搜索
Since R2024b, a Levenberg–Marquardt solver (TrainingOptionsLM) was introduced. The built‑in function trainnet now accepts training options via the trainingOptions function (https://www.mathworks.com/help/deeplearning/ref/trainingoptions.html#bu59f0q-2) and supports the LM algorithm. I have been curious how to use it in deep learning, and the official documentation has not provided a concrete usage example so far. Below I give a simple example to illustrate how to use this LM algorithm to optimize a small number of learnable parameters.
For example, consider the nonlinear function:
y_hat = @(a,t) a(1)*(t/100) + a(2)*(t/100).^2 + a(3)*(t/100).^3 + a(4)*(t/100).^4;
It represents a curve. Given 100 matching points (t → y_hat), we want to use least squares to estimate the four parameters a1–a4.
t = (1:100)';
y_hat = @(a,t)a(1)*(t/100) + a(2)*(t/100).^2 + a(3)*(t/100).^3 + a(4)*(t/100).^4;
x_true = [ 20 ; 10 ; 1 ; 50 ];
y_true = y_hat(x_true,t);
plot(t,y_true,'o-')
- Using the traditional lsqcurvefit-wrapped "Levenberg–Marquardt" algorithm:
x_guess = [ 5 ; 2 ; 0.2 ; -10 ];
options = optimoptions("lsqcurvefit",Algorithm="levenberg-marquardt",MaxFunctionEvaluations=800);
[x,resnorm,residual,exitflag] = lsqcurvefit(y_hat,x_guess,t,y_true,-50*ones(4,1),60*ones(4,1),options);
x,resnorm,exitflag
- Using the deep-learning-wrapped "Levenberg–Marquardt" algorithm:
options = trainingOptions("lm", ...
InitialDampingFactor=0.002, ...
MaxDampingFactor=1e9, ...
DampingIncreaseFactor=12, ...
DampingDecreaseFactor=0.2,...
GradientTolerance=1e-6, ...
StepTolerance=1e-6,...
Plots="training-progress");
numFeatures = 1;
layers = [featureInputLayer(numFeatures,'Name','input')
fitCurveLayer(Name='fitCurve')];
net = dlnetwork(layers);
XData = dlarray(t);
YData = dlarray(y_true);
netTrained = trainnet(XData,YData,net,"mse",options);
netTrained.Layers(2)
classdef fitCurveLayer < nnet.layer.Layer ...
& nnet.layer.Acceleratable
% Example custom SReLU layer.
properties (Learnable)
% Layer learnable parameters
a1
a2
a3
a4
end
methods
function layer = fitCurveLayer(args)
arguments
args.Name = "lm_fit";
end
% Set layer name.
layer.Name = args.Name;
% Set layer description.
layer.Description = "fit curve layer";
end
function layer = initialize(layer,~)
% layer = initialize(layer,layout) initializes the layer
% learnable parameters using the specified input layout.
if isempty(layer.a1)
layer.a1 = rand();
end
if isempty(layer.a2)
layer.a2 = rand();
end
if isempty(layer.a3)
layer.a3 = rand();
end
if isempty(layer.a4)
layer.a4 = rand();
end
end
function Y = predict(layer, X)
% Y = predict(layer, X) forwards the input data X through the
% layer and outputs the result Y.
% Y = layer.a1.*exp(-X./layer.a2) + layer.a3.*X.*exp(-X./layer.a4);
Y = layer.a1*(X/100) + layer.a2*(X/100).^2 + layer.a3*(X/100).^3 + layer.a4*(X/100).^4;
end
end
end
The network is very simple — only the fitCurveLayer defines the learnable parameters a1–a4. I observed that the output values are very close to those from lsqcurvefit.
In our large open-source MATLAB Central community, there are many long-term excellent user groups. I really want to know why you have been using MATLAB for a long time, and what are its absolute advantages?
