66 lines
1.8 KiB
Markdown
66 lines
1.8 KiB
Markdown
|
Yen-chi Chen, Wells Fargo
|
||
|
|
|
||
|
|
|
||
|
|
- Variational Quantum Circuits
|
||
|
|
- Parameterized Quantum Circuit (PQC) - "The other PQC"
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|0> --- encoding gate --- parameterized -----measurement
|
||
|
|
|
||
|
|
Rz = encoding gate
|
||
|
|
Ry = parameterized gate
|
||
|
|
|
||
|
|
|
||
|
|
[ sin(x)] $$ [[\sin x]] \sqrt{ y }$$
|
||
|
|
|
||
|
|
## Quantum gradients =
|
||
|
|
- the gradient of with respect to the parameter theta - the value can be calculated by running two quantum circuits with parameterized circuits - called parameter shift
|
||
|
|
- f (x,y,z) = z x Q(x + y)
|
||
|
|
|
||
|
|
|
||
|
|
Quantum CNN -
|
||
|
|
- use image classification to QCNN wireachhigh accuraccy wuth smaller number of epochs
|
||
|
|
- quantum federarted learning - they don't share the dataset
|
||
|
|
- they will train on their own and they only share the models
|
||
|
|
|
||
|
|
|
||
|
|
## Quantum Fedrated ML with Differential Privacy
|
||
|
|
- During our training we get some noise to get our gradient
|
||
|
|
- to ensure that our data is differential and private
|
||
|
|
- Watkins, Yoo -- *Differential machine learning with differential privacy*
|
||
|
|
|
||
|
|
|
||
|
|
## Quantum LSTM
|
||
|
|
- Quantum long short-term memory
|
||
|
|
- replace the classical neural networks with variational quantum circuits (VQC_)
|
||
|
|
- QLSTM can show ---- read the paper
|
||
|
|
|
||
|
|
Quantum RL
|
||
|
|
- quantum agents in the Gym - a vriational quantum algorithm
|
||
|
|
- quantum rl is similar to classical rl except replace
|
||
|
|
|
||
|
|
|
||
|
|
Paper : *Efficient quantum recurrent reinforcdement learning via quantum reservoir computing*, 2024
|
||
|
|
|
||
|
|
|
||
|
|
When BERT meets quantum - paper
|
||
|
|
|
||
|
|
|
||
|
|
1. H QC P > using classical computer to optimize ]
|
||
|
|
2. Variational quantum circuits aka paramtererized quantum circuits
|
||
|
|
3. can be used with classical components and whole model improved with gradient descent methods
|
||
|
|
4.
|
||
|
|
|
||
|
|
Paper outline for structure:
|
||
|
|
- Motivation
|
||
|
|
- Challenge:
|
||
|
|
- Approach
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
- Create benchmarks for quantum machine learning - very hard to convince classical machine learning people that quantum is useful..
|
||
|
|
- how to mitigate barren plateau
|