AI Paper Atlas

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

Kernel Approximation Methods for Speech Recognition

Avner May, Alireza Bagheri Garakani, Zhiyun Lu, Dong Guo, Kuan Liu, Aurélien Bellet, Linxi Fan, Michael Collins, Daniel Hsu, Brian Kingsbury, Michael Picheny, Fei Sha

JMLR 2019 · Learning theory and algorithms › Kernels and regression theory · approach: Random feature approximation · 56 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
add a linear bottleneck (ablation)Sparse Gaussian kernel model · Bengali devtoken error rate (%): 73.0 -> 71.5Improved
add ERLL-based learning rate decay together with bottleneck (ablation)Sparse Gaussian kernel model · Bengali devtoken error rate (%): 71.5 -> 70.9Improved
add ERLL-based learning rate decay alone (ablation)Sparse Gaussian kernel model · Bengali devtoken error rate (%): 73.0 -> 73.0No clear effect
add random feature selection (ablation)Laplacian kernel model · Cantonese devtoken error rate (%): 69.9 -> 68.4Improved
add random feature selection (ablation)Sparse Gaussian kernel model · Bengali devtoken error rate (%): 73.0 -> 72.9Improved
add random feature selection (ablation)Gaussian kernel model · Bengali devtoken error rate (%): 72.6 -> 74.1Hurt
add linear bottleneck and ERLL decay (ablation)DNN with 4000 hidden units · Bengali devtoken error rate (%): 71.1 -> 70.2Improved
add ERLL-based learning rate decay (ablation)DNN with 2000 hidden units · TIMIT devtoken error rate (%): 19.0 -> 19.2Hurt
use heldout ERLL instead of cross-entropy for stopping/decay (ablation)kernel and DNN acoustic models · correlation with TER on Bengalitoken error rate (%): 0.31 -> 0.91Improved
use heldout ERLL instead of cross-entropy for stopping/decay (ablation)kernel and DNN acoustic models · correlation with TER on TIMITtoken error rate (%): -0.21 -> 0.95Improved
increase number of random features from 5000 to 100000 (ablation)kernel models · BN-50cross-entropy and TERImproved
compare best kernel model to best DNNkernel vs DNN · TIMIT testtoken error rate (%): 20.5 -> 20.4Mixed

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