weixin_43269419的博客

概率密度 reweight

介绍 DDPM( Denoising Diffusion Probabilistic Model )的算法原理

Junction Tree Variational Autoencoder for Molecular Graph GenerationYear: 2018Authors: Wengong Jin, Regina Barzilay, Tommi JaakkolaJournal Name: ICMLInnovation

Variational Graph Auto-EncodersYear: 2016Authors: Thomas N. Kipf, Max WellingJournal Name: NIPSDefinitions定义含有 N=∣V∣N=|V|N=∣V∣ 个节点的无向图 G=(V,E)\mathcal{G}=(\mathcal{V}, \mathcal{E})G=(V,E) ，邻接矩阵 AAA 和度矩阵 DDD ，再引入 N×FN \times FN×F 大小的隐变量矩阵 Z=[z1;z2;...;

De Novo Prediction of RNA 3D Structures with Deep LearningYear: 2022Authors: Julius Ramakers, Christopher Frederik Blum, Sabrina K¨onig, Stefan Harmeling, Markus KollmannJournal Name: bioxiv1 Innovation结合自回归深度生成模型、蒙特卡罗树搜索和分数模型预测 RNA 三维折叠结构。2 Method

Physically informed artificial neural networks for atomistic modeling of materialsYear: 2019Authors: G. P. Purja Pun, R. Batra, R. Ramprasad & Y. MishinJournal Name: Nature CommunicationsInnovation扩展物理模型，使其具有广泛适用性，即在训练集覆盖不到的数据中也能适用。引入与局部结构参数 Gil

RNAcmap: a fully automatic pipeline for predicting contact maps of RNAs by evolutionary coupling analysisYear: 2021Authors: Tongchuan Zhang, Jaswinder Singh, Thomas Litfin, Jian Zhan， Kuldip Paliwal and Yaoqi ZhouJournal Name: BioinformaticsMotivation

Geometric deep learning of RNA structureYear: 2021Authors: Raphael J. L. Townshend, Stephan Eismann, Andrew M. Watkins, Ramya Rangan, Maria Karelina, Rhiju Das, Ron O. DrorJournal Name: ScienceDatasetBackgroundMethodEquivariant convolution等变卷积基于滤波

系列文章GFlowNet Foundation 笔记（一）GFlowNet Foundation 笔记（二）GFlowNet Foundation 笔记（三）GFlowNet Foundation 笔记（四）期望的奖励和奖励最大策略Def 37. 对于在终止状态上的任意分布 Pπ(s)P_{\pi}(s)Pπ​(s) ，期望奖励( expected reward ) 为VPπ(s)=EPπ(S)[R(S)∣S≥s]=∑s′≥sR(s′)Pπ(s′∣s≤s′)V_{P_{\pi}}(s) =

系列文章GFlowNet Foundation 笔记（一）GFlowNet Foundation 笔记（二）GFlowNet Foundation 笔记（三）确定环境和随机环境中的策略Def 34. 策略( policy ) π:A×S↦R\pi: \mathcal{A} \times \mathcal{S} \mapsto \Rπ:A×S↦R 为概率分布 π(a∣s)\pi (a | s)π(a∣s) 。其中，行动 a∈Aa \in \mathcal{A}a∈A ，定义 A(s)\mathcal

系列文章GFlowNet Foundation 笔记（一）GFlowNet Foundation 笔记（二）条件流与自由能Def 24. 已知自由能 F(s)\mathcal{F}(s)F(s)e−F(s)=∑s′:s′≥sR(s′)=∑s′:s′≥se−F(s′)e^{-\mathcal{F}(s)} = \sum_{s': s' \ge s} R(s') = \sum_{s': s' \ge s} e^{-\mathcal{F}(s')}e−F(s)=s′:s′≥s∑​R(s′)=s′:s

