csrdcf基准
在没有数学的情况下，知道何时真正值得使用量子处理器。(Know — without math — when it is actually worthwhile to use a quantum processor.)
This discovery followed some lively Twitter threads regarding my “130,780-point Quantum Classification,” “Superdense Encoding of Classical Data,” and “Quantum MNIST” articles, and whether or not quantum advantage can ever be achieved using SWAP Tests. The conversations referenced some theories and some mathematical proofs, so I decided to run some actual circuits with the intent of estimating the quantum processing unit (QPU) runtime for one of the circuits in question.
这项发现遵循了一些活跃的Twitter话题，涉及我的“ 130,780点量子分类” ，“经典数据的超编码”和“量子MNIST ”文章，以及使用SWAP测试是否可以实现量子优势。 讨论中引用了一些理论和数学证明，因此我决定运行一些实际电路，以估计其中一个电路的量子处理单元(QPU)运行时间。
介绍 (Introduction)
The primary reason to use QPUs, besides to perform calculations out of the range of supercomputers, is to simply perform calculations faster than classical devices and algorithms.
除了在超级计算机范围之外执行计算之外，使用QPU的主要原因还在于，它比传统的设备和算法执行速度更快。
quantum advantage is the demonstration that a quantum device can solve a problem … faster than classical computers
量子优势是量子设备可以解决问题的证明……比传统计算机更快
Source: Wikipedia
资料来源：维基百科
I’ve read arguments that add stipulations to this definition, such as that the algorithm needs certain components to even qualify as being “quantum,” but I’m going to go with this simple Wikipedia definition of “quantum advantage" for this article.
我已经阅读了为该定义添加规定的论点，例如该算法需要某些组件甚至可以被视为“量子”，但是本文将使用Wikipedia的“量子优势”这个简单定义。
您好，量子世界！ (Hello, Quantum World!)
The simplest quantum experiment of them all is the “hello, quantum world!” experiment. Most newcomers to the field run this as their very first quantum circuit.
他们中最简单的量子实验是“你好，量子世界！” 实验。 该领域的大多数新手将其作为第一个量子电路来运行。
For runtime estimation purposes, I intended for this circuit to provide me with the minimum runtime to run any circuit.
为了估计运行时间，我打算让该电路为我提供运行任何电路的最少运行时间。
The important number above is 25.1 seconds. That actually seems rather slow for such a shallow circuit.
上面的重要数字是25.1秒。 对于这么浅的电路，这实际上似乎很慢。
Not knowing what affects runtime, I added a little of everything: qubits, single-qubit gates, multi-qubit gates, and circuit depth. I intended to modify the circuit and then re-run it until I could identify the source of any slowdowns.
不知道什么会影响运行时间，我添加了一些内容：量子位，单量子位门，多量子位门和电路深度。 我打算修改电路，然后重新运行它，直到我能确定任何减速的原因。
Disclaimer: this isn’t an actual algorithm. I just added qubits and gates to the circuit to see how much the runtime would slow down.
免责声明：这不是实际的算法。 我只是在电路中添加了量子位和门，以查看运行时间会变慢多少。
And, this is the same circuit after transpilation. The important takeaway is that this circuit is far more complex than the “hello, quantum world!” circuit.
并且，这是移植后的相同电路。 重要的一点是，该电路比“你好，量子世界！”复杂得多。 电路。
Surprisingly, the runtime only slowed to 27.3 seconds from 25.1 seconds. Not shown here, I ran quite a few 15-qubit circuits, and the runtimes all stayed within this range.
令人惊讶的是，运行时仅从25.1秒减慢到27.3秒。 这里未显示，我运行了很多15个量子比特的电路，并且运行时都保持在此范围内。
Therefore, the next logical test was to try a different device. All I have available is 5-qubit devices.
因此，下一个逻辑测试是尝试其他设备。 我所能提供的只是5个量子比特的设备。
I started with the same “hello, quantum world!” circuit.
