Software Engineer
Google, Mountain View
Hongxiang Gu is currently a software engineer at Google, Mountain View.
Prior to his career at Google, Hongxiang received his Ph.D. degree fromthe computer science department at University of California, Los Angeles (UCLA), advised by Professor Miodrag Potkonjak. He also earned his B.S. and M.S. in computer science at UCLA in 2014 and 2018.
His research mainly focuses on hardware security, IoT, and machine learning. He spent three summers working at Adobe Research conducting research in the domain of computer vision with a focus on video understanding and summarization.
Google, Mountain View
Adobe Research, San Jose
Adobe Research, San Jose
Adobe Research, San Jose
Doctor of Philosophy in Computer Science
University of California, Los Angeles, Department of Computer Science
Master of Science in Computer Science
University of California, Los Angeles, Department of Computer Science
Bachelor of Science in Computer Science
University of California, Los Angeles
The Internet of Things (IoT) technology is everywhere around us, from Nest to Alexa, from smart home to smart city. However, both individuals or organizations are vulnerable to breach through IoT devices, thus security, especially low power security applications are essential to protecting the security of IoT devices. Hongxiang has conducted extensive research in systems in physically unclonable functions (PUF) and has published numerous papers in premier conferences. He is specifically interested in applying machine learning attacks on security systems in IoT as well as utilizing machine learning algorithms to optimize and improve IoT security systems.
Apart from security, Hongxiang also has hands on experience in research in multimedia applications and deep neural networks during his multiple intership at Adobe Research. He is interested in analyzing and understanding video-based multimedia contents.
Proof-of-work is a very important measure in blockchain technology. However, current proof-of-work implementations are very expensive and power hungry, wasting much energy and computational resources in vain. We present a proof-of-ownership method to be used in a private blockchain setup that utilize execution-simulation-gap of some dedicated hardware so that the blockchain is still protected by computation, but in a much more efficient way.
As neural network models are being used in a wide range of applications, model sharing is becoming more needed than ever. However model sharing also put the intellectual property and the training effort in danger. In order to prove that a specific model is trainined by a specific individual using a specific design on a specific training dataset, we propose to use dead Relu neurons as a robust watermark in neural models. Our method is robust against network pruning, model compression, fine-tuning and parameter tuning.
Secure group-oriented communication is crucial to a wide range of applications in Internet of Things (IoT). Security problems related to group-oriented communications in IoTbased applications placed in a privacy-sensitive environment have become a major concern along with the development of the technology. Unfortunately, many IoT devices are designed to be portable and light-weight; thus, their functionalities, including security modules, are heavily constrained by the limited energy resources (e.g., battery capacity). To address these problems, we propose a group key management scheme based on a novel PUF design: multistage interconnected physically unclonable function (MIPUF) to secure group communications in an energy-constrained environment. Our design is capable of performing key management tasks such as key distribution, key storage and rekeying securely and efficiently. We show that our design is secure against multiple attack methods and our experimental results show that our design saves 47.33% of global energy comparing to state-of-the-art Elliptic-curve cryptography (ECC)-based key management scheme on average.
Physical Unclonable Functions (PUFs) are known for their unclonability and light-weight design. Recent advancement in technology has significantly compromised the security of PUFs. Machine learning-based attacks have been proven to be able to construct numerical models that predict various types of PUFs with high accuracy with a small set of challenge-response pairs (CRPs). To address the problem, we present a reconfigurable interconnected PUF network (IPN) design that significantly strengthens the security and unclonability of strong PUFs. While the IPN structure itself provides high resilience against modeling attacks, the reconfiguration mechanism remaps the input-output mapping before an attacker could collect sucient CRPs. Experimental results show that all tested state-of-the-art machine learning attack methods have prediction accuracy of around 50% on a single bit output of a reconfigurable IPN.
