Transfer learning and generalization problems for representation learning, including different data modalities like knowledge graphs, protein sequences, etc.
Excited to announce our ICML2023 Oral paper, ProtST! It's a general framework that enhances protein sequence pre-training and understanding with biomedical texts. Discover more:
Never thought I myself would run into any plagiarism-related issue in such way … (our A*Net paper is plagiarized by an ICLR submission)
@iclr_conf
we all authors call for some serious actions on this matter. Some LINES should NOT be crossed.
Really shocked today...
My paper A*Net has been on arXiv for more than a year, and it was plagiarized in an ICLR submission. There is no reference to A*Net and more than half of the methodology is identical to my paper.
@iclr_conf
any solution?
up: my A*Net. down: plagiarism.
Our foundation model for knowledge graphs (KGs)!✨
It tackles a interesting and important problem that different KGs have different vocabulary of relations and entities, thus cannot be modeled by a shared network in the first glance.
Unifying models for KGs now is possible!
🎉New paper alert!
It’s time to announce ULTRA - a pre-trained foundation model for knowledge graph reasoning that works on _any_ graph and outperforms supervised SOTA models on 50+ graphs.
Arxiv:
Code:
🧵1/n
ULTRA got accepted to
#ICLR2024
- see you in Vienna :)
More importantly though, my dear co-author
@XinyuYuan402
just collected her first 🏆 Grand Slam 🏆 of ML conferences (ICML -> NeurIPS -> ICLR in one cycle)! 🎉
I will be at
#ICML2023
from 23rd to 30th 😃
I will have an oral presentation for our ProtST project with
@MinghaoXu_Alan
and present a poster at KLR workshop () for A*Net, a joint work with
@zhu_zhaocheng
.
Would love to chat about LLM and AI4Science!!
Will be at
@icmlconf
next week and have an oral presentation of our ProtST project along with
@XinyuYuan402
. Happy to chat about protein design, AI4Science, multimodal LLM and meet old/new friends!
ProtST paper:
ProtST code:
📢 Great job! Our awesome collaborator
@santiagomiret
just penned an insightful blog about our ICML spotlight work ProtST. Dive in to see how a user can simply input the description of a protein to retrieve the corresponding proteins
Super super nice, supportive, knowledgeable and aspiring supervisor and collaborator when I had an internship at
@IntelAI
.
⭐️ apply it now and then get to learn all those valuable qualities from him!!!
Our graph team at
@IntelAI
is looking for an intern to work on foundation models for Graph ML including (but not limited to) KG reasoning and scalable GNNs.
Fully remote is ok, starting date: the sooner the better 😉
Application portal:
DMs are open
Our A*Net will be at NeurIPS'23 (as well as another work) 😌 Efficient and scalable graph reasoning (on commodity GPUs) enabled by good ole A*
@zhu_zhaocheng
and
@XinyuYuan402
did a great job, follow them!
latest arxiv:
🎊 At the year-end gala
#NeurIPS
. Let's discuss reasoning, LLM, KG & systems!
We baked two dishes for inductive reasoning: one scales to million-scale KGs (Wed. 3-5pm) and the other generalizes to arbitrary KGs (Fri. GLFrontiers). w/
@michael_galkin
@XinyuYuan402
@tangjianpku
In our new Medium blog post with
@XinyuYuan402
@zhu_zhaocheng
and special guest
@brunofmr
we explore
- the theory of inductive reasoning
- foundation models for KGs
- and explain our recent ULTRA in more detail!
Check out our latest ESM-GearNet project! We conduct a systematical study on the architecture design and learning methods of protein structure-sequence joint representations 👯
So happy to see this: "With the recent upstreams of torch-scatter and torch-sparse to native PyTorch, we are happy to announce that any installation of the extension packages torch-scatter, torch-sparse, torch-cluster and torch-spline-conv is now fully optional"
@zhu_zhaocheng
People always prioritize to chase what they don’t have now and have wanted the most, but realize later something lost or ignorant can then never be pursued back again, like health😅
Make ChatGPT more factual by equipping it with KG reasoning tools ⚒️
We release A*Net and integrate it with ChatGPT. A*Net is a scalable, inductive and interpretable path-based GNN on KGs. It is the first non-embedding method on the OGB leaderboard.
🧵1/6
(4/6) On downstream tasks, ProtST enables both supervised learning tasks and zero-shot prediction, utilizing the aligned representation space of protein sequences and textual descriptions. This versatility of ProtST opens new avenues in protein study and application.
@wangshengpkucn
Hi Prof. Sheng, thank you for the comments. It’s a great work. I think it shares the same essential idea as our ProtST to bring two modality (protein sequences and functional descriptions) into an aligned space, which is the key to zero shot prediction.