Ph.D., LAMDA group

Department of Computer Science

National Key Laboratory for Novel Software Technology

Nanjing University

Social links: sina weibo zhihu twitter github

Google scholar: My Google Scholar

Email: fengji_at_chuangxin.com (work)


                  Ji FENG

" ...because the people who are crazy enough to think they can change the world,

are the ones who do."

[Update, 2021] It was my greatest honor to be the recipient of CCF Outstanding PhD dissertation Award. This is the highest honor one PhD thesis can get in China. This national award gives only 5-10 recipients per year. Thank everyone who helped me in the past, especially my mentor Prof. Zhi-Hua Zhou.

[Update, 2020] On 14/04/2020, I have defended my PhD dissertation, and became a PhD.

[Update, 2018] Starting from Aug 2018, I have officially joined Sinovation Ventures, one of the most influential VC in the world led by Dr. Kai-Fu Lee. (He made an offer I cannot refuse.) I am the founding deputy dean of Nanjing AI Research Institution. Leading an AI institution is one challenging but exciting work, and we are always hiring exceptional people to join us!

[Update, 2021] 本人有幸荣获2020年度 CCF全国优博奖。CCF优秀博士学位论文奖是由中国计算机学会设立,授予在计算机科学与技术及其相关领域的基础理论或应用基础研究方面有重要突破,或在关键技术和应用技术方面有重要创新的中国计算机领域博士学位论文的作者。2006年设立,每年仅5-10位获奖者。感谢导师和所有帮助过我的朋友!

[Update, 2020] 本人于2020年4月14日通过博士论文答辩。

[Update,2018]本人于2018年8月加入创新工场,担任创新工场南京国际人工智能研究院执行院长。欢迎杰出人才加盟研究院,职业发展机会参见 这里

I joined the department as a PhD student of Department of Computer Science and Technology in Nanjing University since 2015 and became a member of LAMDA Group , led by Professor Zhi-Hua Zhou.

Before that, I obtained B.Eng. Degree in Computer Science from Taishan College , Shandong University . (Taishan College is the so-called "pilot class" led by MOE which intakes only 15 students from 10,000 freshmen per year.) My research interest in SDU was mainly on Information Visualization and Visual Analytics. I also did some fun stuffs in differential privacy and computational geometry. Upon graduation at SDU, I was enrolled in the direct PhD program at Nanjing University.

I was a semi-professional young writer for some time. For instance, I had won some quite prestigious nation-wide awards for young writers including this one (first prize). I abandoned this superpower for quite a while and I do NOT have any intention to continue this path. However, I still do enjoy reading great books and explaining ideas (scientific ideas, mostly) in the simplest way.

Professional Services

Vice-Chair, IEEE P3652.1 Committee

IEEE P3652.1 Guide for Architectural Framework and Application of Federated Machine Learning is the world’s first international standard on AI collaboration and security.

Media Coverage (selected) 媒体报道(部分)
[1] People's Daily (人民日报海外版) English Version
[2] (创新工场) (In Chinese)

Research Interest

"Trees are computer scientists' best friends."

---Donald Knuth

My current research interest includes the following:

  • Deep Forest (a.k.a. tree based deep learning)
  • High Performance Computing (from an AI point of view)
  • Deep Neural Networks
  • Quantitative Trading using AI methods
  • Safety and Robust AI


"I have discovered a truly marvelous proof of this,

which this margin is too narrow to contain."

--Pierre de Fermat


Multi-Layered Gradient Boosting Decision Trees
J. Feng, Y. Yu, and Z.-H. Zhou.
In:Advances in Neural Information Processing Systems 31 (NIPS'18), Montreal, Canada, 2018.

Multi-layered representation is believed to be the key ingredient of deep neural networks especially in cognitive tasks like computer vision. In this work, we propose the multi-layered GBDT forest (mGBDTs), with an explicit emphasis on exploring the ability to learn hierarchical distributed representations by stacking several layers of regression GBDTs as its building block. The model can be jointly trained by a variant of target propagation across layers, without the need to derive back-propagation nor differentiability.
[code on github]


Distributed Deep Forest and its Application to Automatic Detection of Cash-out Fraud
Y-L Zhang, J. Zhou, W.H. Zheng, J. Feng, L.F Li, Z.Q. Liu, M. Li, Z.Q. Zhang, C.C. Chen, X.L Li, and Z.-H. Zhou.
arXiv preprint: 1805.04234.

This is a joint work with Ant Financial. We implemented a distributed version of deep forest and successfully applied it on the detection of cash-out fraud with more than 100 millions of training samples. An easy-to-use GUI was also deployed inside Ant Financial and Alibaba. Now data scientists can build the deep forest model with only few mouse clicked.


AutoEncoder by Forest
J. Feng and Z.-H. Zhou.
In: Proceedings of the 32nd AAAI Conference on Artificial Intelligence (AAAI'18), New Orleans, Lousiana, USA, 2018.

We proposed the first forest based auto-encoder by enabling tree ensembles to perform autoencoding tasks. The key motivation is that can we extract useful information along decision paths(rather than the information stored in the leaf node)? Our idea is that by traversing from leaf nodes back to the root and caculating the maximum compatible rules, we can then get an amazingly good estimate of the input pattern.
[code on github]


Deep forest: Towards an alternative to deep neural networks.
Z.-H. Zhou and J. Feng.
In: Proceedings of the 26th International Joint Conference on Artificial Intelligence (IJCAI'17), Melbourne, Australia, 2017, pp.3553-3559.

Deep learning models were usually achieved by constructing multi-layered differentiable modules and been trained via back-propagation. In this work, we proposed the first deep model called gcForest via non-differentiable components without the need for backprop. Compared with deep neural networks, gcForest achieves highly competitive performance with much fewer hyper-parameters.
[code on github]


Deep MIML Network.
J. Feng and Z.-H. Zhou.
In: Proceedings of the 31st AAAI Conference on Artificial Intelligence (AAAI'17), San Francisco, CA, 2017,pp.1884-1890.

We proposed an unified DNN framework called DeepMIML network for Multi-Instance Multi-Label learning. By introducing the sub-concept tensor layer, the network is able to explore the instance-label relationships and can automatically detect key instances for a particular label.
[code on github]

Work Experience

"If you are good at something, never do it for free."

----The Dark Knight

Quant Intern at WizardQuant

Jan 2015 - May 2015

NOTE: This is the work BEFORE my enrollment at LAMDA

I worked as a "quant researcher" at WizardQuant, a private hedge fund.
My primary contribution is introducing machine learning techniques into High Frequency Trade algorithms.

P.S. I still do quant trading using AI systems built by myself in spare time. Automated trading via AI methods (instead of traditional technical methods) should be the only way to go.

P.P.S. My system works.


"Winner Winner, Chicken Dinner"

---Anonymous Quotes

Most Bussiness Potential Award, 2017

National CCF-Intel Parallel Computing Competition

This is a contest on implementing high performance computing softwares. With over 400 participants, this award goes to only 3 teams.

Frist Prize, 2013

Outstanding Undergraduate Research Programme

This is a research grand (¥10,000 per project) organized by MOE to encourage undergraduate students to participant in academic research. My research on differential privacy won the first prize.

Grand Prize, 2013

National Entrepreneurship Chanllenge

The National Entrepreneurship Chanllenge held by MOE, is the biggest bussiness plan competition for students in China. I won the grand prize(特等奖) in province.

Bronze Medal, 2012

ACM/ICPC Asia, Changsha Site

This is a contest on programming and problem solving for college students across the world.

Contact Information

"Send me the raven."

---Game of Thrones