Marco Melis

[Marco Melis @MarcoMelisIT]

@pralab @pluribus_one #machinelearning #NeuralNetworks #advml #Malware #python #attacks

[Marco Melis @MarcoMelisIT]

This week we added new #pretrained #models for @secml in our zoo:
- Support Vector Machines and our Secure SVM from https://t.co/QhsXBZGGvy, on the Drebin #android ds
- #pytorch AlexNet + SVMs from https://t.co/ogQyXLJFgQ, on the #iCub ds

Check them out: https://t.co/7hXBYcFNim

[Marco Melis @MarcoMelisIT]

Despite current difficult times, work on our projects continues!

[secml @secml_py]

secml v0.14 is out!
https://t.co/j6MkKuk2IT
https://t.co/CnEDnkOjzj
https://t.co/UAizk3bEOb

Highlights:
- Foolbox attacks and RobustBench models now included, with notebooks
- New notebook tutorial with an application on Android Malware Detection

[Francesco Crecchi @FrancescoCrecch]

Our new paper “FADER: Fast Adversarial Example Detection” is out! Check it if you want to speed up adversarial examples detection! https://t.co/EUN3d60Dz9
@biggiobattista @MarcoMelisIT @sotgiu_angelo

[secml @secml_py]

New notebook on adversarial Android malware attacks by @MarcoMelisIT!

https://t.co/E0eB4egygp

Paper (w/ attack details): https://t.co/Pg7CuNBW77

With secml we can craft sparse attacks that only inject Boolean features - and explain the classification results.

Check it out!

[secml @secml_py]

v0.13 is out!
https://t.co/3yPY7H8cEK

with major improvements and the implementation of deep neural rejection (DNR) to counter adversarial examples!

Notebook: https://t.co/aEyGMZLedf
Paper: https://t.co/SMhwH6KYoM

[secml @secml_py]

v0.12 is out with many usability and efficiency improvements!
https://t.co/Mo7HjwzBsd

[Marco Melis @MarcoMelisIT]

New hotfix v0.11.2 for #SecML addressing a few issues with attack and optimization classes.

Also, check the new tutorial on advanced attacks using #imagenet and #neuralnetworks from our paper:
https://t.co/ThR6S28Ao6

https://t.co/HZzfdacQ6t

Changelog: https://t.co/APbPN6Ji3T

[arXiv Daily @Arxiv_Daily]

secml: A Python Library for Secure and Explainable Machine Learning
https://t.co/ZBZvpYhuo9
by Marco Melis et al. including @biggiobattista
#MachineLearning #NeuralNetwork

[Marco Melis @MarcoMelisIT]

We just released a hotfix for #SecML addressing compatibility issues with recently released #sklearn v0.22 and #scipy v1.4.

Changelog: https://t.co/APbPN6rGFj

[Marco Melis @MarcoMelisIT]

The list of available pre-trained models in the #modelzoo will be greatly expanded in the future. See https://t.co/HD0zePklPi for more details.

[Marco Melis @MarcoMelisIT]

We just released SecML v0.11, including the new #modelzoo, multiple performance optimizations, and general improvements.

Also check the new tutorial on the use of #cleverhans within SecML: https://t.co/oEdqSTgfAa

Code: https://t.co/77OqeXwJiV
Changelog: https://t.co/APbPN6Ji3T

[Marco Melis @MarcoMelisIT]

@railsimulator And abandoned since then!

[Marco Melis @MarcoMelisIT]

Check out the now available SecML v0.10 with #explainableML and #cleverhans library connector (beta). In addition, multiple enhancements to #pytorch support.

New tutorials available here: https://t.co/oEdqSTgfAa

Code: https://t.co/77OqeXwJiV
Changelog: https://t.co/APbPN6Ji3T

[Marco Melis @MarcoMelisIT]

The new v0.9 update for SecML is now available, which adds support to #NeuralNetworks through #PyTorch.

Many thanks for the positive feedback on the project so far.

Have a look at the new tutorials available here: https://t.co/HD0zePklPi

Code: https://t.co/77OqeXwJiV

[Marco Melis @MarcoMelisIT]

We just released a small update (v0.8.1) for SecML with a fix for documentation not building correctly. Thanks for the feedback.

To send us suggestions and bug reports please use the issue tracker: https://t.co/qXQTGbqYsv

Docs: https://t.co/oEdqSTgfAa

[Battista Biggio @biggiobattista]

We've finally released a beta version of secML, with poisoning and evasion attacks (also on sparse data). Notebooks are available.
Support for DL and other advML libraries will be added soon.
Feedback is welcome!

Code: https://t.co/OuUnVLeqzl
Docs: https://t.co/LoYCJSfzQ7

[Marco Melis @MarcoMelisIT]

Full P2 room for the #adversarialML tutorial at #DeepLearn2019 @PRA_Lab @pluribus_one #deeplearning #NeuralNetworks #MachineLearning

[Marco Melis @MarcoMelisIT]

@PRA_Lab on its way to @DeepLearn2019 #deeplearn2019 #Warsaw #DeepLearning #ArtificialIntelligence #MachineLearning https://t.co/CfHewVMkG0

[Marco Melis @MarcoMelisIT]

@biggiobattista @maiorcasecurity @GiorgioGiacinto

[Marco Melis @MarcoMelisIT]

Our recent paper at #EUSIPCO2018 showing that #ML algorithms overemphasise few features trying to detect #Android #malware, which explains why they are vulnerable and why attacks are transferable across different models. #interpretableML #explainableAI

https://t.co/jiJcyyTwmS