Dimitrios Christofidellis

Accelerating IP generation in materials science using #NLP for patent classification. See how adaptive fine-tuning on #transformers improves over SOTA. @IBMResearch  @QUBelfast @IOPPublishing #IOPPposter #materials

Great to see our work @IBMResearch on prompt driven retrosynthesis in @ACSCentSci highlighted by @OPRD_ACS alongside several former colleagues and collaborators! 🌟Highlight: pubs.acs.org/doi/10.1021/ac… 📰Paper: pubs.acs.org/doi/10.1021/ac… #NLP #MachineLearning #Chemistry #AI

🎉 Wrapping up an incredible 2.5 days of non-stop excitement at #ACSFall2023 ! 🚀 Can you believe it? We rocked: 👥 47 demos with our awesome visitors 🔮1512 model predictions 🧠 2 model retraining And yes, we might've unintentionally set a world record for trashed coffee ☕ 1/3

📢Interested in learning more about Structure Elucidation from IR spectra and at ACS in San Francisco? Come stop by my talk on Wednesday at 8:45 in Hotel Nikko or read the paper: ibm.biz/ir_to_struc @acvaucher @teodorolaino @IBMResearch

Here some of the new technologies you can experience.

📢 If you are attending ACS Fall 2023 in San Francisco and you want to know more about #LanguageModels applied to material design, stop by Nob Hill A - Marriot Marquis at 10:25AM today @IBMResearch

If you are #ACSFall2023 and want to learn how we can utilize AI to predict the right solvent for a chemical reaction, check out my talk today at 3:10Pm at Nikko II - Hotel Nikko or read the preprint doi.org/10.26434/chemr… @IBMResearch @SchwallerGroup

Very excited our Preprint “Leveraging Infrared Spectroscopy for Automated Structure Elucidation” is out. First of its kind model: Automatic Structure determination from IR spectra! Top-1 Accuracy of 45%! @acvaucher @teodorolaino @IBMResearch Paper: ibm.biz/ir_to_struc 🧵⬇️

Explore our latest paper on designing catalysts using deep generative models and computational data. Discover how an RNN-VAE can generate new catalyst candidates for the Suzuki coupling. doi.org/10.1039/D2DD00… @IBMResearch @SchwallerGroup @acvaucher @pschwllr @teodorolaino

🚀 Exciting news! Our paper on GT4SD has been published in @Nature_NPJ Computational Materials! Learn more about it here: doi.org/10.1038/s41524…. Try it out: GitHub - github.com/GT4SD/gt4sd-co… Demo apps on HF Spaces - huggingface.co/GT4SD

A team work featuring: @georgosgeorgos @JannisBorn @OleWinther1 @teodorolaino and @drugilsberg @IBMResearch @DTU

Our paper “Unifying Molecular and Textual Representations via Multi-task Language Modelling” got accepted at #ICML2023🎉🎉. This work presents Text+Chem T5, a multidomain/multitask LM for both the chemical and natural language domains. Try it out now: ibm.biz/textchemt5

Link to the preprint: arxiv.org/pdf/2301.12586… The model is also already available via the #GT4SD python library.

Easter egg🥚in @NatMachIntell: Regression Transformers! Sounds contradicting? Not to us! ➡️ SOTA performance in (bio)chemical regression tasks w/o regression heads. Pure MLM! The best: RT is dichotomous & also excels at conditional molecular generation🧵ibm.biz/rtnmi23

If you're at #ACSSpring2023 next week, consider attending some of the talks from our team! We will be presenting some of the work done at IBM Research Europe in the space of AI and chemistry. Here are the talk details: ibm.biz/acs-spring-2023 (1/3)

🤯 Check out this groundbreaking work by @georgosgeorgos, @jannisborn & more: Unifying Molecular & Textual Representations via Multi-task Language Modelling! 🧠 #Language #Computation #ML deepai.org/publication/un…

Our preprint on retrosynthesis models with boosted diversity is out! preprint: openreview.net/forum?id=Bmk_G… Check out the code at: github.com/rxn4chemistry/… . Open-source models are also available! 🙌 @acvaucher @pschwllr @teodorolaino #compchem #MachineLearning #Chemistry @ForRxn

We are excited to be at #ACSFall2022 with 7 contributions from the Accelerated Discovery team in @IBMResearch Zurich covering a full range of topics. I could not be prouder of all of them. Virtual or in person, don't miss their contributions at #ACSChicago, details in the 🧵 1/8

I'm delighted to share that our paper "pylspack: Parallel algorithms and data structures for sketching, column subset selection, regression and leverage scores" has been finally accepted for publication at ACM TOMS. You can find a preprint at dl.acm.org/doi/10.1145/35…

New feature in #paperscraper (0.2.4), our package for paper keyword searches & scraping publication metadata. Now you can download PDFs, either for individual papers or for scraped metadata dumps 🚀🎉 Install via pip. Code: ibm.biz/paperscraper Thx for suggestions @skepteis

For more information, you can have a look at the preprint ibm.biz/BdPivy, or feel free to reach us!

Do you want to speed up your patent application drafting or just curious about the model? You can access the model via hf’s hub ibm.biz/pgt_model or use it directly from the #GT4SD python library (including our suggested postprocessing steps) ibm.biz/gt4sd-pgt-note…

A super interesting week in Baltimore attending @icmlconf is coming to an end. Yesterday, I had the pleasure to present our Patent Generative Transformer (PGT) model during the KRLM workshop. A Language Model able to understand patents and facilitate the patent drafting process.