[学习]L1 Homotopy for L1 norm minimization

最新推荐文章于 2021-06-02 06:31:27 发布
最新推荐文章于 2021-06-02 06:31:27 发布
阅读量2.9k 收藏 4
点赞数
分类专栏： Convex Optimization
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/bestlinjiayin/article/details/20446229
版权

L1 Homotopy: A MATLAB Toolbox for Homotopy Algorithms in L1 Norm Minimization Problems

  Introduction:

 

This package is a collection of MATLAB routines for solving some L1 norm minimization problems using homotopy techniques. These problems are usually encountered in the recovery of sparse signals from linear incoherent measurements.

 

This package contains scripts for solving the following optimization problems:

 

·        Basis pursuit denoising (BPDN) / LASSO

·        Dantzig selector

·        L1 decoding, robust L1 decoding

·        Re-weighted L1-norm (iterative and adaptive reweighting)

 

In addition to solving these problems for any given set of parameters, we have some dynamic algorithms to update their solution when

 

·        Streaming signal recovery

·        New measurements are sequentially added to the system

·        The unknown signal varies over time and we get a new measurement vector

 

Download:

 

Homotopy package

·        Version 2.0 (zip) released June 2013

o   A single, versatile homotopy program that can solve a variety of dynamic updating problems.

o   We consider the following linear model of observations:

where  is a sparse vector of interest,  is the system matrix,  is the measurement vector, and

