Deren Lei



Deren Lei is currently a graduate student at USC. He is interested in building practical intelligent agents that could learn to reason under uncertainties in the real-world scenario. Specifically, develop efficient deep learning solutions to product development and enhancement. His current focus lies in the area of natural language processing and deep learning.

Work Experiences

NLP Applied Scientist Intern

May 2020 – August 2020
Microsoft Corporation

Research, design, and deploy practical deep learning & NLP solutions for a relevance search product over relational databases, especially in the business domain. (Details omitted due to non-disclosure agreement)

Continue working on a relevant research topic in the natural language generation area after the internship as a side project.

NLP Research Assistant

June 2019 – Present

Focusing on neural symbolic reasoning with an emphasis on natural language and knowledge graph applications:

  • Leveraged symbolic rules with collaborative RL agents for knowledge graph reasoning. (In Submission)
  • Using self-supervised learning to learn relations in knowledge graph and augment existing embedding approach. (In submission)

NLP Research Assistant

Oct 2017 – June 2019

Spend majority of the time on NLP relevant research starting from 3rd year. Mostly interested in natural language generation:

  • Proposed new reinforcement learning approach with distributional semantics for abstract summarization.
  • Assisted other researchers to improve fine-grained entity typing performance from internal inconsistency issues.
  • Expanded the CNN/Dailymail dataset with related images and videos for new multi-model tasks.
  • Observed and analyzed the implicit regularization of vanilla stochastic gradient descent in three common NLP tasks.

Software Development Engineer Intern

Summer 2017
  • Created an internal tool helping developers to do automatic product page status checking.
  • Did web crawling and data preprocessing for softline product recommendation deep learning algorithm


  • "Self-Supervised Relation Learning for Knowledge Graph Reasoning"
    Deren Lei* , Gangrong Jiang*, Kexuan Sun, Xiang Ren (* equal contribution)

  • "Learning Collaborative Agents with Rule Guidance for Knowledge Graph Reasoning" (In submission)
    Deren Lei* , Gangrong Jiang*, Xiaotao Gu, Kexuan Sun, Yuning Mao, Xiang Ren (* equal contribution)
    EMNLP 2020 , short paper, [paper], [code]

  • "Learning to Reason in Round-based Games: Multi-task Sequence Generation for Purchasing Decision Making in First-person Shooters"
    Deren Lei* , Gangrong Jiang*, Beichen Li*, Yilei Zeng*, Emilio Ferrara, Michael Zyda (* equal contribution)
    AIIDE 2020 , research track, [paper], [code]

  • "Deep Reinforcement Learning with Distributional Semantic Rewards for Abstractive Summarization"
    Deren Lei* , Siyao Li*, Pengda Qin, William Yang Wang (* equal contribution)
    EMNLP-IJCNLP 2019 , short paper, [paper]

  • "Imposing Label-Relational Inductive Bias for Extremely Fine-Grained Entity Typing"
    Wenhan Xiong, Jiawei Wu, Deren Lei , Mo Yu, Shiyu Chang, Xiaoxiao Guo and William Yang Wang
    NAACL-HLT 2019 , long paper, [paper]

  • "Implicit Regularization of Stochastic Gradient Descent in NLP: Observations and Implications"
    Deren Lei , Zichen Sun, Yijun Xiao, William Yang Wang
    Technical Report , long paper, [paper]


USC Teaching Assistant

Fall 2019, Fall 2020
USC Viterbi School of Engineering

Principles of Programming for Informatics (Graduate level).

  • Using python and SQL for data processing, modeling, and analysis.

Applied Machine Learning for Games (Graduate Level).

  • Apply deep learning and reinforcement learning to build game AI.
  • I have also taken this course previously and my team's course project has been accepted by AIIDE-20 as a research track paper. It's my honor to work with my amazing collaborators!