I have been using MATLAB for a long time, and there are several reasons for that:
- Fast ramp-up in unfamiliar domains: When I explore an unfamiliar application area or a new topic, MATLAB helps me quickly locate the canonical methods and example workflows. Its comprehensive, professional documentation — along with the related-topic links typically provided at the end of each page — makes it easy to get started intuitively and saves a lot of time that would otherwise be spent hunting for foundational knowledge across the web.
- A relatively cutting-edge yet reliable technical path: MATLAB’s many toolboxes evolve with the field. While updates aren’t always absolutely bleeding-edge, they generally offer approaches that balance modernity and proven reliability. This reduces the risk of wasting time on obscure or unstable algorithms and helps me follow a pragmatic, well-tested technical direction.
- Strong community and technical support: When I encounter a problem I first post on forums like MATLAB Answers and thoroughly investigate the issue myself. If I find a solution, I publish it to contribute back — which deepens my own understanding and helps others. If I can’t solve it alone, experienced community members often respond within hours. As a last resort, MathWorks’ official support is available and typically conducts an in-depth investigation into specific cases to help resolve the issue.
- ......
Also, most individuals have limited time and technical bandwidth, diving deeply into a single, narrow area can be hard to pull back from unless you are committed to that specific direction. For cutting‑edge, highly specialized research it’s often necessary to combine MATLAB with other languages (e.g., Python, C/C++) to go further.
I’m quite curious as to why this particular email was sent directly to my personal inbox. I have never actively subscribed to any online or offline training services nor clicked on any related marketing links. Could it be that because I frequently visit the official MATLAB forums, someone has identified me as a potential customer for their targeted promotions?
I’d love to hear your thoughts and start a discussion on this!
Hey MATLAB enthusiasts!
I just stumbled upon this hilariously effective GitHub repo for image deformation using Moving Least Squares (MLS)—and it’s pure gold for anyone who loves playing with pixels! 🎨✨
- Real-Time Magic ✨
- Precomputes weights and deformation data upfront, making it blazing fast for interactive edits. Drag control points and watch the image warp like rubber! (2)
- Supports affine, similarity, and rigid deformations—because why settle for one flavor of chaos?
- Single-File Simplicity 🧩
- All packed into one clean MATLAB class (mlsImageWarp.m).
- Endless Fun Use Cases 🤹
- Turn your pet’s photo into a Picasso painting.
- "Fix" your friend’s smile... aggressively.
- Animate static images with silly deformations (1).
Try the Demo!
I rarely use MATLAB.
10%
use MATLAB almost every day.
55%
use MATLAB once every 2-3 days.
10%
only use when specific task require
25%
20 个投票
After waiting for a long time, the MathWorks official Community has finally resumed some of its functionalitys! Congratulations! Next, I’d like to share some thoughts to help prevent such outages from happening again, as they have affected far too many people.
- Almost all resources rely solely on MathWorks servers. Once a failure (or a ransomware attack) occurs, everything is paralyzed, and there isn’t even a temporary backup server? For a big company like MathWorks to have no contingency plan at all is eye-opening. This tells us that we should have our own temporary emergency servers!
- The impact should be minimized. For example, many users need to connect to the official servers to download various support packages, such as the “Deep Learning Toolbox Converter for ONNX Model Format.” Could these be backed up and mirrored to the “releases” section of a GitHub repository, so users in need can download them.
- A large proportion of users who have already installed MATLAB cannot access the online help documentation. Since R2023a, installing the help documentation locally has become optional. This only increases the burden on the servers？ Moreover, the official website only hosts documentation for the past five years. That means after 2028, if I haven’t installed the local offline documentation, I won’t be able to access the online documentation for R2023a anymore?
Anything else you’d like to add? Feel free to leave a comment.
What is MATLAB Project?
