Learning Unitaries by Gradient Descent

We study the hardness of learning unitary transformations by performing gradient descent on the time parameters of sequences of alternating operators. Such sequences are the basis for the quantum approximate optimization algorithm and represent one of the simplest possible settings for investigating problems of controllability. In general, the loss function landscape of alternating operator sequences in $U(d)$ is highly non-convex, and standard gradient descent can fail to converge to the global minimum in such spaces. In this work, we provide numerical evidence that -- despite the highly non-convex nature of the control landscape -- when the alternating operator sequence contains $d^2$ or more parameters, gradient descent always converges to the target unitary. The rates of convergence provide evidence for a "computational phase transition." When the number of parameters is less than $d^2$, gradient descent converges to a sub-optimal solution. When the number of parameters is greater than $d^2$, gradient descent converges rapidly and exponentially to an optimal solution. At the computational critical point where the number of parameters in the alternating operator sequence equals $d^2$, the rate of convergence is polynomial with a critical exponent of approximately 1.25.

Deep neural networks are biased towards simple functions

We prove that the binary classifiers of bit strings generated by random wide deep neural networks are biased towards simple functions. The simplicity is captured by the following two properties. For any given input bit string, the average Hamming distance of the closest input bit string with a different classification is at least $\sqrt{n\left/\left(2\pi\ln n\right)\right.}$, where $n$ is the length of the string. Moreover, if the bits of the initial string are flipped randomly, the average number of flips required to change the classification grows linearly with $n$. On the contrary, for a uniformly random binary classifier, the average Hamming distance of the closest input bit string with a different classification is one, and the average number of random flips required to change the classification is two. These results are confirmed by numerical experiments on deep neural networks with two hidden layers, and settle the conjecture stating that random deep neural networks are biased towards simple functions. The conjecture that random deep neural networks are biased towards simple functions was proposed and numerically explored in [Valle P\'erez et al., arXiv:1805.08522] to explain the unreasonably good generalization properties of deep learning algorithms. By providing a precise characterization of the form of this bias towards simplicity, our results open the way to a rigorous proof of the generalization properties of deep learning algorithms in real-world scenarios.

Continuous-variable quantum neural networks

We introduce a general method for building neural networks on quantum computers. The quantum neural network is a variational quantum circuit built in the continuous-variable (CV) architecture, which encodes quantum information in continuous degrees of freedom such as the amplitudes of the electromagnetic field. This circuit contains a layered structure of continuously parameterized gates which is universal for CV quantum computation. Affine transformations and nonlinear activation functions, two key elements in neural networks, are enacted in the quantum network using Gaussian and non-Gaussian gates, respectively. The non-Gaussian gates provide both the nonlinearity and the universality of the model. Due to the structure of the CV model, the CV quantum neural network can encode highly nonlinear transformations while remaining completely unitary. We show how a classical network can be embedded into the quantum formalism and propose quantum versions of various specialized model such as convolutional, recurrent, and residual networks. Finally, we present numerous modeling experiments built with the Strawberry Fields software library. These experiments, including a classifier for fraud detection, a network which generates Tetris images, and a hybrid classical-quantum autoencoder, demonstrate the capability and adaptability of CV quantum neural networks.

Quantum Machine Learning

Fuelled by increasing computer power and algorithmic advances, machine learning techniques have become powerful tools for finding patterns in data. Since quantum systems produce counter-intuitive patterns believed not to be efficiently produced by classical systems, it is reasonable to postulate that quantum computers may outperform classical computers on machine learning tasks. The field of quantum machine learning explores how to devise and implement concrete quantum software that offers such advantages. Recent work has made clear that the hardware and software challenges are still considerable but has also opened paths towards solutions.

Quantum support vector machine for big data classification

Supervised machine learning is the classification of new data based on already classified training examples. In this work, we show that the support vector machine, an optimized binary classifier, can be implemented on a quantum computer, with complexity logarithmic in the size of the vectors and the number of training examples. In cases when classical sampling algorithms require polynomial time, an exponential speed-up is obtained. At the core of this quantum big data algorithm is a non-sparse matrix exponentiation technique for efficiently performing a matrix inversion of the training data inner-product (kernel) matrix.

A Turing test for free will

Before Alan Turing made his crucial contributions to the theory of computation, he studied the question of whether quantum mechanics could throw light on the nature of free will. This article investigates the roles of quantum mechanics and computation in free will. Although quantum mechanics implies that events are intrinsically unpredictable, the `pure stochasticity' of quantum mechanics adds only randomness to decision making processes, not freedom. By contrast, the theory of computation implies that even when our decisions arise from a completely deterministic decision-making process, the outcomes of that process can be intrinsically unpredictable, even to -- especially to -- ourselves. I argue that this intrinsic computational unpredictability of the decision making process is what give rise to our impression that we possess free will. Finally, I propose a `Turing test' for free will: a decision maker who passes this test will tend to believe that he, she, or it possesses free will, whether the world is deterministic or not.