我从相同的“你好，量子世界”开始。 电路。
Despite running an identical circuit, the runtime on a 5-qubit device was slightly quicker, at only 23.3 seconds, than the 15-qubit device’s 25.1-second runtime.
尽管运行相同的电路，但5量子位设备的运行时间比15量子位设备的25.1秒运行时间要快一些，仅23.3秒。
Next, I added randomly complexity.
接下来，我添加了随机复杂性。
And, here is that circuit after transpilation. Once again, the key takeaway is that this circuit is noticeably wider and deeper than the “hello, quantum world!” circuit.
而且，这是移植后的电路。 再次要指出的是，该电路比“你好，量子世界！”明显更宽更深。 电路。
This time, the runtime only slowed from 23.3 seconds to 23.6 seconds. And, again, if you create a multitude of random circuits, you will find that all the runtimes fall within the same range.
这次，运行时间仅从23.3秒减慢到23.6秒。 同样，如果创建大量随机电路，您会发现所有运行时都在同一范围内。
结论 (Conclusion)
At least within IBM Q Experience, running any random circuit on a QPU will give you its approximate runtime for just about any circuit you could want to run on it. This means that you can easily determine whether or not a quantum algorithm will be faster than a classical counterpart: simply run the classical algorithm!
至少在IBM Q Experience中，在QPU上运行任何随机电路将为您提供其几乎可以在其上运行的任何电路的近似运行时间。 这意味着您可以轻松确定量子算法是否会比经典算法更快：只需运行经典算法！
Notwithstanding today’s qubits are not fault-tolerant, you achieve quantum advantage whenever your classical algorithm is slower than the QPUs benchmark runtime. No mathematics is required to make this determination.
尽管当今的量子位不是容错的，但只要您的传统算法比QPU基准测试运行时慢，您就可以获得量子优势。 无需数学即可进行此确定。
量子至上 (Quantum Supremacy)
This conclusion made me think about Google’s “quantum supremacy" experiment. I don’t know the answer, but I wonder if Sycamore’s runtime with 53 qubits is always in the range of 200 seconds. If so, looking at the ibmq_16_melbourne and ibmqx2 experiments that I ran, it would seem that superconducting QPU runtime is determined exclusively — or at least overwhelmingly — by a combination of the number of qubits available and the number of qubits in use. This also suggests that circuit depth — currently the bane of quantum computing due to decoherence — may someday be a boon as it allows additional computation and complexity without noticeably affecting runtime.
这个结论让我想到了Google的“量子至上性”实验，但我不知道答案，但我想知道Sycamore的53量子位的运行时间是否始终在200秒范围内；如果是这样，请查看ibmq_16_melbourne和ibmqx2实验，我跑了下来，似乎超导QPU运行时间完全取决于(或至少是压倒性地)由可用的量子位数量和使用的量子位数量决定的，这也表明电路深度(目前是量子计算的祸根)消除相干性—某天可能会受益匪浅，因为它允许额外的计算和复杂性，而不会明显影响运行时。
未来的工作 (Future Work)
The next logical course of research is identifying classical algorithms with which this simple technique could prove that quantum advantage is possible. Fortunately, I have a few in mind already.
下一步的逻辑研究方向是确定经典算法，用该简单技术可以证明量子优势是可能的。 幸运的是，我已经有了一些想法。
致谢。 (Acknowledgements.)
Thanks, as always, to IBM Q Experience for the free cloud access to real quantum processors. Thanks, also, to Quantum Intuition (@explore_quantum) for the debate that led to this discovery.
与往常一样，感谢IBM Q Experience对真实量子处理器的免费云访问。 另外，还要感谢Quantum Intuition ( @explore_quantum )引发了这一发现的辩论。
翻译自: https://medium.com/@bsiegelwax/benchmarking-quantum-advantage-90e02fb75612
csrdcf基准
191
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
