Homomorphic Encryption

What is homomorphic encryption?

Normal encryption is like locking a file in a safe: to do math, you have to open it, decrypt, then compute. Homomorphic encryption lets you add, subtract, multiply and divide while the box is still locked — and the decrypted result is exactly the same as if you'd computed in the clear first.

What makes it special?

"Encrypted 2" plus "encrypted 3" equals "encrypted 5"
The server gets two strings of gibberish, computes their sum anyway, and hands back the result — still gibberish. You unlock it and get 5. The whole time, the server never saw the real 2, 3, or 5.
Data that's usable but invisible
That's the appeal: outsource the computation to a cloud you don't trust, and the cloud never touches plaintext. Privacy and compute power at once.

What flavors are there?

Partially homomorphic
Supports only one operation (just addition or just multiplication). Efficient, used in specific cases.
Fully homomorphic (FHE)
Supports every operation and can in theory compute any function — but it's expensive. It's the "holy grail" cryptography has chased for decades.
Somewhat homomorphic
Allows a limited number of mixed operations, a middle ground behind many practical schemes.

What's hard, and where is it used?

Performance is the big bottleneck — fully homomorphic math can be orders of magnitude slower than plaintext. But for outsourcing medical data analysis, financial risk screening and cloud machine learning, it's huge: you want the cloud's compute without letting the cloud see your data, and homomorphic encryption is that key.

Bottom line: homomorphic encryption lets you lock your data in a box and still do the math through the box.