binoua · HackerTrans

Yes for now, it's 1000 song, which is already awesome if you think about it, no? As it's like 300 ms, one can increase the DB size by a few order of magnitude, certainly. It will scale to billions of songs thanks to hardware accelerators, which are coming. One can google and see that there is a bunch of companies (small or large) working on accelerating FHE computations.

binoua

2 年前議論

Everything is open-source, you can have a look yourself, and experiment! https://github.com/iamayushanand/Concrete_Shazam/blob/main/M...

Here, the training is not done on encrypted values: the songs are public, what is secret is which song(s) you like

binoua

3 年前議論

Thank you! The python version is quite clear as well: still from the README,

``` from concrete import fhe

def add(x, y): return x + y

compiler = fhe.Compiler(add, {"x": "encrypted", "y": "encrypted"}) inputset = [(2, 3), (0, 0), (1, 6), (7, 7), (7, 1), (3, 2), (6, 1), (1, 7), (4, 5), (5, 4)]

print(f"Compiling...") circuit = compiler.compile(inputset)

print(f"Generating keys...") circuit.keygen()

examples = [(3, 4), (1, 2), (7, 7), (0, 0)] for example in examples: encrypted_example = circuit.encrypt(*example) encrypted_result = circuit.run(encrypted_example) result = circuit.decrypt(encrypted_result) print(f"Evaluation of {' + '.join(map(str, example))} homomorphically = {result}") ```

Here, that's more for non-ML computations.

binoua

3 年前議論

Would you mind elaborating what you prefer in Google's syntax, please?

コメント