Liu Zhe

We developed a two-step ML + Bayesian optimization framework to rapidly optimize eco-friendly, scalable #PerovskiteSolarCell fabrication in ambient air — no anti-solvent, low-toxicity TEP solvent, and 24.26% PCE achieved! Newly online in JEC @jenergychem: doi.org/10.1016/j.jech…

Introducing you to the Co-PAS (Co-Plot for Additive Screening), an ML-driven framework designed to accelerate additive screening for perovskite solar cells. advanced.onlinelibrary.wiley.com/doi/abs/10.100…; github.com/MatAILab-NPU/C… .

Hello, world! Are you looking for an AI copilot while searching the vast chemical space of organic additives for perovskite solar cells? Preprint at: arxiv.org/abs/2412.14109; Open-source Code: github.com/MatAILab-NPU/C…

Buonassisi et al. use transfer learning to overcome data scarcity in high-throughput, automated experimental work in thickness characterisation of optoelectronic thin films. Read more in today's featured article: doi.org/10.1039/D2DD00… #openaccess

Thanks to Curtis @cpbCleanEnergy & Tonio @toniobuonassisi, ASTAR, SERIS & SMART team.

Finally, we got this thicknessML published in Digital Discovery @digital_rsc. We used transfer learning for thin film thickness analysis. Please see an early preview: doi.org/10.1039/D2DD00. Great collaboration with many awesome researchers.

@jpcorreabaena @BaenaLab @GTMSE5 @ACSEnergyLett @CarloPerini3 We are probably going to have papers published in the same issue. Looking forward to it.

📢 Publication alert @BaenaLab @GTMSE5 @ACSEnergyLett @CarloPerini3 studies the role of dimensionality on e- transport across interfaces: Mixed n=1 + n=2 (solution processed) hinders transport; pure n=2 2D (evaporation) improves transport pubs.acs.org/doi/full/10.10…

@Coolphoton @jpcorreabaena @BaenaLab @GTMSE5 @ACSEnergyLett @CarloPerini3 We probably have papers in the same issue.

@jpcorreabaena @BaenaLab @GTMSE5 @ACSEnergyLett @CarloPerini3 So interesting💯

On ML side, inspired by a talk from @prashungorai on avoiding data bias, we supplemented the previous research-intuition-driven dataset with five Latin-Hypercube-sampled ammonium salts. The data diversity gets increased, and the model prediction becomes more robust.

Following up our ML-assisted previous screening research led by @SunShijing @noortitan @toniobuonassisi, a new study found an ML-suggested molecule (2-phenylpropane-1- aminium iodide) experimentally achieves PCEs of 22.36% for FAMACs-based and 24.47% FAMA-based solar cells.

Initializing film homogeneity to retard phase segregation for stable perovskite solar cells - Science news.google.com/__i/rss/rd/art…

@Sergei_Imaging @toniobuonassisi @SterlingBaird1 @JuliaHsu_UTD @UT_Dallas I’m still trying to formalise the idea. But, my intuition is to incorporate device physics into optimisation.

@LiuZhe_MIT @toniobuonassisi @SterlingBaird1 @JuliaHsu_UTD @UT_Dallas Which specific application?

Another demo of a quick and informed optimization with BO and SHAP analysis. This time we tackle the challenge of making flexible perovskite solar cells. Here is a link: authors.elsevier.com/a/1fy8E_L4-wXN…. A great collaboration with Prof. @JuliaHsu_UTD 's team @UT_Dallas.

@Sergei_Imaging @toniobuonassisi @SterlingBaird1 @JuliaHsu_UTD @UT_Dallas I've read these papers. I guess the idea is to use physical laws to accelerate BO, correct? I will be cool to try these concepts in solar cells.

@toniobuonassisi @SterlingBaird1 @LiuZhe_MIT @JuliaHsu_UTD @UT_Dallas Not exactly - use possible physical laws as prior for mean function in GP. Like arxiv.org/abs/2202.01089 And onlinelibrary.wiley.com/doi/abs/10.100…

@SterlingBaird1 @JuliaHsu_UTD @UT_Dallas Cool idea! Lengthscale values are fitted independently for each input dimension. I will check how they are ranked. But, by definition, it should show the covariance on X values at different locations. Not the impact on the output y though. It would be a different perspective.

@LiuZhe_MIT @JuliaHsu_UTD @UT_Dallas Thoughts on when/how to use inverse lengthscales as feature importances vs. Shapley feature importances?

See the figure

Speaking of human-ML interaction, we expanded the search boundaries after the first round for the “PC length” variable. (Semi-)surprised to find out the PC tool have a limitation. So, the take-away: check your BO surrogate at every iteration with contour plot and SHAP analysis.

@John_W_Ward @LivUni @livuniphyssci Congrats.

Excited to announce that I will be joining the University of Liverpool as Lecturer in Chemistry from January 2023! @LivUni @livuniphyssci Keep your eyes peeled for 2 fully funded PhD studentships in organic synthesis 👩‍🔬 and robotics/automation 🤖, soon to be advertised

@Yang1Bai @nanotransport @ASTARsg @Sargent_Group @A3MD_UofT Safe travels. Stay in touch.

Thanks @nanotransport for all of your support from all aspects, I have great time in the past one and half years in @ASTARsg, and really looking forward to the new journey in @Sargent_Group working for the @A3MD_UofT to keep the acceleration of material science🚀🚀🚀

@WeiHe_1 @KCLEngineering @warwickuni Congratz, He Wei

Delighted to start the new position in Oct as Senior Lecturer @kclengineering. Everything went well. Feel grateful for all collaboration, support & guides in the past 3 years at @warwickuni. We have new posts (PDRA+PhD) coming soon, check out our group web shorturl.at/ASX69