How to Protect your Python and R Machine Learning Code
Vložit
- čas přidán 28. 07. 2024
- What options do you have for distributing Python and R code and still keeping some control over the intellectual property (IP). In this video I look at some of the high level options, such as API's, Docker Images, Compiled code, and encryption. If there is interest I may follow this up with more technical low-level tutorials on how to protect code in Python.
0:22 What code should you not share?
0:55 Protecting machine learning code
1:16 What are the parts of a model deploy
1:40 Scoring code vs. training code
2:44 What about lookup tables and other data?
3:02 Deploy in the cloud or on the edge?
3:12 Deploying behind an API
4:28 Protecting access to the API
4:50 Authentication and Throttling
5:24 Advsarial Example Attack
6:20 Edge Deployment
7:50 Preventing copying and modification
8:00 Compiled languages
8:23 Does Docker offer protection?
9:50 Protecting Binary files
10:30 What about encryption?
Complete Eaglesoft (ESI), C64 Cracking Group Intro
• Eagle Soft Inc. [ESI] ...
Follow Jeff Heaton/Subscribe:
/ heatonresearch
github.com/jeffheaton
/ jeffheaton
Support Me on Patreon: / jeffheaton - Věda a technologie
Any interest in a video showing how to hide code in a Docker image? Maybe even a small competition to see how quickly someone can extract my code?
yes please. It would be also useful to learn how to defend model from adversarial and other type of attack techniques.
Do you have existing "how to" docker videos for python gpu learning ? I'm a relative newbie but have an nvidia gpu on a mac I want to get more out of.
Interested!
I do.
This channel is gold and Jeff’s book is even better. Thank you for sharing your knowledge
Always, a joyful and fruitful time is the time that we take to listen to your splendid videos.
The question of security at the edge is really tough. Sometimes it's impractical or too time consuming to rewrite the whole Py code into a compiled language, but... you may rewrite core logic/functions and import them as binary modules to your main python code... just like Numpy and other similar libraries are written in C++ do
Such a great level of content!!!
Thanks!
This is very helpful. I have been scratching my head on how I can do this which is similar to you know when you package / compile your vb codes.
It's amazing how you can stay this awesome this long
You'll be surprised to know that even giants like Google makes the mistake of exposing their tflite models in some of their Android apps without protection.
Great video!
Thanks!
Good one
There are different possibilities to compile the lython codes But Is there any way to protect/hide the trained model files like .pt,.hdf5etc
Could you create more videos about this topic?
How about use R-Shiny package for R and Provide a simple Frontend. No Code or coefficients visible. The User Just Uploads Data and clicks Buttons. P.S. pretty cool you cracked Games!
Hey Jeff, is there any way to swap clothes from lean body to fat one in the same way it happens on e-commerce sites
I have not specifically tried that, but I am guessing some sort of adjustment of GAN latent space.
Oh my goodness! would U listen 2 this!! woohoo!! U'r gett'n in2 the mud bruh!! yup!! piracy, is like a bag of chips! can a pirate, eat just 1?? sure!!
great vid!! super important!! NN prog's, could be the next baby!! good luck....