Lab

In this lab, we will:

• Load and preprocess the AndroidMalware dataset

• Implement classical SVM using scikit-learn

• Implement QSVM in PennyLane

• Compare SVM and QSVM in terms of accuracy

Step 3: Preprocessing the dataset

Before the training, we have to preprocess the dataset so it can be trained on our machine learning model.

Usually, we will drop the empty rows and standardize the data. 

If we have a multidimensional feature space, PCA is used to condense the information and make the training easier.

Besides, if we want to train on the quantum model, we should scale the data because we are using angle embeddings, so angles outside the range [-π, π] will be mapped back into this interval.

Step 4: Build an SVM model

We can directly load the dataset into our classical model. They are already built in the scikit-learn library.

Step 4: Build a Quantum Device

The first part of our quantum computing is setting up the number of qubits. Usually, it equals the dimensions of the features.

Then we load the data into our quantum model.

Firstly, we use a Hadamard gate to map the initial states |0> into a superposition for later data embedding.

Then we use rotation gates to inject data.

We can also use CNOT gates to strengthen the connections between features.

Step 5: Build a Quantum Kernel

In our quantum kernel, the key lies in our matrix operations.

So, we apply our feature map to the first input, and apply the adjoint with the second feature-mapped input. 

Then measuring the expected output, we would get the similarity between these two inputs.

This is the kernel trick in quantum kernel

Then we find our weights and bias in the training set, and use these parameters in our train-test matrix to calculate the similarity. 

The sign of our prediction indicates the class.

Step 6: Replace the Classical Kernel Function with Quantum Kernel

Step 7: Print the Result

details

Link for the code: Code Link (Time took to run the code: 5-10 min)

Link for Dataset: Dataset Link