最低0.47元/天 解锁文章
bestlinjiayin
关注
  • 0
    点赞
  • 踩
  • 4
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
专栏目录
L1 minimization
Areslp的专栏
10-08 1908
Piecewise smooth signal x, D is the differential matrix.  Suppose x is corrupted by noise e, b is the observation b=x+e. Image denoise L1-minimization(Total Variation) is used to recovery x and prese
matlab meas 函数_L1_homotopy_v2.0 这厮压缩感知恢复算法中L1同伦优化的 matlab 238万源代码下载- www.pudn.com...
weixin_39796140的博客
12-30 308
文件名称: L1_homotopy_v2.0下载 收藏√ [5 4 3 2 1]开发工具: matlab文件大小: 975 KB上传时间: 2015-04-17下载次数: 4提 供 者: yang详细说明：这厮压缩感知恢复算法中L1同伦优化的算法-Zhesi restore compressed sensing algorithm Homotopy algorithm optimi...
L1-norm Minimization
06-20
一份介绍L1-norm Minimization很好的文档
matlab解决路径优化代码-L1-homotopy:与使用同伦的L1范数最小化相关的代码
05-23
matlab解决路径优化代码L1同伦 与使用同伦的L1范数最小化相关的代码 ％---------- L1同态软件包------------- + 创建人：Salman Asif @ Georgia Tech。 电子邮件： 代码网站： 和Github仓库： （假设这是） 2013年6月发布的2.0版 Previous versions: v1.1 release: July 2012 v1.0 release: April 2009 ％------------------------------------------ + 参考： - M. Salman Asif and Justin Romberg, "Sparse recovery of streaming signals using L1-homotopy," preprint available at http://users.ece.gatech.edu/~sasif/ and ??? - M. Salman Asif, "Dynamic compressive sensing: Sparse recovery alg
UA MATH567 高维统计专题1 稀疏信号及其恢复1 L0-norm minimization
一个不愿透露姓名的博客
02-05 773
UA MATH567 高维统计专题1 稀疏信号及其恢复1 L0-norm minimizationL0L^0L0-normL0L_0L0​-norm minimizationExhaustive SearchL0L_0L0​-norm minimization的性质L0L_0L0​-norm minimization是NP-hard问题 这个专题我们讨论sparse signal recovery，作为这个专题的开头，我们先简单介绍一下sparse vector的norm；熟悉DSP的同学应该比较清楚，用v
compressive sensing l1 norm
02-18
总结来说，这些资源提供了压缩感知领域的深度学习材料，包括L1范数的理论基础、优化方法以及实际应用。通过深入理解这些内容，我们可以掌握如何利用压缩感知技术在数据采集和信号处理中实现高效和节省资源的解决方案...
Combined Homotopy Method For Solving Nonconvex
02-16
有界区域上优化问题的凝聚动约束组合同伦算法，刘庆怀，王秀玉，对有界区域上优化问题,构造了凝聚动约束函数及组合同伦方程,证明了同伦路径的收敛性,数值算例表明此算法是可行和有效的
Homotopy Based Reinforcement Learning with Maximum Entropy for
01-03
Homotopy Based Reinforcement Learning with Maximum Entropy for Autonomous Air Combat_基于同伦的最大熵强化学习在自主空战中的应用.pdf
L1-norm Regularization
03-12
L1范数正则化，是用于解决目标函数的优化问题的一种方法。
homotopy/同伦方法解非线性方程matlab代码
02-18
homotopy过程是一致的 牛逼在利用积分的方法进行求取，不会出现迭代方法不能收敛的问题而且还可以绘制出积分路径便于比较 并配有相关文档说明
同伦算法matlab源代码
04-29
求解非线性方程组的同伦算法matlab源代码
l1_ls_matlab
09-02
l1范数优化matlab代码，l1_ls算法代码包
高维数据的统计挑战
11-12
高维数据的统计挑战 Modern applications of statistical theory and methods can involve extremely large data-sets, often with huge numbers of measurements on each of a comparatively small number of experimental units. New methodology and accompanying theory have emerged in response: the goal of this theme issue is to illustrate a number of these recent developments.
基于PCA算法的人脸识别（包含人脸库）
11-07
MATLAB编程的以PCA算法为基础的人脸识别，包含人脸库，经典的算法。
UA MATH567 高维统计专题1 稀疏信号及其恢复2 用L1-norm作为L0-norm的convex relexation
一个不愿透露姓名的博客
06-02 381
UA MATH567 高维统计专题1 稀疏信号及其恢复2 用L1-norm作为L0-norm的convex relexationL1L_1L1​-norm minimizationL1L_1L1​-norm是L0L_0L0​-norm的凸包络L1L_1L1​-norm minimization的full recovery 上一讲我们在无噪声的设定下讨论了稀疏信号的恢复，假设yyy是我们对稀疏信号的测量，y=Axoy=Ax_oy=Axo​，系数AAA已知，目标是从测量中还原出信号xox_oxo​，一种可行的操
NonLinear Regression & Lasso & Homotopy Method
weixin_42339449的博客
02-21 523
Part One : Coordinate Descent for Lasso The Lasso optimization problem can be formulated as w^∈argminw∈Rd∑i=1m(hw(xi)−yi)2+λ∥w∥1, \hat{w}\in{ argmin_{w\in R^{d}}\sum_{i=1}^{m}(h_{w}(x_{i})-y_{i})^{2}+...
l1-norm loss & l2-norm loss （l1范数和l2范数作为正则项的比较）
兔角与禅
04-23 7930
l1-norm loss & l2-norm loss （l1范数和l2范数作为正则项的比较） l1-norm 和 l2-norm是常见的模型优化过程中的正则化项，对应到线性回归的领域分别为lasso Regression和 Ridge Regression，也就是 lasso 回归（有的地方也叫套索回归）和岭回归（也叫脊回归）。在深度学习领域也用l1和l2范数做正则化处理。这里简要介绍...
SIS的应用文章
02-22
sure independence screening 的应用
MATLAB函数“l1_homotopy”的代码
最新发布
06-07
