AI Paper Atlas

Research map / Learning theory and algorithms / Kernels and regression theory

Tighter Risk Certificates for Neural Networks

María Pérez-Ortiz, Omar Rivasplata, John Shawe-Taylor, Csaba Szepesvári

JMLR 2021 · Learning theory and algorithms › Kernels and regression theory · approach: PAC-Bayes bounds · 37 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train with f_quad instead of f_classicPBB FCN, Rand.Init. prior · MNISTrisk certificate (0-1 loss): .3304 -> .3155Improved
train with f_quad instead of f_classicPBB CNN, Learnt prior · MNISTrisk certificate (0-1 loss): .0166 -> .0155Improved
train with f_quad instead of f_classicPBB FCN, Learnt prior · MNISTstochastic predictor test 0-1 error: .0230 -> .0202Improved
train with f_lambda instead of f_quadPBB CNN, Rand.Init. prior · MNISTstochastic predictor test 0-1 error: .0513 -> .0397Improved
train with f_lambda instead of f_quadPBB FCN, Learnt prior · MNISTrisk certificate (0-1 loss): .0279 -> .0354Hurt
train with f_bbb (KL coefficient 0.1) instead of f_quadPBB FCN, Rand.Init. prior · MNISTstochastic predictor test 0-1 error: .0921 -> .0293Mixed
use data-dependent (Learnt) prior instead of data-free random prior (ablation)f_quad FCN · MNISTrisk certificate (0-1 loss): .3155 -> .0279Improved
use data-dependent (Learnt) prior instead of data-free random prior (ablation)f_quad CNN · MNISTstochastic predictor test 0-1 error: .0513 -> .0104Improved
use CNN instead of FCN (ablation)f_quad, Learnt prior · MNISTrisk certificate (0-1 loss): .0279 -> .0155Improved
sample stochastic predictor from posterior instead of prior mean predictor (ablation)f_quad FCN, Rand.Init. prior · MNISTtest 0-1 error: .8792 -> .0921Improved
add KL attenuating coefficient 0.0001 to training objective (ablation)f_quad CNN, Rand.Init. prior · MNISTstochastic predictor test 0-1 error: .0513 -> .0305Mixed
train with PBB f_quad and SGD instead of SGLD (D&R 2018)FCN with data-dependent prior · MNISTrisk certificate (0-1 loss): 0.2100 -> 0.0279Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy