28 / 28
2026 8
我在生物医学研究里怎么用 AI 整理自我的分享“AI for Biomedicine”：我怎么用 AI 规划、跑实验和写报告，怎么用它做研究工具和 slides，以及为什么分数好的模型，到了医院不一定有用。 How I work with AI in biomedical research From my talk AI for Biomedicine: how I plan, run and report experiments with AI, build research tools and slides with it, and why a model that scores well is not yet a model that helps in a hospital. Predicting severe COVID-19 around admission from a CT scan, the chart, and a blood test Models that combine chest CT, admission records and lab results, using patient data collected from 27 December 2019 to 31 March 2020, to flag which patients were likely to need intensive care, a ventilator, or not survive. 入院时用 CT、病历和化验预测新冠重症 这些模型结合胸部 CT、入院记录和化验结果，使用 2019 年 12 月 27 日至 2020 年 3 月 31 日期间收集的患者数据，用来标出哪些患者可能需要重症监护、需要呼吸机，或者可能无法存活。 Teaching a model to redraw lung tumors, then reusing what it learned A 3D variational autoencoder learned lung lesion features from CT scans without outcome labels, and those features held up against radiomics for predicting clinical outcomes. 让模型学会重画肺部病灶，再把学到的东西拿去用 一个 3D 变分自编码器在不使用结局标签的情况下，从 CT 影像中学到了肺部病灶的特征。在预测临床结局时，这些特征的表现和影像组学相当。 Eight chest X-ray models, nine datasets, and what one score leaves out We ran eight published chest X-ray AI models on six public and three private datasets, and found that which model looks best depends on the findings you count, the metric you use and the patients you test on. 八个胸片模型、九个数据集，以及一个分数看不到的东西 我们在六个公开数据集和三个私有数据集上运行了八个已发表的胸片 AI 模型，发现哪个模型看起来最好，取决于统计哪些征象、用什么指标，以及在哪些患者身上测试。
2021 3
Brief Walkthrough of Famous ImageNet Contenders Briefly introduce the most famous/widely used image nets over the years Using HDF5 format for python file saving and loading Numpy data saving and loading with its advantages LSTM Walk Through Brief walk-through of LSTM, inspired by a great post and a great lecture.
2020 17
Notes About the Logics Behind the Development of Tree-Based Models A summary following the development of tree-based methods, good for understanding the reasoning Projects Archive Briefly describe as well as record some of my course projects in Stanford Complex Heatmap, flexible package for heatmap based on R Example based instructions for using ComplexHeatmap for heatmaps Pyradiomics Simple Usage Brief for quick start in pyradiomics package Further Discussion of Relative Importance Discussed two more methods of estimating relative importance of linear regression features. Commonality Analysis and Dominance analysis. Multivariate Linear Regression -- Collinearity and Feature Importance How to perform linear regression and addressing the problem of collinearity and find feature importance. CS229 Problem Sets Spring 2020 My answers to the problem sets for Stanford CS229 course in Spring 2020 Survival Analysis Case Study Using R Survival Analysis Case Study Using R Survival Analysis -- the Basics The basics of survival analysis, why do we need it and how does it work EM Algorithm Notes Notes of GMM, and it's generalized form EM algorithm Kernel Method Note Note down some facts about kernel method. Generative Models -- Gaussian Discriminant Analysis The differences of generative models and discriminative models. How are they related. Logistic Regression Updated with Newton's Method Insight of logistic regression and Newton's method Store Almost Any Objects of Python in Files Cache outputs of time consuming code results Probabilistic Interpretation of Sum of Square Loss Function One way to interpret the form of "square" in square loss. Positive Semidefinite Matrix in Machine Learning Positive Semidefinite Matrix is closely associated with some of machine learning algorithms. SMOTE with Python An over-sampling method, inspired by KNN, to deal with imbalanced data.