40%
Never use it
28%
Only use existing from others' proj
3%
Use it occasionally
13%
Use it frequently
16%
90 个投票
Provide insightful answers
9%
Provide label-AI answer
9%
Provide answer by both AI and human
21%
Do not use AI for answers
46%
Give a button "chat with copilot"
10%
use AI to draft better qustions
5%
1561 个投票
%% 清理环境
close all; clear; clc;
%% 模拟时间序列
t = linspace(0,12,200); % 时间从 0 到 12，分 200 个点
% 下面构造一些模拟的"峰状"数据，用于演示
% 你可以根据需要替换成自己的真实数据
rng(0); % 固定随机种子，方便复现
baseIntensity = -20; % 强度基线（z 轴的最低值）
numSamples = 5; % 样本数量
yOffsets = linspace(20,140,numSamples); % 不同样本在 y 轴上的偏移
colors = [ ...
0.8 0.2 0.2; % 红
0.2 0.8 0.2; % 绿
0.2 0.2 0.8; % 蓝
0.9 0.7 0.2; % 金黄
0.6 0.4 0.7]; % 紫
% 构造一些带多个峰的模拟数据
dataMatrix = zeros(numSamples, length(t));
for i = 1:numSamples
% 随机峰参数
peakPositions = randperm(length(t),3); % 三个峰位置
intensities = zeros(size(t));
for pk = 1:3
center = peakPositions(pk);
width = 10 + 10*rand; % 峰宽
height = 100 + 50*rand; % 峰高
% 高斯峰
intensities = intensities + height*exp(-((1:length(t))-center).^2/(2*width^2));
end
% 再加一些小随机扰动
intensities = intensities + 10*randn(size(t));
dataMatrix(i,:) = intensities;
end
%% 开始绘图
figure('Color','w','Position',[100 100 800 600],'Theme','light');
hold on; box on; grid on;
for i = 1:numSamples
% 构造 fill3 的多边形顶点
xPatch = [t, fliplr(t)];
yPatch = [yOffsets(i)*ones(size(t)), fliplr(yOffsets(i)*ones(size(t)))];
zPatch = [dataMatrix(i,:), baseIntensity*ones(size(t))];
% 使用 fill3 填充面积
hFill = fill3(xPatch, yPatch, zPatch, colors(i,:));
set(hFill,'FaceAlpha',0.8,'EdgeColor','none'); % 调整透明度、去除边框
% 在每条曲线尾部标注 Sample i
text(t(end)+0.3, yOffsets(i), dataMatrix(i,end), ...
['Sample ' num2str(i)], 'FontSize',10, ...
'HorizontalAlignment','left','VerticalAlignment','middle');
end
%% 坐标轴与视角设置
xlim([0 12]);
ylim([0 160]);
zlim([-20 350]);
xlabel('Time (sec)','FontWeight','bold');
ylabel('Frequency (Hz)','FontWeight','bold');
zlabel('Intensity','FontWeight','bold');
% 设置刻度（根据需要微调）
set(gca,'XTick',0:2:12, ...
'YTick',0:40:160, ...
'ZTick',-20:40:200);
% 设置视角（az = 水平旋转，el = 垂直旋转）
view([211 21]);
% 让三维坐标轴在后方
set(gca,'Projection','perspective');
% 如果想去掉默认的坐标轴线，也可以尝试
% set(gca,'BoxStyle','full','LineWidth',1.2);
%% 可选：在后方添加一个浅色网格平面 (示例)
% 这个与题图右上方的网格类似
[Xplane,Yplane] = meshgrid([0 12],[0 160]);
Zplane = baseIntensity*ones(size(Xplane)); % 在 Z = -20 处画一个竖直面的框
surf(Xplane, Yplane, Zplane, ...
'FaceColor',[0.95 0.95 0.9], ...
'EdgeColor','k','FaceAlpha',0.3);
%% 进一步美化（可根据需求调整）
title('3D Stacked Plot Example','FontSize',12);
constantplane("x",12,FaceColor=rand(1,3),FaceAlpha=0.5);
constantplane("y",0,FaceColor=rand(1,3),FaceAlpha=0.5);
constantplane("z",-19,FaceColor=rand(1,3),FaceAlpha=0.5);
hold off;
Have fun! Enjoy yourself!
