Princeton UniversitySept. 2014 - Current
Ph. D. Candidate
Automated neural network architecture synthesis
Deep neural network compression
Reconfigurable neural network and online learning
Advisor: Prof. Niraj K. Jha
Peking UniversitySept. 2010 - Jul. 2014
Bachelor of Science
Work and Research Experience
Facebook AI (Mobile Vision) May 2018 - Current
Princeton University Sept. 2014 - Current
Project: Automated architecture synthesis for CNN, RNN, and LSTM
Train both weights and architecture with gradient descent
Enable neural network to adaptively adjust their structure during training
Generate compact yet accurate neural network in an automated flow
Project: Convolutional neural network compression
Propose an architecture compression algorithm based on a grow-and-prune paradigm
Reduce both memory and computation cost of deep neural networks
Achieve 15.7x (4.6x) parameters (FLOPs) reduction on AlexNet (current state-of-the-art)
Project: Online learning with reconfigurable neural networks
Design reconfigurable neural networks that can grow and prune neurons/connections
Allow online learning systems to adaptively adjust neural network capability
Generate inference models with high accuracy/cost ratio in real-time
Tensilica at Cadence Design System Inc. May 2016 - Aug. 2016
DSP Software Engineer Intern
Simultaneous localization and mapping (SLAM)
Targeted at map establishment and localization problem in robot navigation
Implemented scan-matching, extended Kalman filter, and computer vision based solutions
Built and tested a real SLAM platform with LIDAR and camera in a real indoor environment
Other Computer Skills
- Android development
X. Dai, H. Yin, and N. K. Jha, "Grow and Prune Compact, Fast, and Accurate LSTMs" arXiv prepring arXiv:1805.11797, May 2018.
X. Dai, H. Yin, and N. K. Jha, “Nest: A Neural network synthesis tool based on a grow-and-prune paradigm” arXiv preprint arXiv:1711.02017, Sept. 2017.
X. Dai and N. K. Jha, “Device state library: Mechanism to burst the productivity of TCAD mixed-mode simulation,” Accepted by IEEE Trans. VLSI Syst., May. 2017.
X. Dai and N. K. Jha, “Improving convergence and simulation time of quantum hydrodynamic simulation: Application to extraction of best 10-nm FinFET parameter values” IEEE Trans. VLSI Syst., vol. pp, no. 99, pp. 1–11, May 2016.
A. M. Nia, X. Dai, P. Mittal, and N. K. Jha, ”PinMe: Tracking a smartphone user around the world”, Submitted to IEEE Trans. Multi-scale Computing, Nov. 2016.
J. Wu, D. Hou, Z. Qin, X. Dai, Z. Zhang, and J. Zhao“Erbium-fiber-laser-based direct frequency comb spectroscopy of Rb two-photon transitions,” Optics Lett., Vol. 38, Issue 23, pp. 5028-5031, Sept. 2013.