Method using sparse autoencoders to discover language-specific features from monolingual data for controlling LLM output language without parallel data.
Discrete diffusion language model using tree-structured token prediction to reduce parameters and memory in language generation.
Video diffusion transformer framework for synthesizing diverse bimanual robot manipulation demonstrations from limited real data.
Framework combining score-based diffusion models as priors in plug-and-play optimization for imaging inverse problems.
Method for LLMs to dynamically compress intermediate reasoning thoughts into compact representations while maintaining reasoning quality.
Graph learning approach for melanoma detection in medical images using graph signal processing.
Framework for injecting bit-flip faults into DNNs used in autonomous driving systems to identify critical failure points.
Deep reinforcement learning framework for optimizing land-use allocation in Lake Malawi Basin to maximize ecosystem service value.
Debiased machine learning approach for conformal prediction of counterfactual outcomes under confounding.
Theoretical analysis of Sinkhorn-Knopp algorithm efficiency for entropically regularized optimal transport.
Method using Rényi attention entropy for patch pruning in transformers to reduce quadratic self-attention cost.
Theoretical analysis reconciling practitioner and statistician perspectives on Elo ranking algorithms.
Study on adversarial attacks against transformer-based malware detectors using control flow graphs, examining robustness of RoBERTa models.
SecureAFL: Asynchronous federated learning framework addressing straggler problem while maintaining security.
Study comparing LLM probed representations with performance on narrative analogical reasoning tasks.
PhaseFlow4D: Latent diffusion for 4D particle beam reconstruction from sparse 2D projections with physical constraints.
Machine learning attacks on Learning with Errors problem using data repetition and stepwise regression.
Secure-by-design GenAI framework integrating PromptShield for LLM-based cloud security and forensic analysis.
Fused multinomial logistic regression leveraging summary-level external machine-learning predictions for data integration.
Cross-Modal Graphical Lasso for learning interpretable multimodal representations by disentangling shared and specific topologies.
Value-based safety forecasting for streaming LLM outputs, improving response moderation on partial generations.
Fixed-confidence best arm identification in semiparametric bandits with instance-optimal sample complexity bounds.
Computer vision method addressing object occlusion through pattern masking and severity-informed classification.
Causal graph-attention approach to detect and mitigate hallucinations in LLMs for improved factual reliability.
Jellyfish: Zero-shot federated unlearning scheme using knowledge disentanglement for privacy-preserving federated learning.
Study on connectome-constrained neural networks and biological graph topology effects on learning efficiency.
TORA: Topology-first framework for 3D shape assembly using flow-matching and pretrained 3D encoders.
Fault detection framework for hybrid dynamical systems combining Petri nets with semi-supervised anomaly detection.
FactReview: LLM-based peer review system that grounds claims in evidence from papers, related work, and code to improve ML paper reviewing.
Real-time traffic monitoring system using YOLOv11 object detection and multi-object tracking in PyTorch/OpenCV.
Fine-tuned language models enhance embeddings for cognitive diagnosis in online education systems by incorporating semantic representations.
Event camera and neuromorphic hardware approach for efficient spacecraft pose estimation during autonomous rendezvous operations in space.
Comparison of CNN and CNN-Transformer models for robust speech recognition from MEG brain signals under distribution shift using LibriBrain benchmark.
Mathematical framework for primal-dual optimization methods handling nonsmooth nonconvex problems with orthogonality constraints on Stiefel manifolds.
Test-time adaptation approach for improving deep learning model generalization across geographic regions in remote sensing land surface temperature prediction.
Framework using non-equilibrium stochastic dynamics to address stability-plasticity dilemma in continual learning via Kramers escape theory.
First comprehensive benchmark for evaluating AI models on professional graphic design tasks including layout, typography, and design intent translation.
ML-driven workflow for multi-objective discovery in materials science using automated microscopy and characterization, avoiding premature convergence.
Study analyzing bias toward American English in LLMs through postcolonial lens, examining how data curation and geopolitical histories shape model development.
Asymptotic convergence analysis of Q-learning with linear decay to zero learning rates addressing persistent bias and slow convergence issues.
Formal framework and metrics for pedagogical safety in educational reinforcement learning, introducing Reward Hacking Severity Index to detect misalignment.
Extension of transmission neural network model incorporating inhibitory connections and neurotransmitter populations with firing probability characterization.
Combee framework for scaling prompt learning in LLM agents enabling efficient self-improvement through system prompt optimization across parallel runs.
MC-CPO method for constrained reinforcement learning in tutoring systems preventing reward hacking through mastery-conditioned safety constraints.
Theoretical work on Expectation Propagation convergence avoiding non-integrable beliefs and optimizing constrained Bethe Free Energy.
Position paper analyzing failure modes in agentic IR systems where early errors cascade despite linguistic fluency, causing misalignment between reasoning and execution.
Open foundation models for Radio Access Network time-series forecasting enabling AI-native optimization and closed-loop control with improved generalization.
System and analysis of personalized LLM customization for individual investor decision-making, identifying fundamental limitations in current personalization paradigms.
Out-of-air computation framework for structured extraction from wireless superposition using joint source-channel coding without pre-embedded computation.
Neural Willmore flow approach using neural architectures and PINN-style loss functions to minimize Willmore energy on 2D surfaces.