It is time to support the cameraIntrinsics function to accept a 3-by-3 intrinsic matrix K as an input parameter for constructing the object. Currently, the built-in cameraIntrinsics function can only be constructed by explicitly specifying focalLength, principalPoint, and imageSize. This approach has drawbacks, as it is not very intuitive. In most application scenarios, using the intrinsic matrix
K=[fx,0,cx;
0,fy,cy;
0,0,1]
is much more straightforward and effective!
intrinsics = cameraIntrinsics(K)
Check out the result of "emoji matrix" multiplication below.
- vector multiply vector:
a = ["😁","😁","😁"]
b = ["😂";
"😂"
"😂"]
c = a*b
d = b*a
- matrix multiply matrix:
matrix1 = [
"😀", "😃";
"😄", "😁"]
matrix2 = [
"😆", "😅";
"😂", "🤣"]
resutl = matrix1*matrix2
enjoy yourself!
Since May 2023, MathWorks officially introduced the new Community API(MATLAB Central Interface for MATLAB), which supports both MATLAB and Node.js languages, allowing users to programmatically access data from MATLAB Answers, File Exchange, Blogs, Cody, Highlights, and Contests.
I’m curious about what interesting things people generally do with this API. Could you share some of your successful or interesting experiences? For example, retrieving popular Q&A topics within a certain time frame through the API and displaying them in a chart.
If you have any specific examples or ideas in mind, feel free to share!
MATLAB FEX(MATLAB File Exchange) should support Markdown syntax for writing. In recent years, many open-source community documentation platforms, such as GitHub, have generally supported Markdown. MATLAB is also gradually improving its support for Markdown syntax. However, when directly uploading files to the MATLAB FEX community and preparing to write an overview, the outdated document format buttons are still present. Even when directly uploading a Markdown document, it cannot be rendered. We hope the community can support Markdown syntax!
BTW，I know that open-source Markdown writing on GitHub and linking to MATLAB FEX is feasible, but this is a workaround. It would be even better if direct native support were available.
Currently, according to the official documentation, "DisplayName" only supports character vectors or single scalar string as input. For example, when plotting three variables simultaneously, if I use a single scalar string as input, the legend labels will all be the same. To have different labels, I need to specify them separately using the legend function with label1, label2, label3.
Here's an example illustrating the issue:
x = (1:10)';
y1 = x;
y2 = x.^2;
y3 = x.^3;
% Plotting with a string scalar for DisplayName
figure;
plot(x, [y1,y2,y3], DisplayName="y = x");
legend;
% To have different labels, I need to use the legend function separately
figure;
plot(x, [y1,y2,y3], DisplayName=["y = x","y = x^2","y=x^3"]);
% legend("y = x","y = x^2","y=x^3");
In the past two years, large language models have brought us significant changes, leading to the emergence of programming tools such as GitHub Copilot, Tabnine, Kite, CodeGPT, Replit, Cursor, and many others. Most of these tools support code writing by providing auto-completion, prompts, and suggestions, and they can be easily integrated with various IDEs.
As far as I know, aside from the MATLAB-VSCode/MatGPT plugin, MATLAB lacks such AI assistant plugins for its native MATLAB-Desktop, although it can leverage other third-party plugins for intelligent programming assistance. There is hope for a native tool of this kind to be built-in.
What is the side-effect of counting the number of Deep Learning Toolbox™ updates in the last 5 years? The industry has slowly stabilised and matured, so updates have slowed down in the last 1 year, and there has been no exponential growth.Is it correct to assume that? Let's see what you think!
releaseNumNames = "R"+string(2019:2024)+["a";"b"];
releaseNumNames = releaseNumNames(:);
numReleaseNotes = [10,14,27,39,38,43,53,52,55,57,46,46];
exampleNums = [nan,nan,nan,nan,nan,nan,40,24,22,31,24,38];
bar(releaseNumNames,[numReleaseNotes;exampleNums]')
legend(["#release notes","#new/update examples"],Location="northwest")
title("Number of Deep Learning Toolbox™ update items in the last 5 years")
ylabel("#release notes")
Create a struct arrays where each struct has field names "a," "b," and "c," which store different types of data. What efficient methods do you have to assign values from individual variables "a," "b," and "c" to each struct element? Here are five methods I've provided, listed in order of decreasing efficiency. What do you think?
