Lai, Lifeng

UC Davis
UC Davis Electrical and Computer Engineering logo
Menu

Advancement to Professor Step 6 Review

Dear Esteemed Colleague,

Thank you for considering a reference letter for my advancement to Professor Step 6 review at UC Davis. I appreciate your time and support. I hope that you will find the following material useful in writing your letter. In my department, recommendations are anonymous, so I will never know who you are, but if there’s additional information you may find useful in writing your letter, please contact my department chair Prof Saif Islam.

At UC Davis, in line with most other universities, promotion to associate and full professor is a barrier step and requires external letters. Different from other universities, University of California operates “UC Step System” to review faculty members before and after tenure for promotion and advancement. In particular, advancement to Professor Step 6 is also a barrier step. UC Davis will evaluate me primarily on my record since my promotion to full professor 6 years ago.

Background 

I joined ECE at UC Davis as an Associate Professor in 2016 and was promoted to Professor in July 2020. For advancement to the Professor Step 6 (the case at hand), my institution considers the candidate’s overall record in research, service and teaching since promotion to the Professor rank. For my current review case, that would be July 2019 (since activities up to July 2019 were considered for my promotion to the professor rank that took effect in 2020) to present date. Here are my Curriculum Vitae and google scholar link.

Research 

My main research focus is on the design and analysis of secure and efficient communication and learning systems, using tools from information theory, signal processing and statistics. In the past, my research group mainly focused on information theoretic security and sequential analysis. In the last few years, my research group has also gradually expanded to other research topics such as trustworthy machine learning, robust and communication efficient federated learning, and theoretical understanding of machine learning algorithms.

Funding summary: Since July 2019 (this review period), a total of over $6 M grants were active (over $4 M were newly awarded during this review period), among which I served as PI for over $3 M of them.

Publication summary: In this review period, my group published 26 journal papers, all of them were published in top journals in the field (TIT, TSP, JSAC, JMLR and TMLR). In addition, my group published 7 papers in the most prestigious machine learning conferences (ICML, NeurIPS and ICLR) and several other conference papers in ISIT, ICASSP etc.

In the following, I only provide a brief summary of my group’s work in new directions initiated after I became a Professor (Please see my CV for my group’s work on other topics).

1. Trustworthy Machine Learning

Machine learning algorithms are increasingly woven into the fabric of our daily lives, influencing everything from the content we consume and the services we utilize to more sensitive areas like finance and healthcare. This widespread use makes the security of these algorithms paramount for trustworthiness. Securing ML involves protecting the algorithms, their training data, and their outputs against threats like adversarial attacks, data poisoning, model theft, and system vulnerabilities. Developing trustworthy algorithms is essential to ensure the reliability, integrity, and safety of these increasingly influential technologies. In this important domain, my group has carried out the following research directions.

  • Trustworthy Sequential Decision Systems:  Sequential decision systems such as reinforcement learning (RL) have been applied in many safety critical and security related applications such as autonomous driving, finance and business management etc. In these critical systems, some questions naturally arise: 1) Should we trust decisions made by RL agents? 2) Can an adversary mislead RL agents? 3) How can we design RL algorithms that are robust to adversarial attacks? While there are many existing work addressing adversarial attacks on supervised learning models, the understandings of vulnerabilities of RL models and their corresponding mitigation strategies are less complete, partially due to the significant differences between online RL and supervised learning. In particular, compared with the supervised learning setting, the design and analysis of attack/defense mechanisms for RL models have to handle the challenges such as long-term rewards, no access to future data and unknown dynamics etc. Our group has been systematically investigating potential vulnerabilities of RL models and algorithms and developed robust RL algorithms that can mitigate impacts of adversarial attacks. In particular, we have: 1) investigated novel action manipulation attacks and defense for stochastic bandits (TSP’20a); 2) Designed provably efficient black-box attacks again single user RL (NeurIPS’21); 3) Investigated vulnerability of multi-agent RL (NeurIPS’23); and 4) Designed novel robust RL against policy execution uncertainty (TMLR’24).
  • Trustworthy Statistical Inference: The goal of this direction is to investigate the fundamental limits and design efficient algorithms for adversarially robust statistical inference. As machine learning and statistical inference algorithms are being increasingly used in safety critical applications and security related applications, there is a pressing need to study the robustness of machine learning and statistical inference algorithms in adversarial environments. In many recent data analytical applications including safety critical applications, we are facing severe situations where an adversary can observe the whole data and then devise its attack vector to modify all entries in the data point. The existence of such powerful adversaries calls for new models and methodologies for adversarially robust inference. To address this need, my research group has carried out systematic study on adversarially robust inference in the presence of powerful adversaries. We aim to answer the following questions: 1) What is the attacker’s optimal attack strategy in choosing the attack vectors?; 2) What are the impacts of these attacks?; and 3) How shall we design inference algorithms to minimize the impact? Our group has made significant progresses in addressing these pressing questions: 1) We characterized adversarial robustness of subspace learning (TSP’20b); 2) Developed a general methodology to measure adversarial robustness of estimators (TIT’20a); 3) Developed robust Lasso based feature selection method (TSP’21); and 4) Designed fairness aware regression robust to adversarial attacks (TSP’23).

