gradients and uncertainty scoring to improve parameter-efficient LLM adaptation.
3/17/2026
Computation and Communication Efficient Federated Unlearning via On-server Gradient Conflict Mitigation and Expression
Federated unlearning framework reducing computational and communication costs for removing participant data from federated learning models while maintaining privacy.
3/17/2026
Prototypical Exemplar Condensation for Memory-efficient Online Continual Learning
ML research: memory-efficient continual learning via prototypical exemplar condensation, reducing samples-per-class requirements below 20.
3/17/2026
Collapse or Preserve: Data-Dependent Temporal Aggregation for Spiking Neural Network Acceleration
ML research: proposes Temporal Aggregated Convolution to efficiently execute spiking neural networks on SIMD architectures.
3/17/2026
Evaluating Semantic Fragility in Text-to-Audio Generation Systems Under Controlled Prompt Perturbations
ML research: evaluates semantic robustness of text-to-audio generation under prompt variations; reveals fragility to linguistic changes.
3/17/2026
Fine-tuning is Not Enough: A Parallel Framework for Collaborative Imitation and Reinforcement Learning in End-to-end Autonomous Driving
ML research: parallel framework combining imitation and reinforcement learning for autonomous driving, addressing policy drift in sequential fine-tuning.
3/17/2026
Sirens' Whisper: Inaudible Near-Ultrasonic Jailbreaks of Speech-Driven LLMs
SWhisper: Framework for inaudible near-ultrasonic prompt injection attacks against speech-driven LLMs using commodity hardware.
3/17/2026
APEX-Searcher: Augmenting LLMs' Search Capabilities through Agentic Planning and Execution
APEX-Searcher: Multi-round retrieval system combining agentic planning and LLM execution for complex information seeking tasks.
3/17/2026
Step-CoT: Stepwise Visual Chain-of-Thought for Medical Visual Question Answering
Step-CoT dataset and method for medical visual question answering with expert-curated structured reasoning supervision.
3/17/2026
Multi-Modal Character Localization and Extraction for Chinese Text Recognition
Multi-modal method for Chinese text character localization and extraction addressing complexity of scene text recognition.
3/17/2026
Large Language Models Reproduce Racial Stereotypes When Used for Text Annotation
Study showing LLMs reproduce racial stereotypes in text annotation across 19 models and 4M+ judgments, demonstrating bias issues.
3/17/2026
UVLM: A Universal Vision-Language Model Loader for Reproducible Multimodal Benchmarking
UVLM: Unified framework for loading and benchmarking multiple vision-language models across architectures in Google Colab.
3/17/2026
Robust Self-Training with Closed-loop Label Correction for Learning from Noisy Labels
Self-training framework with label correction for training neural networks on noisy datasets using bilevel optimization.
3/17/2026
Pixel-level Scene Understanding in One Token: Visual States Need What-is-Where Composition
Visual state representation for robotic agents encoding what-is-where composition from video for sequential decision making.
3/17/2026
FedPBS: Proximal-Balanced Scaling Federated Learning Model for Robust Personalized Training for Non-IID Data
FedPBS federated learning algorithm addressing statistical heterogeneity and uneven client participation in distributed training.
3/17/2026
SmoothVLA: Aligning Vision-Language-Action Models with Physical Constraints via Intrinsic Smoothness Optimization
SmoothVLA: Training method for vision-language-action robotic models balancing stability and exploration via smoothness constraints.
3/17/2026
Discriminative Flow Matching Via Local Generative Predictors
Discriminative flow matching framework reformulating classification and object detection as conditional transport problems.
3/17/2026
Iterative Semantic Reasoning from Individual to Group Interests for Generative Recommendation with LLMs
Recommendation system using LLMs for semantic reasoning from individual to group interests for generative recommendations.
3/17/2026
LLM-Guided Reinforcement Learning for Audio-Visual Speech Enhancement
Speech enhancement framework combining audio-visual processing with reinforcement learning and LLM-based interpretable reward models.
3/17/2026
Chunk-Guided Q-Learning
Chunk-Guided Q-Learning algorithm for offline reinforcement learning addressing bootstrapping error and policy class restrictions.
3/17/2026
FLUX: Data Worth Training On
FLUX: Data preprocessing pipeline for LLM training that balances scale and quality without sacrificing token efficiency or data integrity.
3/17/2026
Human-like Object Grouping in Self-supervised Vision Transformers
Behavioral benchmark comparing self-supervised vision transformer object grouping with human perception using same/different judgments on naturalistic scenes.
3/17/2026
A Benchmark for Multi-Party Negotiation Games from Real Negotiation Data
Benchmark for multi-party negotiation games with configurable generator and value-function approximations for evaluating agent decision-making.
3/17/2026
Towards Agentic Honeynet Configuration
Dynamic honeypot deployment system using AI agents to adaptively select assets based on evolving attacker tactics and threat intelligence.
3/17/2026
Diffusion Reinforcement Learning via Centered Reward Distillation
Reinforcement learning fine-tuning for diffusion models using centered reward distillation to improve prompt fidelity and compositional correctness.
3/17/2026
Align Forward, Adapt Backward: Closing the Discretization Gap in Logic Gate Networks
Analysis of training-inference mismatch in logic gate networks using hard vs soft selection and Gumbel noise.
3/17/2026
Citation-Enforced RAG for Fiscal Document Intelligence: Cited, Explainable Knowledge Retrieval in Tax Compliance
Citation-enforced RAG system for tax compliance enabling explainable and cited knowledge retrieval from fiscal documents.