Video summaries come in many forms, from traditional single-image thumbnails, animated thumbnails, storyboards, to trailer-like video summaries. Content creators use the summaries to display the most attractive portion of their videos; the users use them to quickly evaluate if a video is worth watching. All forms of summaries are essential to video viewers, content creators, and advertisers. Often video content management systems have to generate multiple versions of summaries that vary in duration and presentational forms. We present a framework ReconstSum that utilizes LSTM-based autoencoder architecture to extract and select a sparse subset of video frames or keyshots that optimally represent the input video in an unsupervised manner. The encoder selects a subset from the input video while the decoder seeks to reconstruct the video from the selection. The goal is to minimize the difference between the original input video and the reconstructed video. Our method is easily extendable to generate a varietyof applications including static video thumbnails, animated thumbnails, storyboards and "trailer-like" highlights. We specifically study and evaluate two most popular use cases: thumbnail generation and storyboard generation. We demonstrate that our methods generate better results than the state-of-the-art techniques in both use cases.
A list of my publications during my Ph.D career is listed here, you can directly download the paper by clicking the download button. Please feel free to contact me if you have any questions regarding the paper.
Secure group-oriented communication is crucial to a wide range of applications in Internet of Things (IoT). Security problems related to group-oriented communications in IoT-based applications placed in a privacy-sensitive environment have become a major concern along with the development of the technology. Unfortunately, many IoT devices are designed to be portable and light-weight; thus, their functionalities, including security modules, are heavily constrained by the limited energy resources (e.g., battery capacity). To address these problems, we propose a group key management scheme based on a novel PUF design: multistage interconnected physically unclonable function (MIPUF) to secure group communications in an energy-constrained environment. Our design is capable of performing key management tasks such as key distribution, key storage and rekeying securely and efficiently. We show that our design is secure against multiple attack methods and our experimental results show that our design saves 47.33% of global energy comparing to state-of-the-art Elliptic-curve cryptography (ECC)-based key management scheme on average.
Today’s mobile devices are equipped with cameras capable of taking very high-resolution pictures. For computer vision tasks which require relatively low resolution, such as image classification, sub-sampling is desired to reduce the unnecessary power consumption of the image sensor. In this paper, we study the relationship between subsampling and the performance degradation of image classifiers that are based on deep neural networks (DNNs). We empirically show that subsampling with the same step size leads to very similar accuracy changes for different classifiers. In particular, we could achieve over 15× energy savings just by subsampling while suffering almost no accuracy lost. For even better energy accuracy trade-offs, we propose AdaSkip, where the row sampling resolution is adaptively changed based on the image gradient. We implement AdaSkip on an FPGA and report its energy consumption.
Video summaries come in many forms, from traditional single-image thumbnails, animated thumbnails, storyboards, to trailer-like video summaries. Content creators use the summaries to display the most attractive portion of their videos; the users use them to quickly evaluate if a video is worth watching. All forms of summaries are essential to video viewers, content creators, and advertisers. Often video content management systems have to generate multiple versions of summaries that vary in duration and presentational forms. We present a framework ReconstSum that utilizes LSTM-based autoencoder architecture to extract and select a sparse subset of video frames or keyshots that optimally represent the input video in an unsupervised manner. The encoder selects a subset from the input video while the decoder seeks to reconstruct the video from the selection. The goal is to minimize the difference between the original input video and the reconstructed video. Our method is easily extendable to generate a variety of applications including static video thumbnails, animated thumbnails, storyboards and ”trailer-like” highlights. We specifically study and evaluate two most popular use cases: thumbnail generation and storyboard generation. We demonstrate that our methods generate better results than the state-of-the-art techniques in both use cases.
Physical Unclonable Functions (PUFs) are known for their unclonability and light-weight design. Recent advancement in technology has significantly compromised the security of PUFs. Machine learning-based attacks have been proven to be able to construct numerical models that predict various types of PUFs with high accuracy with a small set of challenge-response pairs (CRPs). To address the problem, we present a reconfigurable interconnected PUF network (IPN) design that significantly strengthens the security and unclonability of strong PUFs. While the IPN structure itself provides high resilience against modeling attacks, the reconfiguration mechanism remaps the input-output mapping before an attacker could collect su cient CRPs. Experimental results show that all tested state-of-the-art machine learning attack methods have prediction accuracy of around 50% on a single bit output of a reconfigurable IPN.
