wmdi.github.io/index.html at master · wmdi/wmdi.github.io · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
---
layout: home
author_profile: true
---

<div style="padding-left: 10%;">
  <h1> About Me </h1>
  <p>
    I am Mengdi Wu (吴梦迪), a third-year Ph.D. student in the <a href="https://csd.cmu.edu/">Computer Science Department</a> at Carnegie Mellon University, advised by <a href="https://www.cs.cmu.edu/~zhihaoj2/">Zhihao Jia</a>. I received my undergraduate degree from the <a href="https://iiis.tsinghua.edu.cn/en/yaoclass/">Yao Class</a> at Tsinghua University in 2023. I am honored to be supported by the Amazon Fellowship for my research.
  </p>
  <p>
    I work on machine learning systems, with a focus on ML compilers and superoptimization.
    I am interested in understanding the computational structure of machine learning applications and developing techniques to make them more efficient and reliable.
  </p>

  <h2> Research Experience </h2>

  <table>
      <th>2022.02~2022.08</th> <td> I visited CMU as a research intern, where I worked with <a href="https://www.cs.cmu.edu/~zhihaoj2/">Zhihao Jia</a> on accelerating distributed DL training. </td>
    </tr>
    <tr>
      <th>2020.11~2021.11</th> <td> I joined <a href="https://group.iiis.tsinghua.edu.cn/~maks/">ARChip Lab</a> led by <a href="https://group.iiis.tsinghua.edu.cn/~maks/leader.html">Kaisheng Ma</a> at Tsinghua University, and worked on designing architectures and schedulers for DNN accelerators. </td>
    </tr>
  </table>

  <h2> Awards </h2>
  <div class="award__">
    <ul>
      <li>
        <a href="https://compiler.educg.net/2020CSCC">National Student Computer System Capability Challenge 2020 (Compiler track)</a>, 2nd Price [<a href="https://github.com/Cutepiler/Cutepiler-Sysy2020">Cutepiler</a>]
        <p class="comment__">With Jiatu Li, Runda Liu, and Zhidong Wang. </p>
      </li>
      <li>
        National Olympiad in Informatics, China, Gold Medal
      </li>
    </ul>
  </div>

  <h2> Publications </h2>
  <div class="project__">
    <ul>
      <li class="project__title">
        <a href="https://arxiv.org/pdf/2506.04529">
          Identity Testing for Circuits with Exponentiation Gates
        </a>
      </li>
      <li class="project__author">
        Jiatu Li<sup>*</sup>, Mengdi Wu<sup>*</sup>
      </li>
      <li class="project__publish">
        ITCS 2026
      </li>
    </ul>
  </div>
  <div class="project__">
    <ul>
      <li class="project__title">
        <a href="https://arxiv.org/pdf/2402.18789">
          Collie: Token-Level Co-Serving of LLM Inference and Fine-Tuning with SLO Guarantees
        </a>
      </li>
      <li class="project__author">
        Gabriele Oliaro, Xupeng Miao, Xinhao Cheng, Vineeth Kada, Ruohan Gao, Yingyi Huang, Remi Delacourt, April Yang, Yingcheng Wang, Mengdi Wu, Colin Unger, Zhihao Jia
      </li>
      <li class="project__publish">
        NSDI 2026
      </li>
    </ul>
  </div>
  <div class="project__">
    <ul>
      <li class="project__title">
        <a href="https://www.usenix.org/system/files/osdi25-wu-mengdi.pdf">
          Mirage: A Multi-Level Superoptimizer for Tensor Programs
        </a>
      </li>
      <li class="project__author">
        Mengdi Wu, Xinhao Cheng, Shengyu Liu, Chunan Shi, Jianan Ji, Kit Man Ao, Praveen Velliengiri, Xupeng Miao, Oded Padon, Zhihao Jia
      </li>
      <li class="project__publish">
        OSDI 2025
      </li>
    </ul>
  </div>
  <div class="project__">
    <ul>
      <li class="project__title">
        <a href="https://dl.acm.org/doi/pdf/10.1145/3669940.3707220">
          GraphPipe: Improving the Performance and Scalability of DNN Training with Graph Pipeline Parallelism
        </a>
      </li>
      <li class="project__author">
        Byungsoo Jeon<sup>*</sup>, Mengdi Wu<sup>*</sup>, Shiyi Cao<sup>*</sup>, Sunghyun Kim<sup>*</sup>, Sunghyun Park, Neeraj Aggarwal, Colin Unger, Daiyaan Arfeen, Peiyuan Liao, Xupeng Miao, Mohammad Alizadeh, Gregory R. Ganger, Tianqi Chen, Zhihao Jia
      </li>
      <li class="project__publish">
        ASPLOS 2025
      </li>
    </ul>
  </div>
  <div class="project__">
    <ul>
      <li class="project__title">
        <a href="https://dl.acm.org/doi/pdf/10.1145/3676536.3676692">
          LACO: A Latency-Constraint Offline Neural Network Scheduler towards Reliable Self-Driving Perception
        </a>
      </li>
      <li class="project__author">
        Zhanhong Tan, Zijian Zhu, Mengdi Wu, Kaisheng Ma
      </li>
      <li class="project__publish">
        ICCAD 2024
      </li>
    </ul>
  </div>
  <div class="project__">
    <ul>
      <li class="project__title">
        <a href="https://proceedings.mlr.press/v162/dong22a.html">
          Finding the Task-Optimal Low-Bit Sub-Distribution in Deep Neural Networks
        </a>
      </li>
      <li class="project__author">
        Runpei Dong<sup>*</sup>, Zhanhong Tan<sup>*</sup>, Mengdi Wu, Linfeng Zhang, Kaisheng Ma
      </li>
      <li class="project__publish">
        ICML 2022
      </li>
    </ul>
  </div>

</div>