Create an array of 10,000 structures, each containing each of the elements corresponding to the a,b,c variables.
num = 10000;
a = (1:num)';
b = string(a);
c = rand(3,3,num);
Here are the methods;
%% method1
t1 =tic;
s = struct("a",[], ...
"b",[], ...
"c",[]);
s1 = repmat(s,num,1);
for i = 1:num
s1(i).a = a(i);
s1(i).b = b(i);
s1(i).c = c(:,:,i);
end
t1 = toc(t1);
%% method2
t2 =tic;
for i = num:-1:1
s2(i).a = a(i);
s2(i).b = b(i);
s2(i).c = c(:,:,i);
end
t2 = toc(t2);
%% method3
t3 =tic;
for i = 1:num
s3(i).a = a(i);
s3(i).b = b(i);
s3(i).c = c(:,:,i);
end
t3 = toc(t3);
%% method4
t4 =tic;
ct = permute(c,[3,2,1]);
t = table(a,b,ct);
s4 = table2struct(t);
t4 = toc(t4);
%% method5
t5 =tic;
s5 = struct("a",num2cell(a),...
"b",num2cell(b),...
"c",squeeze(mat2cell(c,3,3,ones(num,1))));
t5 = toc(t5);
%% plot
bar([t1,t2,t3,t4,t5])
xtickformat('method %g')
ylabel("time(second)")
yline(mean([t1,t2,t3,t4,t5]))
As far as I know, starting from MATLAB R2024b, the documentation is defaulted to be accessed online. However, the problem is that every time I open the official online documentation through my browser, it defaults or forcibly redirects to the documentation hosted site for my current geographic location, often with multiple pop-up reminders, which is very annoying!
Suggestion: Could there be an option to set preferences linked to my personal account so that the documentation defaults to my chosen language preference without having to deal with “forced reminders” or “forced redirection” based on my geographic location? I prefer reading the English documentation, but the website automatically redirects me to the Chinese documentation due to my geolocation, which is quite frustrating!
----------------2024.12.13 update-----------------
Although the above issue was resolved by technical support, subsequent redirects are still causing severe delays...
In the past two years, MATHWORKS has updated the image viewer and audio viewer, giving them a more modern interface with features like play, pause, fast forward, and some interactive tools that are more commonly found in typical third-party players. However, the video player has not seen any updates. For instance, the Video Viewer or vision.VideoPlayer could benefit from a more modern player interface. Perhaps I haven't found a suitable built-in player yet. It would be great if there were support for custom image processing and audio processing algorithms that could be played in a more modern interface in real time.
Additionally, I found it quite challenging to develop a modern video player from scratch in App Designer.（If there's a video component for that that would be great）
-----------------------------------------------------------------------------------------------------------------
BTW，the following picture shows the built-in function uihtml function showing a more modern playback interface with controls for play, pause and so on. But can not add real-time image processing algorithms within it.
As far as I know, the MATLAB Community (including Matlab Central and Mathworks' official GitHub repository) has always been a vibrant and diverse professional and amateur community of MATLAB users from various fields globally. Being a part of it myself, especially in recent years, I have not only benefited continuously from the community but also tried to give back by helping other users in need.
I am a senior MATLAB user from Shenzhen, China, and I have a deep passion for MATLAB, applying it in various scenarios. Due to the less than ideal job market in my current social environment, I am hoping to find a position for remote support work within the Matlab Community. I wonder if this is realistic. For instance, Mathworks has been open-sourcing many repositories in recent years, especially in the field of deep learning with typical applications across industries. I am eager to use the latest MATLAB features to implement state-of-the-art algorithms. Additionally, I occasionally contribute through GitHub issues and pull requests.
In conclusion, I am looking forward to the opportunity to formally join the Matlab Community in a remote support role, dedicating more energy to giving back to the community and making the world a better place! (If a Mathworks employer can contact me, all the better~)