Physical unclonable functions (PUFs) take advantage of the effect of process variation on hardware to obtain their unclonability. Traditional PUF design only focuses on the analog signals of circuits. An arbiter PUF, for example, generates responses by racing delay signals. Implementations of such PUFs usually employ large area and power consumption while providing very low throughput. To address this problem, we propose an energy efficient PUF design in such a way that it races analog signals and computes digital logic simultaneously. More importantly, the analog portion of the circuit (racing) shares a large amount of hardware resources with the digital portion of the circuit (computing) by introducing only small overhead in terms of area and power. Our test results on Spartan-6 field-programmable gate array (FPGA) platforms indicate that by combining the two outputs, our design enables much larger PUF output throughput, better randomness and less power consumption compared to traditional PUFs.
Physical unclonable functions (PUFs) take advantage of the effect of process variation on hardware to obtain their unclonability. Traditional PUF design only focuses on the analog signals of circuits. An arbiter PUF, for example, generates responses by racing delay signals. Implementations of such PUFs usually employ large area and power consumption while providing very low throughput. To address this problem, we propose an energy efficient PUF design in such a way that it races analog signals and computes digital logic simultaneously. More importantly, the analog portion of the circuit (racing) shares a large amount of hardware resources with the digital portion of the circuit (computing) by introducing only small overhead in terms of area and power. Our test results on Spartan-6 field-programmable gate array (FPGA) platforms indicate that by combining the two outputs, our design enables much larger PUF output throughput, better randomness and less power consumption compared to traditional PUFs.
We have proposed a new security platform: physical unclonable function (PUF) matching using programmable delay lines (PDL). Our platform inherits good security properties of standard PUFs, such as low energy, low delay, and unclonability. However, standard PUF-based security protocols induce high computational resources of at least one involved party. To resolve this issue, we take advantage of PDL technology to match standard PUFs in such a way that two PUFs have the same challenge response mapping function. The matched pair of PUFs enables a majority of protocols to be executed in an ultra low energy, low latency manner for all the involved parties.
Data security and privacy have emerged to become an important issue in various types of applications. Although many cryptographic cyphers are proposed to leverage the issue, they normally suffer from the problems of either requiring large power/bandwidth consumption or employing linear system which is easy to break. To solve the problem of traditional cyphers, we have proposed a new hardware security primitive: recursive inverse function (RIF) designed on the field-programmable gate array (FPGA). The RIF takes advantage of a pair of inverse functions, building a recursive scheme for message encryption and decryption. The inverse functions are defined as a pair of functions where each function implements a mapping being inverse to the mapping of the other function. On the top of it, the recursive structure guarantees the input-output mapping to be statistically extremely hard to predict. The RIF can be easily implemented using hierarchical lookup-table (LUT) structures with low delay and power overhead. Using our proposed RIF structure, we have demonstrated how the RIF can be incorporated into a processor design to enable the data protection. Finally, we implement our scheme on a Xilinx Spartan-6 FPGA device to analyze the performance and the overhead.
I now work as a software engineer at Google. I do spend my spare time working on research projects. Prior to my current position, I spent most of my summers doing internship in industrial research labs. During my undergraduate years, however, my internships were mostly software engineering positions. After graduation from college, I took a few months off co-founding a small startup.
Software Engineer
Research Scientist Intern. Understanding, editing and summarizing online videos with deep reinforcement learning.
Data Science Research Intern. Unsupervised video summarization using generative deep neural networks.
Research Scientist Intern. Secure video streaming in Trusted Execution Environment (TEE) for Ultra High Definition (UHD) videos.
Co-founder & backend engineer. Design & Develop an iOS app to create Interactive images.
Software Development Engineer Intern, Search team. Hyperlocal topic trend detection system.
SQA Intern. Infrastructural security modules within the company.
I enjoy traveling around the world exploring different cultures and histories. I have visited more than 20 countries and I am trying hard to extend my list of visited places every year. I love taking photos whereever I go, you can find some of photos below.
I would be happy to talk to you through email or in person if you are interested in my research, teaching or life.
I now live in Sunnyvale, CA with my two cats: Mio and Miso.
Please contact me via email if you are interested in talking about research, entrepreneurship and more.