学习流量Def 18. GFlowNet 可用 (F^(s),P^F(st+1∣st))(\hat{F}(s), \hat{P}_F(s_{t+1} | s_t))(F^(s),P^F​(st+1​∣st​)) 表示。从终止流量估计转移概率终止流量对应终止奖励函数 RRRR(s)=F(s→sf)R(s) = F(s \rightarrow s_f)R(s)=F(s→sf​)推论3. 上面的式子可以推导出总流量Z=F(s0)=F(sf)=∑s∈Par(sf)R(s)Z = F(s_0) =

马尔可夫流量的测量轨迹( trajectories )和流量( flows )Def 1. 一个有向图用 (S,A)(\mathcal{S}, \mathbb{A})(S,A) 表示，其中 S\mathcal{S}S 为状态的集合， A\mathbb{A}A 为 S×S\mathcal{S} \times \mathcal{S}S×S 大小的有向边的子集。 A\mathbb{A}A 中的元素表示为 s→s′s \rightarrow s's→s′ ，叫做 边缘( edges ) 或 转移( transi

变分推断是为了近似获得 P(Z∣X)P(Z | X)P(Z∣X) ，即隐状态的后验分布。logP(X)=logP(X,Z)−logP(Z∣X)=logP(X,Z)q(Z)−logP(Z∣X)q(Z)\begin{aligned} log P(X) &= log P(X, Z) - log P(Z | X) \\ &= log \frac{P(X, Z)}{q(Z)} - log \frac{P(Z | X)}{q(Z)}\end{aligned}logP(X)​=logP(X,Z

花书 5.4.45.4.45.4.4 中说均方误差( MSE )度量着估计 θ^\hat{\theta}θ^ 和真实参数 θ\thetaθ 之间平方误差的总体期望偏差，但没有进行证明，现将推导过程展示在下面MSE=E[(θ^−θ)2]=E[((θ^−E(θ^))+(E(θ^)−θ))2]=E[(θ^−E(θ^))2]+E[(E(θ^)−θ)2]+2E[(θ^−E(θ^))(E(θ^)−θ)]=E[(θ^−E(θ^))2]+(E(θ^)−θ)2+2(E(θ^)−E(θ^))(E(θ^)−θ)=Var(θ^)

最近在看花书，其中 4.3 中提到了方向导数当 α=0\alpha = 0α=0 时， ∂∂αf(x+αu)=uT∇xf(x)\frac{\partial}{\partial \alpha}f(\textbf{x} + \alpha \textbf{u}) = \textbf{u}^T
abla_xf(\textbf{x})∂α∂​f(x+αu)=uT∇x​f(x) 是怎么得出的？根据全微分∂f(x+αu)=∂f(x1+αu1)∂x1⋅∂(x1+αu1)+⋯+∂f(xn+αun)∂xn⋅∂(xn+

系列文章《Graph Representation Learning》笔记 Chapter2《Graph Representation Learning》笔记 Chapter3《Graph Representation Learning》笔记 Chapter4《Graph Representation Learning》笔记 Chapter5《Graph Representation Learning》笔记 Chapter6传统方法概述我们可以将生成过程指定为计算 P(A[u,v]=1)P(A

系列文章《Graph Representation Learning》笔记 Chapter2《Graph Representation Learning》笔记 Chapter3《Graph Representation Learning》笔记 Chapter4《Graph Representation Learning》笔记 Chapter5目录Applications and Loss FunctionsGNNs for Node ClassificationGNNs for Graph Cla

系列文章《Graph Representation Learning》笔记 Chapter2《Graph Representation Learning》笔记 Chapter3《Graph Representation Learning》笔记 Chapter4目录Permutation invariance and equivarianceNeural Message PassingOverview of the Message Passing FrameworkPermutation invar

Basic Local Alignment Search ToolYear: 1990Authors: Stephen F. Altschul, Warren Gish, Webb Miller, Eugene W. Myers and David J. LipmanJournal Name: Journal of Molecular BiologyAbstract一个快速进行序列比较的新方法：基本局部比对搜索工具( basic local alignment search tool, BLAST