Stat.ML Papers@StatMLPapers

Unofficial updates of statistical machine learning papers on arXiv

arxiv.org/list/stat.ML/recen…
Joined June 2012
Why Constrained Updates Still Forget: A Functional-Path Analysis of Continual SFT ift.tt/rWQdpKI

Why Constrained Updates Still Forget: A Functional-Path Analysis...

Supervised fine-tuning adapts language models to specialized tasks but can disrupt previously acquired behavior. Existing approaches constrain parameters, gradients, or replay objectives, yet the...

arxiv.org
2
1,534
Differentially Private Synthetic Data via Generalized Moment Matching ift.tt/cmfJqLC

Differentially Private Synthetic Data via Generalized Moment Matching

We study $(\varepsilon,δ)$-differentially private synthetic data for arbitrary databases of $n$ records in $[-1,1]^d$ under a specified integral probability metric (IPM). We introduce...

arxiv.org
2
970
High-Value Training Data from Noisy, Diverse Sources ift.tt/qFWn8ul

High-Value Training Data from Noisy, Diverse Sources

Training state-of-the-art models increasingly requires learning from large, heterogeneous datasets whose examples vary in quality, domain relevance, and source of supervision....

arxiv.org
1
5
44
5,001
Flow-based Spectral Kernel Learning for Nonstationary Attention ift.tt/ifBErRk

Flow-based Spectral Kernel Learning for Nonstationary Attention

Kernelized attention provides a useful perspective for understanding self-attention as a similarity-based aggregation mechanism, but existing approaches often rely on fixed or stationary kernels,...

arxiv.org
10
991
xVGAE: A Hierarchical Variational Graph Autoencoder for Exchangeable Graphs ift.tt/cO2jG6B

xVGAE: A Hierarchical Variational Graph Autoencoder for Exchangeable Graphs

Graphs play a crucial role in many applications, ranging from drug discovery to social network modeling and astronomy. Generative models for graphs have substantially advanced in recent years, but...

arxiv.org
5
999
Continuous-depth Deep Gaussian Processes ift.tt/oKBqrzw

Continuous-depth Deep Gaussian Processes

Deep Gaussian processes (DGPs) are appealing Bayesian regression models, but their standard discrete-depth parameterization often becomes harder to optimize as depth grows. We argue that depth is...

arxiv.org
14
1,724
Random Projections for Spectral Algorithms: Optimal Rates in Sobolev Norms ift.tt/wF7Sotp

Random Projections for Spectral Algorithms: Optimal Rates in Sobolev Norms

Random projections (RP) offer an effective method for reducing computational and storage overhead while preserving the geometric structure of data. However, existing studies have primarily focused...

arxiv.org
2
17
1,339
Prediction Uncertainty Controlled Computer Model Calibration ift.tt/xzTSZFG

Prediction Uncertainty Controlled Computer Model Calibration

Computer-model calibration is usually undertaken in service of prediction, yet neither Bayesian Kennedy-O'Hagan (KOH) calibration nor amortized simulation-based inference is built to control the...

arxiv.org
905
Deep Learning for Random Objects in Metric Spaces: Beyond Manifold Assumptions ift.tt/oMPYLTr

Deep Learning for Random Objects in Metric Spaces: Beyond Manifold...

Deep learning (DL) methods have substantially improved predictive performance across a wide range of downstream prediction tasks, enabling new discoveries in multiple domains. However, principled...

arxiv.org
9
66
4,033
Metric Conformal Prediction via the Conditional Expected Local Radius ift.tt/miDHRkK

Metric Conformal Prediction via the Conditional Expected Local Radius

We introduce the conditional expected local radius (CELR) and develop a new conformal prediction method for non-Euclidean responses in general metric spaces. CELR measures local conditional...

arxiv.org
3
898
Dimension-free regimes for unadjusted Langevin in ridge-separable models: A subspace robust Wasserstein analysis ift.tt/1nRSF6s

Dimension-free regimes for unadjusted Langevin in ridge-separable...

The Unadjusted Langevin Algorithm (ULA) is widely used for high-dimensional Bayesian inference. Classical analyses in Wasserstein-2 distance predict an asymptotic bias and a mixing time that grow...

arxiv.org
1
912
Nonparametric autoregressive estimation of probability distributions with parallel gradient boosting ift.tt/mYa205O

Nonparametric autoregressive estimation of probability...

Nonparametric estimation of joint probability distributions is a core problem in statistics, which is notoriously challenging with high-dimensional data. Such challenges also arise with...

arxiv.org
1
892
Reliable AI-Assisted Decisions via Selective Auditing ift.tt/bpxGQuE

Reliable AI-Assisted Decisions via Selective Auditing

We consider online monitoring of an AI system that produces a sequence of decisions. At each round, the system outputs a proposed decision together with a confidence score estimating the...

arxiv.org
1
928
AI Mindmeld: A Latent Interconnect Protocol ift.tt/AvbF8nI

AI Mindmeld: A Latent Interconnect Protocol

LLMs can act as scientific agents using tools to consult specialists and run experiments. Yet specialist information must enter their context, generally as text, images, audio or video: a...

arxiv.org
4
923
When Reconstruction Is Not Enough: Doubly Robust Counterfactual and Interventional Mediation Analysis with Learned Multimodal Representations ift.tt/bf6x0iQ

When Reconstruction Is Not Enough: Doubly Robust Counterfactual...

Learned representations of high-dimensional multimodal health records are increasingly used to adjust for confounding, yet they are usually trained to reconstruct the data rather than preserve the...

arxiv.org
893
Lifetime-valid decisions by continually evolving AI analysts ift.tt/vIDuQa4

Lifetime-valid decisions by continually evolving AI analysts

AI agents increasingly act as analysts whose decision can often be cast as hypothesis tests. Classical tests control the Type I error rate, the probability of rejecting a true null hypothesis, for...

arxiv.org
864
Residual Calibration via Local Feature-Space Refinement ift.tt/13QedrI

Residual Calibration via Local Feature-Space Refinement

Deep neural networks often struggle to provide reliable confidence estimates, making post-hoc calibration essential for reliable decision-making. Existing post-hoc methods largely rely on...

arxiv.org
1
13
1,270
Width and Depth Limits Commute During Training ift.tt/eKmzaYc

Width and Depth Limits Commute During Training

Recent work established that the infinite-width and infinite-depth limits of residual networks commute at initialization when residual branches are scaled by $1/\sqrt{depth}$. In particular, the...

arxiv.org
2
18
2,244
Transformers with CoT Discover Heavy-Ball Acceleration Through Training ift.tt/F9PoTQ4

Transformers with CoT Discover Heavy-Ball Acceleration Through Training

Modern transformers often solve complex tasks by generating and refining intermediate reasoning steps. To develop a theoretical understanding of chain-of-thought (CoT) reasoning, a notable recent...

arxiv.org
4
1,466
What Comes Next and Why: Interpretable Next-Event Prediction with Neuro-Symbolic Rules ift.tt/OYDto8e

What Comes Next and Why: Interpretable Next-Event Prediction with...

In real-world applications, being able to explain why a prediction is made is often as important - if not more so - than achieving high accuracy. However, existing methods for sequential event...

arxiv.org
2
13
1,297
Load more