archive
Every paper Pith has read. Search by title, abstract, or pith.
2685 papers in stat.ML · page 14
-
Local Fisher geometry reveals sensitivity differences missed by activation alignment
Beyond Activation Alignment: The Geometry of Neural Sensitivity
-
New Riemannian score detects local sensitivity gaps beyond global alignment
Beyond Activation Alignment: The Geometry of Neural Sensitivity
-
Gap-aware transformer reduces Alzheimer's prediction error by 13%
Forecasting Medium-Horizon Alzheimer's Disease Progression: Residual Gap-Aware Transformers for 24-Month CDR-SB Change from ADNI Clinical and Biomarker Histories
-
Framework lets users set confidence in Bayesian uncertainty sources
Bayesian inference with sources of uncertainty: from confidence modelling to sparse estimation
-
One local tensor triples as KL term
From Information Geometry to Jet Substructure: A Triality of Cumulant Tensors, Energy Correlators, and Hypergraphs
-
Vine copulas spot higher-order time-varying links in neural data
Dynamic Vine Copulas: Detecting and Quantifying Time-Varying Higher-Order Interactions
-
Optimized sampling and surrogate cut leading error in value estimates
First-Order Efficiency for Probabilistic Value Estimation via A Statistical Viewpoint
-
Bayesian reflex unifies online AI learning like autonomic nervous system
The Bayesian Reflex: Online Learning as the Autonomic Nervous System of Modern and Future AI
-
Divergence measures derive training for regression to diffusion
Information Theory and Statistical Learning
-
Neural networks converge to Gaussians with explicit Wasserstein bounds
Universality in Deep Neural Networks: An approach via the Lindeberg exchange principle
-
Per-sample clipping yields optimal SGD rates under heavy tails
Robust and Fast Training via Per-Sample Clipping
-
Fréchet algorithm gives consistent random effects for metric-space objects
Random-Effects Algorithm for Random Objects in Metric Spaces
-
ParaRNN adds additive structure to RNNs for interpretability and bounds
ParaRNN: An Interpretable and Parallelizable Recurrent Neural Network for Time-Dependent Data
-
Online DEM tracks hidden states in chaotic systems
Online Generalised Predictive Coding
-
One label flip shifts DPO gradient independently of parameters
Efficient Preference Poisoning Attack on Offline RLHF
-
POO optimizes noisy functions with unknown smoothness within sqrt(log n) of optimal
Black-box optimization of noisy functions with unknown smoothness
-
RL routes parcels by modeling logistics as goal-conditioned MDPs
Middle-mile logistics through the lens of goal-conditioned reinforcement learning
-
Algorithm adapts to unknown ranks across simultaneous matrix completions
Active multiple matrix completion with adaptive confidence sets
-
Neural nets detect GR deviations 33x better using response functions
Testing General Relativity Through Gravitational Wave Classification: A Convolutional Neural Network Framework
-
This paper generalizes distributional alignment games to remove systematic bias from…
Generalized Distributional Alignment Games for Unbiased Answer-Level Fine-Tuning
-
The paper builds a dataset of 150 homicide cases annotated with 17 legal concepts and…
Can Causal Discovery Algorithms Help in Generating Legal Arguments?
-
Kernel embeddings compare conditional distributions with multiple metrics
Measuring Differences between Conditional Distributions using Kernel Embeddings
-
Doubly stable selection guards features against design noise
2D Stability Selection: Design Jittering for Doubly Stable Feature Selection
-
Curvature penalties smooth KAN activations without accuracy loss
KANs need curvature: penalties for compositional smoothness
-
Causal models need constant bits for observations but quadratic for interventions
The Causal Description Gap: Information-Theoretic Separations Across Pearl's Hierarchy
-
The paper gives matching upper and lower bounds on the number of samples needed for…
On the Optimal Sample Complexity of Offline Multi-Armed Bandits with KL Regularization
-
Flexible per-class thresholds improve graph-based classifiers
Large margin classifier with graph-based adaptive regularization
-
Mira score ranks conditional distributions by fidelity to joint samples
MIRA: A Score for Conditional Distribution Accuracy and Model Comparison
-
Extreme value theory enables extrapolation beyond training data
Extrapolation in Statistical Learning with Extreme Value Theory
-
Estimator recovers latent clusters and matches oracle rates
Adaptive Estimation and Inference in Semi-parametric Heterogeneous Clustered Multitask Learning via Neyman Orthogonality
-
Stable blankets extend to hidden variables and cycles
Stable Blanket with Hidden Variables and Cycles
-
Diffusion model shapes rewards to schedule AI workloads in data centers
Joint Energy Management and Coordinated AIGC Workload Scheduling for Distributed Data Centers: A Diffusion-Aided Reward Shaping Approach
-
Semi-supervised kernel test uses covariates for higher power
A Semi-Supervised Kernel Two-Sample Test
-
Generative models reconstruct full outcome distributions under interventions
Distributional Causal Mediation via Conditional Generative Modeling
-
Auditing tool finds 25% reasoning gaps in DTI models despite matched accuracy
ISAAC: Auditing Causal Reasoning in Deep Models for Drug-Target Interaction
-
Bounded loss certifies GFlowNet fidelity to target rewards
Stable GFlowNets with Probabilistic Guarantees
-
Bayesian model imputes missing data with consistent posterior uncertainty
Missingness-aware Data Imputation via AI-powered Bayesian Generative Modeling
-
Learned per-edge trust lets causal discovery use imperfect priors safely
PRCD-MAP: Learning How Much to Trust Imperfect Priors in Causal Discovery
-
The paper proves that scale-invariant upper bounds on self-normalized martingales exist…
Self-Normalized Martingales and Uniform Regret Bounds for Linear Regression
-
Graph choice and distance metric shape persistent-homology features for time series
Persistent Homology of Time Series through Complex Networks
-
Rule-based selection picks wrong models in time series forecasts
Why Model Selection Fails in Time Series Forecasting: An Empirical Study of Instability Across Data Regimes
-
Anisotropic pre-distortion stabilizes private LASSO
Stabilizing Private LASSO under Heterogeneous Covariates via Anisotropic Objective Perturbation
-
Random walks let LLMs estimate properties on million-node graphs
Evaluating LLMs on Large-Scale Graph Property Estimation via Random Walks
-
SMO for MAPE-loss SVR needs only bound and feasibility changes
Sequential Minimal Optimization for $\varepsilon$-SVR with MAPE Loss and Sample-Dependent Box Constraints
-
SMO for MAPE ε-SVR changes only feasibility sets and clipping bounds
Sequential Minimal Optimization for $\varepsilon$-SVR with MAPE Loss and Sample-Dependent Box Constraints
-
Truncation bias sets floor for high-dimensional mean testing
Mean Testing under Truncation beyond Gaussian
-
Randomized tests plus simulators fix bias in model logs
The Partial Testimony of Logs: Evaluation of Language Model Generation under Confounded Model Choice
-
Deep net saddle escape time set by bottleneck layer count
A Theory of Saddle Escape in Deep Nonlinear Networks
-
Norm imbalance identity sets deep-net escape time by bottleneck depth
A Theory of Saddle Escape in Deep Nonlinear Networks
-
SGD generalizes via fast signal drift and slow noise diffusion in NTK
A Theory of Generalization in Deep Learning