minibatchpredict memory leak bug?

basically as the title says. not sure whether this is intended.

the problem can be reproduced using the code below.

"net_2btest.mat" can be obtained here

https://github.com/TFEI-Terminal/minibatchpredict-demo/blob/83e961be22ac7c9f1109a0e22c71ed20d79c0812/net_2btest.mat

===============

load('net_2btest.mat','net');
a = rand(101,1,100);
for i = 1 : 10000000
    if ~rem(i,100)
        i
        memory 
    end
    b = dlarray(a,"TCB");
    c = minibatchpredict(net,b);
end

===============

if you keep attention on memory. i.e., Memory used by MATLAB: xxxx MB

you`d find every minibatchpredict call uses ~500kb-1mb more memory.

Answer 1

[Abhiram]

Hi @Felix,

I have noticed a similar behaviour with the ‘minibatchpredict’ function when working with MATLAB R2024b. Upgrading the MATLAB version to R2025a Prerelease has solved the issue.

Hope this helps!