When Randomness meets Automation- The Intersection of Cryptography with Machine Learning

Author- Anuj Srivastava

Here is a brief summary, or rather the cornerstone of what Dr. Arpita Patra, a renowned faculty and esteemed researcher of IISc Bangalore had rendered us in her motivating lecture on 26/06/19 along with some farrago of my outlook. The 16th day of ACM India summer school(IIIT Bangalore).

We are talking about two stalwarts of Computer Science. Yes! the Machine Learning and the Cryptography. With the unprecedented use of machine learning in almost every domain like facial recognition, medical, finance, banking, etc. So, machine learning is touching our lives and making it better. But recall one thing, an ML algorithm can build a very good model only when it is fed with a sufficiently huge amount of data. Otherwise, it will be not suitably accurate. But the question is from where do you get the data? Data is the private property of individuals or organizations. It is often said that "Data is the new Currency" They generally don’t like to share. So, its where the cryptography comes into play. It promises the person or the organization to keep their data secure and private. It gives us a technique where we can jointly run a cryptographic protocol that will allow us to train a model on the joint database altogether hold but without compromising the privacy of the individual databases. This is called as Federated learning or Secure ML. That’s the promise that cryptography gives to ML. The data on which ML algorithms are used are not unlawfully used. For instance in the area of self-driving cars Companies like Tesla, BMW, Volkswagen, etc are coming together to make this possible and all of them hold their individual databases and want to build a model but at the same time, they are in business competitors so they don’t want to share their databases with each other. But cryptography is a technique which can and is bringing them together to train their model on a joint database and they need not compromise the privacy of their databases. To make life more comfortable.

There are many techniques like practical instantiations of secure multi-party computation (MPC), fully homomorphic encryption (FHE), etc. by which this can be achieved.

Cryptography can safely be said as "The Science of Secrets"

From art to science, from informal to become a part of formal computer science. Randomness is very important in cryptography as water in our lives. 

Let’s say there is a smart guy called X who secretly got access to your communication channel. Since this guy has access to your communication, he can do much more than just eavesdropping, for example, he can try to change the message.  Now, this is just a small example. What if Eave gets access to your private information? The result could be catastrophic.

So, Encryption is essentially important because it secures data and information from unauthorized access and thus maintains confidentiality. 

Cryptocurrencies use cryptography for three main purposes; to secure transactions, to control the creation of additional units, and to verify the transfer of assets. To accomplish all of these things, cryptocurrencies rely on what is called “public key cryptography.” With the advent of Bitcoin and the recent launch of Facebook's Libra as cryptocurrencies. I believe the responsibility of cryptography has increased and at the same time, it has to explore new possibilities to make the world a better and safer place.

 Machine Learning  "The art of automating automation"

Thanks to machine learning, there's never been a more exciting time in the history of computer science. Every day, new breakthroughs are changing what's possible with computers.