Yile Gu

Yile Gu

Auto-generated profile, based on publicly available information

Yile Gu

Amazon,USA

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 March 2019 in Powder Technology
Verified
Journal Article
Closed access
YJYundi Jiang
Yundi Jiang
JKJari Kolehmainen
Jari Kolehmainen
YGYile Gu
Yile Gu
YKYannis G. Kevrekidis
Yannis G. Kevrekidis
AOAli Ozel
Ali Ozel
SSSankaran Sundaresan
Sankaran Sundaresan

4 December 2018 in Journal of Fluid Mechanics
Verified
Journal Article
Closed access
PDF file
YGYile Gu
Yile Gu
AOAli Ozel
Ali Ozel
JKJari Kolehmainen
Jari Kolehmainen
SSSankaran Sundaresan
Sankaran Sundaresan
Developing constitutive models for particle phase rheology in gas-fluidized suspensions through rigorous statistical mechanical methods is very difficult when complex inter-particle forces are present. In the present study, we pursue a computational approach based on results obtained through Eulerian–Lagrangian simulations of the fluidized state. Simulations were performed in a periodic domain for non-cohesive and mildly cohesive (Geldart Group A) particles. Based on the simulation results, we propose modified closures for pressure, bulk viscosity, shear viscosity and the rate of dissipation of pseudo-thermal energy. For non-cohesive particles, results in the high granular temperature $T$ regime agree well with constitutive expressions afforded by the kinetic theory of granular materials, demonstrating the validity of the methodology. The simulations reveal a low $T$ regime, where the inter-particle collision time is determined by gravitational fall between collisions. Inter-particle cohesion has little effect in the high $T$ regime, but changes the behaviour appreciably in the low $T$ regime. At a given $T$ , a cohesive particle system manifests a lower pressure at low particle volume fractions when compared to non-cohesive systems; at higher volume fractions, the cohesive assemblies attain higher coordination numbers than the non-cohesive systems, and experience greater pressures. Cohesive systems exhibit yield stress, which is weakened by particle agitation, as characterized by $T$ . All these effects are captured through simple modifications to the kinetic theory of granular materials for non-cohesive materials.

16 December 2023
Verified
Conference Paper
Open access
YYyuan yu
yuan yu
CYChao-Han Huck Yang
Chao-Han Huck Yang
JKJari Kolehmainen
Jari Kolehmainen
PSPrashanth G. Shivakumar
Prashanth G. Shivakumar
YGYile Gu
Yile Gu
IBIvan Bulyko
Ivan Bulyko
We propose a neural language modeling system based on low-rank adaptation (LoRA) for speech recognition output rescoring. Although pretrained language models (LMs) like BERT have shown superior performance in second-pass rescoring, the high computational cost of scaling up the pretraining stage and adapting the pretrained models to specific domains limit their practical use in rescoring. Here we present a method based on low-rank decomposition to train a rescoring BERT model and adapt it to new domains using only a fraction (0.08%) of the pretrained parameters. These inserted matrices are optimized through a discriminative training objective along with a correlation-based regularization loss. The proposed low-rank adaptation RescoreBERT (LoRB) architecture is evaluated on LibriSpeech and internal datasets with decreased training times by factors between 5.4 and 3.6.

6 June 2021
Verified
Conference Paper
Open access
LLLinda Liu
Linda Liu
YGYile Gu
Yile Gu
AGAditya Gourav
Aditya Gourav
AGAnkur Gandhe
Ankur Gandhe
SKShashank Kalmane
Shashank Kalmane
IBIvan Bulyko
Ivan Bulyko
As voice assistants become more ubiquitous, they are increasingly expected to support and perform well on a wide variety of use-cases across different domains. We present a domain-aware rescoring framework suitable for achieving domain-adaptation during second-pass rescoring in production settings. In our framework, we fine-tune a domain-general neural language model on several domains, and use an LSTM-based domain classification model to select the appropriate domain-adapted model to use for second-pass rescoring. This domain-aware rescoring improves the word error rate by up to 2.4% and slot word error rate by up to 4.1% on three individual domains – shopping, navigation, and music – compared to domain general rescoring. These improvements are obtained while maintaining accuracy for the general use case.

30 August 2021
Verified
Conference Paper
Closed access
SNScott Novotney
Scott Novotney
YGYile Gu
Yile Gu
IBIvan Bulyko
Ivan Bulyko