Undergrad./High School Research

List of projects

Scientific

• JUNO water phase data analysis

JUNO水相数据分析

During comissioning, JUNO will be filled with water and form a 20kt water Cherenkov detector. Given its high photocathode coverage, JUNO has potentials to reach lower energy threshold than SuperK. It is thus interesting to see if JUNO may probe CNO solar neutrinos with this data.

• JUNO water phase reconstruction algorithm

JUNO水相重建算法

During comissioning, JUNO will be filled with water and form a 20kt water Cherenkov detector. Given its high photocathode coverage, JUNO has potentials to reach lower energy threshold than SuperK. It is thus necessary to build the reconstruction algorithm to support the data analysis.

• Sensitivity to NSI with JUNO B-8 solar neutrinos

利用JUNO硼-8太阳中微子寻找非标准相互作用

JUNO can lower the detection threshold of B-8 solar neutrinos from 3 MeV to 2 MeV, allowing constraining survival probability of solar neutrinos of around 6 MeV in the transition region. This result can be used to constrain the non-standard interaction.

• Global fit of solar neutrino fluxes

太阳中微子流强全局拟合

After Borexino, SAGE, and SuperK updated their results about solar neutrinos, it is time to update the global fit results of solar neutrino fluxes. This result can be further used to provide input to Standard Solar Models and provide hint to Solar Metallicity Problem.

Machine Learning

• Transformer based JUNO cosmogenic background suppression

基于Transformer的JUNO宇生本底压低策略

Transformer underpins GPT-4 and ParT. The later is a jet tagging architecture used by CMS/ATLAS and outperforms other machine learning algorithms. Cosmogenic backgrounds are important backgrounds of several physics targets, including solar neutrinos, DSNB, atmohpheric neutrinos, invisible decays etc. It is expected that Transformer may further improve the signal to background ratio and benefit the physics targets.

• Systematic uncertainty of ML based algorithms

Machine learning based algorithms outperforms traditional methods, yet lack of the interpretability makes it vulnerable to systematic uncertainties. Depending on the strategies to train the models, factors impacting the precisoin of models vary. A systematic review of factors contributing to systematic uncertainties of ML algorthms is thus necessary.

Fun

• Literature review with LLaMA and Alpaca

利用LLaMA和Alpaca作文献调研

Literature review is critial in exploring new fields and looking for existing solutions to specific problems. Literature review usually takes a lot of time, because the search engine may not be smart enough to understand the user’s target. The large language models demonstrate ability to understand human’s language, and thus may outperform search engine and saves time of researchers.

• JUNO IP (Intellectual Property) made with Stable Diffusion

基于Stable Diffusion制作JUNO IP

IP is a concept used in marketing. It is useful to introduce the JUNO project and its scientific targets to the general public, and an IP would speed up such a process and make it more attractive. Stable diffusion is an AI based iamge generation model. It would be interesting to see how stable diffusion and JUNO can come together and become an hot IP.

• Online game: Design of Neutrino detector

在线游戏：中微子探测器设计

Design of neutrino detectors is complex, because its cross-section is too small. The volume, target, readout, and costs are all constraints. The neutrino fluxes, energies, and flavors, the distance between source and detectors, the depths, are all factors that need to be considered. Putting all these together, it is a puzzle, and thus can be a hard-core online game.