Name
Abstract | ||
|---|---|---|
Replacing hand-engineered pipelines with end-to-end deep learning systems has enabled strong results in applications like speech and object recognition. However, the causality and latency constraints of production systems put end-to-end speech models back into the underfitting regime and expose biases in the model that we show cannot be overcome by scaling up, i.e., training bigger models on more data. In this work we systematically identify and address sources of bias, reducing error rates by up to 20% while remaining practical for deployment. We achieve this by utilizing improved neural architectures for streaming inference, solving optimization issues, and employing strategies that increase audio and label modelling versatility. |
Year | Venue | Field |
|---|---|---|
2017 | arXiv: Computation and Language | Causality,Pipeline transport,Software deployment,Inference,Latency (engineering),Computer science,Artificial intelligence,Deep learning,Scaling,Machine learning,Cognitive neuroscience of visual object recognition |
DocType | Volume | Citations |
Journal | abs/1705.04400 | 3 |
PageRank | References | Authors |
0.41 | 9 | 16 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| eric battenberg | 1 | 312 | 12.30 |
| Rewon Child | 2 | 38 | 3.79 |
| Adam Coates | 3 | 2493 | 160.95 |
| christopher fougner | 4 | 266 | 9.52 |
| Yashesh Gaur | 5 | 15 | 9.06 |
| Jiaji Huang | 6 | 3 | 2.77 |
| Heewoo Jun | 7 | 11 | 1.53 |
| Ajay Kannan | 8 | 30 | 3.31 |
| Markus Kliegl | 9 | 3 | 0.41 |
| Atul Kumar | 10 | 7 | 3.12 |
| Hairong Liu | 11 | 374 | 17.41 |
| Vinay Rao | 12 | 8 | 1.50 |
| Sanjeev Satheesh | 13 | 5591 | 233.55 |
| david seetapun | 14 | 274 | 9.42 |
| Anuroop Sriram | 15 | 39 | 3.10 |
| Zhenyao Zhu | 16 | 567 | 26.75 |