VISOR: efficiency method for vision-language models using sparse dynamically selected interactions instead of visual token reduction.
Data-driven and empirical formula models to quantify momentum in competitive tennis matches.
ML model for virtual screening in drug discovery handling out-of-distribution regions with extrapolatory pseudo-label matching.
Minimal Frame Averaging: framework for constructing provably minimal frames achieving exact equivariance in ML systems efficiently.
Continuous action representation for 3D floorplanning addressing scalability bottlenecks from discrete canvas coordinates.
Convergence analysis of linear temporal difference learning in reinforcement learning without requiring linearly independent features.
HFLDD: hybrid federated learning framework using dataset distillation to address non-IID data heterogeneity and label distribution skew.
DART-Eval: benchmark for evaluating DNA language models on regulatory DNA prediction, interpretation and design tasks.
MSA-CNN: lightweight multi-scale CNN with attention mechanism for automatic sleep stage classification from signal data.
BalanceKV: streaming algorithm for approximating attention computations in LLMs using geometric process to reduce memory requirements for long-context generation.
Federated learning approach for designing data-driven feedforward control in vehicle lateral dynamics using distributed data across multiple systems.
Expectation Reflection: multiplicative parameter update paradigm for ML optimization using observation-prediction ratios instead of additive gradient descent.
Analysis of geometric properties and flatness in neural architecture search spaces.
Information-theoretic framework for characterizing and measuring information leakage in concept-based models.
Federated LoRA fine-tuning method for large language models with communication-efficient sparsified updates.
Meta-optimization approach for generating generalizable heuristics using LLMs for combinatorial optimization.
Multi-experiment equation learning method for deriving analytical models from agent-based simulation data.
State representation learning from trajectories using minimum action distance metric for MDPs.
Foundation model adapter for time series forecasting with heterogeneous covariates and multimodal data.
Red teaming framework for systematically discovering diverse vulnerabilities in large language models.
Privacy-preserving graph structure learning using differential privacy at data publishing stage.
Spiking neural network framework with hyperparameter optimization for fraud detection.
Testing framework for deep learning systems using topographical feature discrimination.
Fourier-embedded operator learning framework for solving partial differential equations.
Deep temporal graph method for correcting GNSS positioning errors from jamming attacks.
Graph neural network architecture for modeling spatio-temporal signals with dynamic structure.
Federated learning framework for training deep models on resource-constrained edge devices.
Method for identifying counterfactuals from observational data using optimal transport theory.
Novel sampling algorithm for masked diffusion models improving generation quality and efficiency.
Deep learning model for channel state information prediction in MIMO systems with robustness testing.
Optimizes on-device semantic selection with cross-encoder rerankers for retrieval, agent memory, and recommendations via monolithic forwarding.
Scalable framework for automated desktop UI exploration to generate training data for LLM-based GUI understanding and automation.
Parameter-free clustering framework using self-supervised consensus maximization without requiring hyperparameter tuning.
Pathlet dictionary learning approach for robust and interpretable trajectory generation in privacy-preserving urban mobility applications.
Studies model inversion attacks on latent diffusion models, showing non-uniform memorization patterns in latent space.
Theoretical analysis of feature learning mechanisms and implicit bias in deep networks, proposing analytical scaling arguments.
Presents Arc Gradient Descent optimizer with phase-aware step dynamics, evaluated on Rosenbrock function and ML datasets.
Proposes differentially private federated learning optimization using regularized Fisher information matrix for faster convergence under privacy constraints.
Uses Chernoff information to characterize trade-offs between fairness, privacy, and accuracy in machine learning systems.
Theoretical analysis of input-connected MLPs with direct connections from input to hidden neurons and universal approximation properties.
Introduces homomorphism error metric to measure representational inconsistencies and predict compositional generalization failures in transformers.
Analyzes forecast uncertainty in ML model explainability, arguing uncertainty at decision boundaries explains LIME/SHAP instability.
Brain-inspired routing method with temporal-ensemble experts for general continual learning from non-stationary data streams.
Proposes one-to-one channel-head binding method for imputing missing values in multivariate time series data.
Studies robustness of PPO reinforcement learning under sensor drift using temporal sequence models to handle partial observability.
arXiv paper MJ1: multimodal judge trained with RL enforcing visual grounding through structured verification chains and counterfactual consistency rewards.
arXiv paper proposing WiFi CSI sensing framework handling station-wise feature missingness and limited labeled data in multi-station deployments.
arXiv paper SAFE-PIT-CM: autoencoder with frozen Euler solver for recovering material diffusion coefficients from continuum mechanics data.
arXiv paper proposing key deletion approach for machine unlearning designed at model development stage rather than post-hoc, addressing privacy regulations and data errors.
arXiv paper PRISM: empirical study of mid-training design choices across 7 LLM base models showing consistent +15 to +40 point gains from 27B token sequences.