2. Robust and Communication Efficient Federated Learning

Due to the grow of modern dataset size and the desire to harness computing power of multiple machines, there is a recent surge of interest in the design of distributed machine learning algorithms such as federated learning. There are many algorithms in this research area. First order methods use gradient information for update in each iteration. Second order methods employ second order information to harness the computer power of each worker and reduce the cost of communication. Zeroth order methods use estimated gradient information to solve the problem where the gradient is not available. However, since distributed algorithms require communication between workers and server, they face several challenges:  1) They are sensitive to Byzantine attackers who can send falsified data to prevent the convergence of algorithms or lead the algorithms to converge to value of the attackers’ choice; 2) The communication overhead could be significant. My group has been investigating how to design robust and communication efficient distributed learning algorithms. In particular, we have: 1) designed first-order based federated learning algorithms that are robust to an arbitrary number of Byzantine attackers (TSP’19); 2) designed robust and efficient second order based  method (TSP’20c); 3) characterized the impact of network topology and communication delays on the convergence of distributed dual coordinate ascent algorithms (JSAC’21).

3. Fundamental Understanding of Machine Learning Algorithms

Given the increasing impact of machine learning algorithms on our lives, a fundamental understanding of these algorithms is crucial. In this research thrust, our group aims to understand the fundamental limits of machine learning algorithms using tools from Information Theory, optimization and Statistics.

  • k Nearest Neighbor (kNN) Based Algorithms: kNN methods are a class of nonparametric statistical methods that are widely used in a large variety of statistical problems, including functional estimation (such as mutual information, KL divergence etc) and machine learning problems. Despite the widespread use of kNN type methods and many existing investigations of the properties, several theoretical problems still need further study. In particular, 1) The theoretical convergence rates of kNN methods for functional estimations, classification and regression etc are still not fully established; 2) For many applications of practical interests, it is not clear under what conditions this type of methods are optimal; 3) Most of the existing analysis of kNN methods rely on availability of independent and identically distributed (i.i.d) training data, while in certain applications (such as applications involving Markov chains) the available data are dependent; 4) While there are many applications and analysis of kNN methods for supervised learning, the applications and analysis of kNN methods for reinforcement learning etc are limited. Providing a theoretical framework of these methods is a fundamental and important task, which ensures a formal guarantee of these methods, often in terms of convergence rates. Furthermore, by addressing the main sources of estimation errors identified by these investigations, we can design improved kNN methods that have better convergence rates. My group has made significant contributions in tackling these challenges. In particular, we have: 1) characterized convergence rate of kNN based mutual information estimator (TIT’20b); 2)  Established minimax optimality of kNN based KL divergence estimator (TIT’20c); 3) Designed novel minimax rate optimal kNN based classification and regression (TIT’21); 4) Provided novel analysis of kNN based density estimator (TIT’22); and 5) Established the optimality of kNN based Q-learning algorithm for reinforcement learning (TIT’25a).
  • Risk-sensitive Reinforcement Learning Algorithms: Reinforcement learning (RL) is an area of machine learning where agents learn from interacting with environment to determine actions. Existing risk-neutral approaches aim to minimize the expected total discounted cost and do not take the tail of the distributions of the cost into consideration. In the tail of the distribution, the cost may be prohibitively high, even though the probability of happening is low. Realizing the potential drawbacks of risk-neutral approach in safety critical applications, the risk-sensitive approach is gaining more research attentions. In risk-sensitive approach, instead of merely minimizing the expected cost, one aims to design decision policies that take the risk into consideration. Even though a multitude of risk measures have been extensively studied in the literature and successfully applied to RL, existing work face the following challenges: 1) Most of these applied risk measures, with the exception of CVaR, are not coherent; 2) While there is an increasing demand for a broader range of choices in risk measures to better align with individual risk preferences in complex scenarios, there is a lack of unified framework that enables efficient design of risk-sensitive RL algorithms tailoring to users’ different choices of risk measures suitable for their applications; and 3) There may be model uncertainties and model shifting in practical applications, but existing risk-sensitive RL algorithms are not robust to such scenarios. To address these challenges, our group has been focusing on systematically employing a class of risk measure named coherent risk measures that are widely studied in economic, to design efficient algorithms for risk-sensitive RL. In particular, we have: 1) developed a unified framework for the efficient design of risk-sensitive RL using coherent measures (TIT’25b); and 2) based on the developed framework, we have designed robust risk sensitive RL algorithms that are robust to model uncertainties and model mis-specifications (ICML’24).

