diff --git a/assets/img/project/traincheck_logo.png b/assets/img/project/traincheck_logo.png new file mode 100644 index 0000000..9dbe3ed Binary files /dev/null and b/assets/img/project/traincheck_logo.png differ diff --git a/assets/img/team/yuxuan.jpg b/assets/img/team/yuxuan.jpg index 892e150..87c00e7 100644 Binary files a/assets/img/team/yuxuan.jpg and b/assets/img/team/yuxuan.jpg differ diff --git a/index.html b/index.html index cfa386a..08bd848 100644 --- a/index.html +++ b/index.html @@ -85,6 +85,28 @@

Recent Projects

+
+
+ TrainCheck +
+
+

+ Catching Silent Errors in Deep Learning Training +

+

+ Silent errors in deep learning training can silently waste + thousands of GPU hours and produce low-quality models. We + introduce TrainCheck, a proactive checking framework that learns + semantic invariants from correct training runs and enforces them + at runtime to catch failures early—before they silently + accumulate cost and damage model reliability. +

+
+ Read More +
+
+
+
diff --git a/news.html b/news.html index 1b09fd1..34f7793 100644 --- a/news.html +++ b/news.html @@ -5,6 +5,14 @@
    +
  • + Mar 2025 + TrainCheck is accepted to appear at OSDI '25 +
    + [...] + Training deep learning (DL) models is a complex task involving multiple steps and various libraries, making DL training pipelines prone to silent bugs that lead to suboptimal or incorrect models. These issues are challenging to detect and diagnose. TrainCheck is the first framework that takes a proactive checking approach to systematically address silent issues. TrainCheck automatically infers invariants tailored for DL training. It uses these invariants to enhance a training task and proactively detect silent issues while providing debugging help. +
    +
  • May 2024 Yigong will join Boston University as an Assistant Professor! diff --git a/paper/traincheck-osdi25-preprint.pdf b/paper/traincheck-osdi25-preprint.pdf new file mode 100644 index 0000000..a5dbbd5 Binary files /dev/null and b/paper/traincheck-osdi25-preprint.pdf differ diff --git a/paper/traincheck-osdi25.bib b/paper/traincheck-osdi25.bib new file mode 100644 index 0000000..3face46 --- /dev/null +++ b/paper/traincheck-osdi25.bib @@ -0,0 +1,10 @@ +@inproceedings{TrainCheckOSDI2025, + author = {Jiang, Yuxuan and Zhou, Ziming and Xu, Boyu and Liu, Beijie and Xu, Runhui and Huang, Peng}, + title = {Training with Confidence: Catching Silent Errors in Deep Learning Training with Automated Proactive Checks}, + booktitle = {Proceedings of the 19th USENIX Symposium on Operating Systems Design and Implementation}, + series = {OSDI '25}, + month = {July}, + year = {2025}, + address = {Boston, MA, USA}, + publisher = {USENIX Association}, +} diff --git a/pubs.html b/pubs.html index 16fa45a..73a6fb5 100644 --- a/pubs.html +++ b/pubs.html @@ -7,9 +7,10 @@

    2025

+
+
+
+

TrainCheck [OSDI '25]

+

TrainCheck is an innovative tool for detecting silent errors in deep learning training. We are excited to open-source TrainCheck–explore the project and get involved on GitHub!

+
+
+