How to convert program (write by python and tf) to quantum program?

PennyLane Development
1

Dear Everyone,

I have a simple speech transmission program that wirite by python and tensorflow.
it includes three main parts such as: Encoding part, transsimision channel (e.g, wireless channel) part, and Decoding part.
For example of code source for encoding part:

def init(self, frame_length, stride_length, arg):
self.num_frame = args.num_frame
self.frame_length = frame_length
self.stride_length = stride_length
def call(self, _input):

    # preprocessing _intput
    _input, batch_mean, batch_var = wav_norm(_input)
    _input = enframe(_input, self.num_frame, self.frame_length, self.stride_length)
    _input = tf.expand_dims(_input, axis=-1)
    _output = conv_bn_layer(_input, filters=self.sem_enc_outdims[0],
                            strides=(2, 2), name="sem_enc_cnn1")
    _output = tf.nn.relu(_output)
    _output = conv_bn_layer(_output, filters=self.sem_enc_outdims[1],
                            strides=(2, 2), name="sem_enc_cnn2")
    _output = tf.nn.relu(_output)
           
    return _output, batch_mean, batch_var

Now, I want to convert all to quantum program.
Who can give me some suggesion/advisor to mine can complete it?

2

Hey @baba_tnvn, welcome to the forum! :rocket:

A great place to start is our demos: Demos | PennyLane

There’s a bunch there, but the quantum machine learning ones are good to start with: PennyLane — Results

In particular, this one about integrating PennyLane and Tensorflow: Turning quantum nodes into Keras Layers | PennyLane Demos

Let me know if that helps!

3

Dear @isaacdevlugt ,
Thank for your reply, I saw it and practice following with Quanvolutional Neural Networks demo.
I have a question. I have one convolution layer that defined likes

def conv_bn_layer(inputs, filters, strides, name):
    conv = Conv2D(filters=filters, kernel_size=(3, 3), strides=strides,
                  padding="same", use_bias=False, name="{}_conv".format(name))(inputs)
    conv_bn = BatchNormalization(name="{}_bn".format(name))(conv)
    return conv_bn

After I quantum this layer, can I use it to compute for other program (note that, this program is classical program)

4

It seems like that later is purely classical? In that sense, you can’t really make it quantum. PennyLane’s QNN functionality is meant to do the opposite: make quantum circuits “behave” like classical layers :slight_smile:.

5

Hi @isaacdevlugt,

This program is purely classical, My main target is want to convert it to quantum.
Please, can you share with me steps to do that?

Thank you so much, sir

6

Hey @baba_tnvn as far as I’m aware, there’s no general way to embed a purely classical layer onto a quantum circuit. Even if there is, I don’t think it would be that efficient :thinking:

I think you should look into how you can trick a quantum layer to behaving like a classical layer. This demo that I shared earlier goes into how that’s possible: Turning quantum nodes into Keras Layers | PennyLane Demos

Let me know if that helps!