Update in Jun 2024

_pages/about.md

@@ -10,7 +10,7 @@ redirect_from:
 
 About Me
 ------
-I am a Ph.D. student in Mechanical Engineering at University of Michigan, working on Molecular Machine Learning in [EEG](https://www.cmu.edu/me/venkatgroup/) advised by [Prof. Viswanathan](https://aero.engin.umich.edu/people/viswanathan-venkat/). Earlier, I received M.S. in Computational Materials Science and Engineering at Carnegie Mellon University and did research in [Mechanical and AI Lab](https://sites.google.com/view/barati) advised by [Prof. Barati Farimani](https://www.meche.engineering.cmu.edu/directory/bios/barati-farimani-amir.html). Besides, I received B.Eng in Materials Science and Engineering at South China University of Technology.
+I am a Ph.D. student in Mechanical Engineering at University of Michigan, working on Molecular Machine Learning in [EEG](https://eeg.engin.umich.edu/) advised by [Prof. Viswanathan](https://aero.engin.umich.edu/people/viswanathan-venkat/). Earlier, I received M.S. in Computational Materials Science and Engineering at Carnegie Mellon University and did research in [Mechanical and AI Lab](https://sites.google.com/view/barati) advised by [Prof. Barati Farimani](https://www.meche.engineering.cmu.edu/directory/bios/barati-farimani-amir.html). Besides, I received B.Eng in Materials Science and Engineering at South China University of Technology.
 
 My research interest lies in combining Artificial Intelligence with interdisciplinary science and engineering problems. My current research focuses on implementing and improving foundation models for material discovery and leveraging scientific machine learning for electrolyte optimization. I believe that deep learning models are able to learn representations from data so that we can understand scientific problems from a data science view, which is a significant transformation from traditional research strategies. 
 

_pages/cv.md

@@ -7,15 +7,15 @@ redirect_from:
   - /resume
 ---
 
-You can download my [CV](http://ChangwenXu98.github.io/files/CV.pdf) (last updated September 2023)
+You can download my [CV](http://ChangwenXu98.github.io/files/CV.pdf) (last updated June 2024)
 
 {% include base_path %}
 
 Education
 ======
 * Ph.D. in Mechanical Engineering, University of Michigan, 2027 (expected)
-* B.E. in Materials Science and Engineering, South China University of Technology, 2021
 * M.S. in Computational Materials Science and Engineering, Carnegie Mellon University, 2022
+* B.E. in Materials Science and Engineering, South China University of Technology, 2021
 
 Work experience
 ======
@@ -53,7 +53,7 @@ Blog Posts
 
 Professional Services
 ======
-* Reviewer: NeurIPS'23, ICLR ML4Materials Workshop'23, ICML SPIGM Workshop'23
+* Reviewer: NeurIPS'23-24, ICLR'24, ICML'24, ICLR ML4Materials Workshop'23, ICML SPIGM Workshop'23-24, ICML AI4Science Workshop'24
 
 Skills
 ======

_talks/2024-04-03-Cloud.md

@@ -0,0 +1,14 @@
+---
+title: "CLOUD: A Scientific Foundation Model for Crystal Property Prediction"
+collection: talks
+type: "Poster"
+permalink: /talks/2024-04-03-Cloud
+venue: "MICDE Scientific Foundation Model Conference"
+date: 2024-04-03
+location: "Ann Arbor, MI"
+---
+
+
+**Abstract**
+
+Property prediction of crystals is crucial for material design. However, developing machine learning models for these tasks is hampered by the need for labeled data from costly experiments or Density Functional Theory (DFT), resulting in limited data size and poor generalization to new crystals. Foundation models (FMs) present a potential solution with their self-supervised pretraining on unlabeled datasets for better representation learning and transferability. Yet, applying FMs to crystals is challenging due to the sparse number of valid structures for pretraining and the inadequacy of existing representations to capture critical structural information like symmetry. Herein, We propose the CrystaL fOUnDation model (CLOUD), a Transformer-based foundation model for crystal property prediction. CLOUD utilizes a novel symmetry-aware string representation that efficiently encodes symmetry, equivalent sites, and constituting atoms, eliminating the need for coordinate information or equivariant models. Pretrained on million-scale crystal data from various databases via Masked Language Modeling (MLM), CLOUD is then fine-tuned and assessed on eight MatBench datasets. The model not only significantly outperforms structure-agnostic models and achieves near state-of-the-art results on two datasets, but also demonstrates robust scaling with data and model size. This suggests CLOUD's potential as a scalable solution for crystal foundation models, capable of learning from billions of unlabeled crystal data.

_talks/2024-06-19-Cloud.md

@@ -0,0 +1,15 @@
+---
+title: "CLOUD: A Scientific Foundation Model for Crystal Property Prediction"
+collection: talks
+type: "Poster"
+permalink: /talks/2024-06-19-Cloud
+venue: "Molecular Machine Learning Conference"
+date: 2024-06-19
+location: "Montreal, Quebec"
+---
+
+Spotlight Paper at MoML 2024
+
+**Abstract**
+
+Property prediction of crystals is crucial for material design. However, developing machine learning models for these tasks is hampered by the need for labeled data from costly experiments or Density Functional Theory (DFT), resulting in limited data size and poor generalization to new crystals. Foundation models (FMs) present a potential solution with their self-supervised pretraining on unlabeled datasets for better representation learning and transferability. Yet, applying FMs to crystals is challenging due to the sparse number of valid structures for pretraining and the inadequacy of existing representations to capture critical structural information like symmetry. Herein, We propose the CrystaL fOUnDation model (CLOUD), a Transformer-based foundation model for crystal property prediction. CLOUD utilizes a novel symmetry-aware string representation that efficiently encodes symmetry, equivalent sites, and constituting atoms, eliminating the need for coordinate information or equivariant models. Pretrained on million-scale crystal data from various databases via Masked Language Modeling (MLM), CLOUD is then fine-tuned and assessed on eight MatBench datasets. The model not only significantly outperforms structure-agnostic models and achieves near state-of-the-art results on two datasets, but also demonstrates robust scaling with data and model size. This suggests CLOUD's potential as a scalable solution for crystal foundation models, capable of learning from billions of unlabeled crystal data.