4. Strengthening Minority Serving Institute (MSI) Research Capability Initiative

In collaboration with Professor Michael Cho of California State University Northbridge (CSUN), a minority serving institute, and Professor Weiyu Xu of University of Iowa, we developed a large proposal to leverage research capacity of UC Davis and U Iowa to strengthen the research capability at CSUN and create a pathway for students from MSI to pursue graduate studies at R1 universities.

Service

1. Professional Service:

I am very active in providing professional service to the research community.

  • Editorship: 1) Senior area editor: I am currently serving as senior area editor for IEEE Transactions on Information Forensics and Security since 2024, senior area editor for IEEE Transactions on Signal and Information Processing over Networks since 2025. In this review period. 2) Associate editor: I am currently serving as an associate editor for IEEE Transaction on Information Theory since 2021. In this review period, I have also served an associate editor for IEEE Transactions on Information Forensics and Security from 2015 to 2020, an associate editor for IEEE Transactions on Signal and Information Processing over Networks from 2021 to 2024, and an associate editor for IEEE Transactions on Mobile Computing from 2021 to 2025.
  • Proposal Review: In this review period, I served in more than 10 panels for several different National Science Foundation programs (CAREER, CCF, CPS, ECCS, and TIP etc). I also regularly review proposals for Israel Science Foundation. In addition, I reviewed proposals for private foundation such as C3.ai Digital Transformation Institute.
  • Technical Program Committee/Chair and Paper Reviews: I served in technical program committees as member or chair for numerous conferences. I also regularly review papers for a variety of journals and conferences.

2. Internal Service:

In this review period, I took substantial internal services. I highlight several of them:

  • Vice Chair for Graduate Studies and Chair Designate of ECE Graduate Program. Different from other universities, graduate programs at UC Davis are organized in an interdisciplinary manner. For the ECE Graduate Program, we have around 35 faculty members from the ECE department and around 40 non-ECE faculty members. In the ECE department, Vice Chair for Graduate Studies also serves as the Chair  Designate of the ECE graduate program simultaneously. I served in these capacities from 2022 to 2024. I was responsible for the overall operation (please see page 5 of ECEGP bylaw for the list of duties) of the ECE graduate program consisting of around 250 PhD students and more than 100 MS students. During my term, to improve the quality of the program and to improve the efficiency of the program administration, working closely with other colleagues and program staff in the program, I initiated and implemented many changes. For example, I led the final approval process of ECE graduate degree requirement updates (this multi-year process started two chairs before me, and went through several rounds of revisions before the final campus approval). After the campus approval, I worked closely with faculty and students for the success implementation of the new degree requirements. I also worked with Graduate Student Associate (GSA) to develop the first ever ECE graduate program conflict management guideline to help faculty and students better handle faculty-student conflicts. Moreover, my term as the chair was coincident with the UC academic works strike and unionization. These events brought significant challenges to the operation of the program and many university policy changes. As the program chair, I led the efforts to manage and improve faculty-student relations during these challenging time, and led the implementations of university policy changes such as graduate student funding, research unit requirements etc.
  • Other Committee Services. I also served in many other department and college committees, such as faculty recruitment committee, college graduate studies committee, chair of unit 18 lectures recruitment committee, and undergraduate studies committee etc. Please see my CV for a detailed list.

Teaching and Mentoring

  • Classroom Instruction. I regularly teach both undergraduate and graduate courses related to communications, information theory, signal processing and machine learning. The courses that I taught tend to be large and math heavy (for example, I teach undergraduate level probability course with more than 100 students enrolled almost every year). Despite the challenges associated with large and math heavy courses, I consistently received high teaching evaluations from students. I also developed a new permanent graduate course on reinforcement learning.
  • Graduate Education. I regularly maintained a research group with 5-6 PhD students and several MS students. In this review period, I graduated 8 PhD students and 4 MS students. All of them landed good jobs upon graduation.
  • Education Outreach. I am very active in teaching and mentoring outreach for students ranging from high school, undergraduate to graduate schools. In this review period, I mentored 3 high school students for them to gain research experience. I advised several undergraduate students from various universities such as Brown University and Tsinghua University for undergraduate research. Furthermore, I was invited by UC Davis Associate Dean of Graduate Studies to serve as a faculty mentor for Professors for the Future (PFTF) program from 2023 to 2024. This program is organized by the UC Davis Graduate Studies. This is a competitive, leadership-development program that provides outstanding Ph.D. students and postdoctoral scholars with an opportunity to develop leadership prowess, refine problem-solving skills and acquire an advanced understanding of the university system.