De Jure is an automated pipeline for extracting structured regulatory rules from legal documents using LLM self-refinement, requiring no human annotation or domain-specific data.
Analysis of token initialization strategies for new vocabulary in language models used for generative recommendation systems.
Video diffusion research on multi-agent action binding in generative video games for world models.
Risk-aware linear bandits optimization with application to smart order routing in financial decision-making.
ML research on double acceleration of distributed optimization through local training and compression in gradient descent.
Theoretical analysis showing transformers can solve non-linear non-Markovian stochastic filtering for conditionally Gaussian signals.
Moonwalk: inverse-forward differentiation technique addressing backpropagation's memory requirements for training deeper networks.
Industrial anomaly detection approach using ExIFFI for fast, interpretable explanations in Industry 5.0 processes.
Comprehensive benchmark of 17 graph pooling methods across 28 datasets evaluating effectiveness, robustness and generalization.
Neural architecture using selective logical operators for interpretable classification with adaptive AND/OR operations.
Study of in-context learning in LLMs with spurious correlations, examining transformer robustness to spurious features in classification.
Automated neural architecture selection for time series forecasting comparing LSTMs, GRUs, Transformers, and State-Space Models.
ML research on classification metrics accounting for confidence in incorrect predictions for safety-critical applications.
Foundation model for single-cell RNA sequencing analysis in disease biology and drug discovery applications.
Scientific ML research on training neural differential-algebraic equation systems extending neural ODE methods.
GradPower: lightweight gradient transformation technique for accelerating LLM pre-training via sign-power transformation, requiring minimal code changes.
ML research introducing Reliable Policy Iteration variants that maintain theoretical guarantees under function approximation in reinforcement learning.
Research on improving CNN interpretability by accounting for softmax dependency on logit differences rather than absolute values in class activation mapping.
Investigation of Generalized Gaussian mechanism for differential privacy in machine learning as alternative to Laplace/Gaussian.
GAN-based residual guided training strategy for Physics-Informed Transformers solving nonlinear PDEs.
Low-rank amortized Bayesian meta-learning for LLMs enabling few-shot learning across multiple datasets efficiently.
TANDEM: attention-guided neural differential equations for time series classification with missing data handling.
DeepIMC: machine learning framework for fast calibration of agent-based epidemic models via bidirectional LSTM.
GCond: scalable gradient conflict resolution for large-scale multi-task learning using accumulation-based stabilization.
StelLA: geometry-aware extension of LoRA using three-factor decomposition on Stiefel Manifold for efficient model fine-tuning.
Reinforcement learning approach for task offloading on Internet of Wearable Things to overcome battery and computation constraints.
Analysis of what temporal graph learning models learn, addressing reliability concerns in benchmark evaluation protocols.
Rigorous approach for determining rank identifiability in tensor factorization using prior predictive moment matching.
Strategic Doubly Robust estimator framework for causal inference in strategic equilibrium systems with endogenous treatment.
Theoretical framework connecting sublinear graph algorithms to test-time LLM methods like RAG and tool use via prior knowledge.
Theoretical analysis of when Transition Matching outperforms Flow Matching for generative models with fewer sampling steps.
Performance metric for binary classifiers with precision and capacity constraints, extending ROC curve analysis.
Framework using Discrete Empirical Interpolation Method for interpretability in neural ODEs and dynamical system analysis.
Review of neural network approaches for precipitation prediction, comparing traditional NWP with deep learning methods.
Research on jointly learning sequential and relational data for prediction tasks involving entities, integrating sequence and graph modeling.
ML approach to optimize power line de-energization decisions for wildfire risk mitigation by solving mixed-integer linear programs faster for operational power systems.
Research applying model-based reinforcement learning to improve variable selection heuristics in branch-and-bound solvers for mixed-integer linear programming and combinatorial optimization problems.
Invertible flow-based method for learning data-driven manifolds in irregularly-sampled time series classification.
Study on impact of label quality versus model complexity in time series anomaly detection with limited labels.
Adversarial robustness evaluation of CSI-based wireless sensing models for human activity recognition.
Deep RL study for dynamic algorithm configuration using DDQN and PPO on evolutionary algorithm parameter control.
Large-scale evaluation of hybrid quantum-classical autoencoders for unsupervised network intrusion detection.
Unified positional encoding framework using group actions for transformers, unifying rotational and additive approaches.
Causal framework for interpretable and controllable generative models with theoretical guarantees.
Technique addressing output logit divergence instability during LLM pretraining via embedding centering.
One-shot reinforcement learning approach for improving LLM reasoning across multiple domains with minimal data.
Fairness-aware federated learning method with calibrated server updates across demographic groups.
Theoretical framework jointly optimizing source weights and transfer quantities in multi-source transfer learning.
Multigrade deep learning framework for structured error refinement in neural network training.
Pre-trained transformer for brain motor decoding that generalizes across recording sites and subjects.