mike + cryptography   14

Encrypt your Machine Learning – corti.ai – Medium
Assume your customers are unable to give you their data for privacy or security reasons. Which means, if you want to apply your models on their data, you have to bring the model to them. But if sharing your valuable model is impossible or you are limited by privacy concerns, encrypting your model might be an option. You can train your model, encrypt it, and send it to your customers. In order for the customer to actually use the prediction, you have to provide them with a decryption service.

The first fully homomorphic algorithm was incredibly slow, taking 100 trillion times as long to perform calculations of encrypted data than plaintext analysis.⁶

IBM has sped things up considerably, making calculations on a 16-core server over two million times faster than past systems.

For the smallest parameter set, the time required for a homomorphic multiplication of ciphertexts was measured to be 3.461 milliseconds.

Computational requirements are not the only concern — we also have to consider the size of the encrypted data or model.

In conclusion, the encrypted data is one to three orders of magnitude larger than the unencrypted data. The exact factor depends on what is considered a natural representation of the data in its raw form.
machinelearning  cryptography  security  federatedlearning 
7 weeks ago by mike
Critique My Plan: API Key for Authentication | Hacker News
We hash passwords because passwords are valuable across sites; it's a big deal to compromise someone's password, even on a random low-value application. That's not true of API keys. If your database is compromised, the API keys don't matter anymore. Don't bother encrypting them.
programming  security  cryptography 
april 2018 by mike
A (Relatively Easy To Understand) Primer on Elliptic Curve Cryptography
The gap between the difficulty of factoring large numbers and multiplying large numbers is shrinking as the number (i.e. the key's bit length) gets larger. As the resources available to decrypt numbers increase, the size of the keys need to grow even faster. This is not a sustainable situation for mobile and low-powered devices that have limited computational power. The gap between factoring and multiplying is not sustainable in the long term.

All this means is that RSA is not the ideal system for the future of cryptography. In an ideal Trapdoor Function, the easy way and the hard way get harder at the same rate with respect to the size of the numbers in question. We need a public key system based on a better Trapdoor.

The gap between the difficulty of factoring large numbers and multiplying large numbers is shrinking as the number (i.e. the key's bit length) gets larger. As the resources available to decrypt numbers increase, the size of the keys need to grow even faster. This is not a sustainable situation for mobile and low-powered devices that have limited computational power. The gap between factoring and multiplying is not sustainable in the long term.

The elliptic curve discrete logarithm is the hard problem underpinning elliptic curve cryptography. Despite almost three decades of research, mathematicians still haven't found an algorithm to solve this problem that improves upon the naive approach. In other words, unlike with factoring, based on currently understood mathematics there doesn't appear to be a shortcut that is narrowing the gap in a Trapdoor Function based around this problem. This means that for numbers of the same size, solving elliptic curve discrete logarithms is significantly harder than factoring.

To visualize how much harder it is to break, Lenstra recently introduced the concept of "Global Security." You can compute how much energy is needed to break a cryptographic algorithm, and compare that with how much water that energy could boil. This is a kind of cryptographic carbon footprint. By this measure, breaking a 228-bit RSA key requires less energy to than it takes to boil a teaspoon of water. Comparatively, breaking a 228-bit elliptic curve key requires enough energy to boil all the water on earth. For this level of security with RSA, you'd need a key with 2,380-bits.
cryptography 
april 2018 by mike
Build a Better Monster: Morality, Machine Learning, and Mass Surveillance
This is the text version of "Build a Better Monster: Morality, Machine Learning, and Mass Surveillance", a talk I gave on April 18, 2017, at the Emerging Technologies for the Enterprise conference in Philadelphia.
facebook  advertising  google  politics  Privacy  cryptography 
march 2018 by mike
Learning to protect communications with adversarial neural cryptography | the morning paper
The kind of network setup shown here is a general pattern for learning goals of the form in which we want to maximise performance in task A without permitting task B to be accomplished.
neuralnetworks  cryptography 
february 2017 by mike

Copy this bookmark:



description:


tags: