# Biography

**Hong-Ming Chiu** is a Electrical and Computer Engineering Ph.D. student at the **University of Illinois Urbana-Champaign (UIUC)**, working with
Prof. Richard Y. Zhang. He received his Bachelor's Degree in Electronics Engineering from National Chiao Tung University, Hsinchu,
Taiwan (NCTU). He worked with Prof. Carrson C. Fung and
Prof. Tian-Sheuan Chang during his undergrads. He is a member of
**IEEE-Eta Kappa Nu, Mu Sigma Chapter** and a former member of **IEEE Student Bench at NCTU**. His research interests include
**Machine Learning** and **Optimization**.

# Experiences

# Publications

To certify the robustness of neural networks to adversarial perturbations, most state-of-the-art techniques rely on a linear programming (LP) relaxation of the ReLU activation. Recent results suggest that LP relaxation faces an inherent "convex relaxation barrier". In this paper, we propose a nonconvex relaxation for the ReLU relaxation, based on a low-rank restriction of a semidefinite programming (SDP) relaxation. We show that the nonconvex relaxation has a similar complexity to the LP relaxation, and can almost completely overcome the "convex relaxation barrier" faced by the LP relaxation.

The matrix completion problem seeks to recover a $d \times d$ ground truth matrix of low rank $r \ll d$ from observations of its individual elements. Stochastic gradient descent (SGD) is one of the few algorithms capable of solving matrix completion on a huge scale. Unfortunately, SGD experiences a dramatic slow-down when the underlying ground truth is ill-conditioned; it requires at least $O(\kappa \log(1/\epsilon))$ iterations to get $\epsilon$-close to ground truth matrix with condition number $\kappa$. In this paper, we propose a preconditioned version of SGD that is agnostic to $\kappa$. For a symmetric ground truth and the Root Mean Square Error (RMSE) loss, we prove that the preconditioned SGD converges in $O(\log(1/\epsilon))$ iterations.

A graph learning and augmentation (GLA) technique is proposed herein to solve the received signal power interpolation problem, which is important for preemptive resource allocation in location-aware communications. A graph parameterization results in the proposed GLA interpolator having superior mean-squared error performance and lower computational complexity than the traditional Gaussian process method. Simulation results and analytical complexity analysis are used to prove the efficacy of the GLA interpolator.

Modern convolutional neural network (CNN) models offer significant performance improvement over previous methods, but suffer from high computational complexity and are not able to adapt to different run-time needs. To solve the above problem, this paper proposes an inference-stage pruning method that offers multiple operation points in a single model, which can provide computational power-accuracy modulation during run time. This method can perform on shallow CNN models as well as very deep networks such as Resnet101. Experimental results show that up to 50% savings in the FLOP are available by trading away less than 10% of the top-1 accuracy.

# Projects

We built a simple oscilloscope using Nexys 4 DDR board and PCB. First, the PCB transforms the input voltage signal into a signal with an acceptable voltage range for the FPGA board input, as well as generating knobs' control signal. The FPGA board then takes the processed signal and control signal to display waveforms, change voltage scale, adjust sweep time, etc. This work is the final project of the Digital Laboratory class at NCTU.

This work contains the impelmentation and comparison of two graph learning algorithms, Causal Graph Process (CGP) and Sparse Vector Autoregressive model (SVAR). These two graph learning methods can be used to derive the graph representation among a large number of unstructured time series data, and then make predictions on the future data.

**Hong-Ming Chiu**

Implemented 8-bit Huffman coding algorithm using System Verilog. The system takes an image as an input, the image contains 100 pixels and each pixel value is an integer between 1 to 6 (inclusive). The system then outputs the Huffman Code for each pixel value based on the source probability distribution (more frequent pixel values will have the shorter codewords). This is the final project of the Digital Circuit and Systems class in NCTU.