Comments URL: https://news.ycombinator.com/item?id=48726155

Points: 6

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=48636982

Points: 38

# Comments: 39

Comments URL: https://news.ycombinator.com/item?id=48533829

Points: 42

# Comments: 95

Comments URL: https://news.ycombinator.com/item?id=48186479

Points: 4

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=47152055

Points: 2

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=42923329

Points: 2

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=40749324

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=36374360

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=25440009

Points: 1

# Comments: 0

I can't say much more at the moment, but we should have a few more plugins released in the next few weeks that targets hardware from other QC vendors.

The D-Wave question in an interesting one, though. Unlike the QC hardware available from IBM, Rigetti, Google, etc, which uses a universal circuit model, D-Wave has focused on a particular application - quantum annealing. While our theoretical quantum gradient results only apply to the qubit model, it is an interesting question whether they can be extended to the quantum annealing framework.

For a while now, QML has been getting a lot of hype --- at the Quantum2Business conference the other day, a quote that made the rounds was "QML: most overhyped and underestimated field at the same time" (attributed to Iordanis Kerenidis, I believe).

However, current research has been showing a lot of promise, especially as an application for near-term quantum devices, that doesn't require an exceptionally large number of fault tolerant qubits.

At the moment, the main approach to QML has been the so-called 'variational circuit' approach, where a parameterised quantum circuit is evaluated on quantum hardware, with optimization/machine learning then performed by an external classical ML library, such as TensorFlow/PyTorch. However, this is not the most optimal approach - the most optimal approach is to take advantage of the quantum hardware to also perform the optimization.

This was our goal with PennyLane. Before we could even start designing the library, we needed to know how to analytically evaluate gradients on quantum circuits; so we performed the research, discovered some cool analytic tricks, and published this separately [1]. This forms the backbone of PennyLane - the exact same quantum circuits used in the machine learning model are also used to calculate the gradient during backpropagation. As a result, you can construct arbitrarily complex classical-quantum models, with both the quantum and classical parts natively 'backpropagation aware'.

Even more ambitiously, we wanted an environment where you can build a hybrid classical-quantum computational model, using not only different quantum hardware devices at once, but different hardware devices from different hardware vendors. By taking advantage of all near-term quantum hardware currently available - even those using fundamentally different models, such as qubits vs. photonic modes - you can build significantly more powerful computations. Currently, we have plugins available for [ProjectQ](https://projectq.ch), [Strawberry Fields](https://github.com/XanaduAI/strawberryfields), [Qiskit](https://qiskit.org/), and more to come.

Feel free to ask any questions you might have on PennyLane, the state of QML, and quantum computation in general!

[1] Evaluating analytic gradients on quantum hardware (https://arxiv.org/abs/1811.11184)

[2] Check out the PennyLane documentation for the nitty-gritty on our analytic gradient approach to QML: https://pennylane.readthedocs.io

Comments URL: https://news.ycombinator.com/item?id=18695494

Points: 8

# Comments: 4

Comments URL: https://news.ycombinator.com/item?id=18503535

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=18350232

Points: 1

# Comments: 0