“l1_homotopy”函数是一个MATLAB程序包，用于求解一般L1最小化问题，其代码如下： ```matlab function [x,status,history] = l1_homotopy(A, b, lambda, opts) % l1_homotopy: Solve L1-regularized least squares problem % [x,status,history] = l1_homotopy(A, b, lambda, opts) % % Solves the problem % minimize || A*x - b ||_2^2 + lambda * || x ||_1 % using a homotopy method that tracks the solution path as lambda decreases % from infinity to lambda. The solution is returned in the vector x. % % The "status" output gives the status of the solver: % status = 0: the solution is known to be exact. % status = 1: the solution is known to be accurate to within the given % tolerance tol. % status = 2: the solution is accurate to within a certain tolerance, % but the algorithm did not converge. % status = 3: the algorithm did not converge and did not find a solution. % % The "history" output is a structure containing the following fields: % iter: the iteration number % obj: the objective function value at each iteration % gap: the duality gap at each iteration % time: the time elapsed at each iteration % % The input "opts" is a structure that can contain the following fields: % tol: the convergence tolerance (default: 1e-6) % maxiter: the maximum number of iterations (default: 1000) % x0: an initial guess for x (default: all zeros) % print: whether to print progress information (default: 0) % update: how often to update the progress information (default: 10) % stopcond: the stopping condition to use (default: 3) % 1 = || x_k - x_k-1 ||_2 / || x_k ||_2 < tol % 2 = || x_k - x_k-1 ||_2 < tol % 3 = duality gap < tol % % Example usage: % n = 1000; p = 200; % A = randn(p,n); x0 = sprandn(n,1,0.1); b = A*x0 + 0.01*randn(p,1); % lambda = 0.1; % [x,status,history] = l1_homotopy(A, b, lambda); % % Written by: Justin Romberg, Caltech % Email: jrom@acm.caltech.edu % Created: February 2007 t0 = tic; % Set default options if nargin < 4 opts = struct(); end if ~isfield(opts, 'tol') opts.tol = 1e-6; end if ~isfield(opts, 'maxiter') opts.maxiter = 1000; end if ~isfield(opts, 'x0') opts.x0 = []; end if ~isfield(opts, 'print') opts.print = 0; end if ~isfield(opts, 'update') opts.update = 10; end if ~isfield(opts, 'stopcond') opts.stopcond = 3; end % Initialize solver [m,n] = size(A); x = zeros(n,1); if ~isempty(opts.x0) x = opts.x0; end w = x; v = zeros(m,1); gamma = 0; delta = 0; obj = 0.5 * norm(b - A*x)^2; gap = 0; iter = 0; history.iter = iter; history.obj = obj; history.gap = gap; history.time = 0; status = 3; % Compute initial values of quantities Atb = A'*b; AtA = A'*A; norm_w = norm(w); norm_v = norm(v); % Main loop while iter < opts.maxiter iter = iter + 1; % Update w and v w_old = w; v_old = v; [w,v,gamma,delta] = l1_homotopy_update(A, b, Atb, AtA, w, v, gamma, delta, lambda); % Compute objective value and duality gap obj_old = obj; obj = 0.5 * norm(b - A*x)^2 + lambda * norm(w,1); gap = abs(obj - gamma); % Update x x_old = x; x = w; % Compute convergence criteria if opts.stopcond == 1 crit = norm(x - x_old, 2) / norm(x, 2); elseif opts.stopcond == 2 crit = norm(x - x_old, 2); elseif opts.stopcond == 3 crit = gap; end % Save history information history.iter(iter) = iter; history.obj(iter) = obj; history.gap(iter) = gap; history.time(iter) = toc(t0); % Print progress information if opts.print && (mod(iter, opts.update) == 0 || iter == 1 || iter == opts.maxiter) fprintf('Iteration %d: obj = %0.6g, gap = %0.6g, crit = %0.6g\n', iter, obj, gap, crit); end % Check stopping criteria if crit < opts.tol status = 0; break; end end % Set status and final history information if status == 3 if gap < opts.tol status = 1; else status = 2; end end history.iter = history.iter(1:iter); history.obj = history.obj(1:iter); history.gap = history.gap(1:iter); history.time = history.time(1:iter); ``` 这个函数的输入包括：矩阵A和向量b（表示L1最小化问题），正则化参数lambda，以及一些选项opts（包括收敛容差、最大迭代次数、初始猜测、输出选项等）。输出包括：解向量x、状态status和历史记录history。函数中使用了一个“l1_homotopy_update”函数来更新w、v、gamma和delta等变量。主要的求解过程在主循环中进行。

“相关推荐”对你有帮助么？

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值