The Hidden Story Behind the Suicide Ph.D. Candidate Huixiang Chen
On June 13, 2019, just before the ISCA 2019 conference in Phoenix, a doctoral candidate, whose paper was supposed to be published at this conference, hanged himself in the campus building of the University of Florida.
As more and more information comes out, we find that there should be a lot of hidden stories behind his suicide. The most important one may be related to his struggle against academic misconduct.
A suicide note was sent to his friends, lab colleges, and especially his mentor Doctor Tao Li, a tenured professor (with preeminence professorship) in the Department of Electrical and Computer Engineering, also the last author of the paper mentioned in Huixiang’s suicide note.
The original version of the suicide note is in Chinese:
Here we translated it into English (This is not a certified legal translation):
Dear fellow colleagues:
If you get this message, that means I have already passed away.
As you know, since last year December, I wrote an ISCA paper in a very short time and it was submitted and accepted very quickly. As all of you know the reason, it was owe to our professor Dr.Tao Li ’s networking. Among the six reviewers of my paper, four of them are friends of Tao Li. But this paper has very severe issues: the design doesn’t make any sense and the reviewers also pay no attention to it. The paper was accepted at no reason which can make anyone feel guilty in this case.
After the deadline, crazy modifications of the paper were made and I started to make up for the missing experiments. During those experiments, I found that the phenomenon of the experiments and the design was different from what was claimed before, which made this paper doesn’t make any sense from title to characterization and to design. For example, It was claimed that this (the accelerator) is targeted for 3D CNN, which can maximize the performance by changing Diffy architecture to achieve the dynamic control of the temporal delta and spatial delta. But there are too many issues with this. First of all, the Diffy architecture has a serious problem of synchronization. If the width of the input feature is too narrow, the overhead of lane synchronization … . But it was assumed that the temporal dimension of 3D CNN (the maximum is generally 16 and it will decrease with layer) will expand with Diffy. Secondly, it can’t skip zero, because of the synchronization of lane and the incremental of spatial/temporal delta. So if I want to make a change the whole paper need to be rewrite. It is unrealistic to propose a totally new idea before the final version. After I communicated with Tao Li, it turned out that I have to fix the paper by myself. Until today I am still unable to patch up any of those issues. These issues are too obvious which can be easily caught by any experts of accelerator. I have no words to comment that Dr. Li acquiesced to the data which makes no sense.
Considering that this will have an impact on my career in the future and my reputation in the area of Computer Architecture, my future life will be worse than death and I will be totally in a dilemma. I considered all the cases, I really can’t work them around. In order to make up for the fault, I decide to suicide. I hope you can learn my lessons and don’t mess around things. We shall never claim too much in the paper before we can really achieve it. I hope this will make a change in this world.
I hope you can keep simple and stay honest in this society. I will bless you in another world.
Best
Huxiang
After the tragedy, some dialogues between Huixiang Chen and his friends are exposed, which also shed some light on his struggling against academic misconduct and the intense relationship between him and Dr. Tao Li. (Translations here are not certified legal translation. To protect the information providers, we cover the content of their words.)
Only a suicide note and several pieces of dialogues may not uncover all the truth. University of Florida and the ECE department should have an elaborate and responsible investigation on this event.
