Quantum computing is a promising paradigm to solve computationally intractable problems. Various companies such as, IBM, Rigetti and D-Wave offer quantum computers using a cloud-based platform that possess several interesting features namely, (i) quantum hardware with various number of qubits and coupling maps exist at the cloud end that offer different computing capabilities; (ii) multiple hardware with identical coupling maps exist in the suite; (iii) coupling map of larger hardware with more number of qubits can fit the coupling map of many smaller hardware; (iv) the quality of each of the hardware is distinct; (v) user cannot validate the origination of the result obtained from a quantum hardware. In other words, the user relies on the scheduler of the cloud provider to allocate the requested hardware; (vi) the queue of quantum programs at the cloud end is typically long and maximizing the throughput, which is the key to reducing costs and helping the scientific community in their explorations. The above factors motivate a new threat model with following possibilities: (a) in future, less-trustworthy quantum computers from 3rd parties can allocate poor quality hardware to save on cost or towards satisfying their falsely-advertised qubit or quantum hardware specifications; (b) the workload scheduling algorithm could have a bug or malicious code segment which will try to maximize throughput at the cost of allocation to poor fidelity hardware. Such bugs are possible for trustworthy providers; (c) a rogue employee in trusted cloud vendor could try to sabotage the vendor's reputation by degrading the user compute fidelity just by tampering with the scheduling algorithm or rerouting the program; (d) a rogue employee can steal information by redirecting the programs to a 3rd party quantum hardware where they have full control. If the allocated hardware is inferior in quality, the user will suffer from poor quality result or longer convergence time. We propose two flavors of a Quantum Physically Unclonable Function (QuPUF) to address this issue- one based on superposition and another based on decoherence. Our experiments on real quantum hardware reveal that temporal variations in qubit quality can degrade the quality of the proposed QuPUF. We add a parametric rotation to the QuPUF for stability. Experiments on real IBM quantum hardware show that the proposed QuPUF can achieve inter-die Hamming Distance (HD) of 55% and intra-HD as low as 4%, as compared to ideal cases of 50% and 0% respectively. The proposed QuPUFs can also be used as a standalone solution for any other application.
Quantum Random Access Memory (QRAM) has the potential to revolutionize the area of quantum computing. QRAM uses quantum computing principles to store and modify quantum or classical data efficiently, greatly accelerating a wide range of computer processes. Despite its importance, there is a lack of comprehensive surveys that cover the entire spectrum of QRAM architectures. We fill this gap by providing a comprehensive review of QRAM, emphasizing its significance and viability in existing noisy quantum computers. By drawing comparisons with conventional RAM for ease of understanding, this survey clarifies the fundamental ideas and actions of QRAM. QRAM provides an exponential time advantage compared to its classical counterpart by reading and writing all data at once, which is achieved owing to storage of data in a superposition of states. Overall, we compare six different QRAM technologies in terms of their structure and workings, circuit width and depth, unique qualities, practical implementation, and drawbacks. In general, with the exception of trainable machine learning-based QRAMs, we observe that QRAM has exponential depth/width requirements in terms of the number of qubits/qudits and that most QRAM implementations are practical for superconducting and trapped-ion qubit systems.
Quantum Machine Learning (QML) has recently emerged as a rapidly growing domain as an intersection of Quantum Computing (QC) and Machine Learning (ML) fields. Hybrid quantum-classical models have demonstrated exponential speedups in various machine learning tasks compared to their classical counterparts. On one hand, training of QML models on real hardware remains a challenge due to long wait queue and the access cost. On the other hand, simulation-based training is not scalable to large QML models due to exponentially growing simulation time. Since the measurement operation converts quantum information to classical binary data, the quantum circuit is executed multiple times (called shots) to obtain the basis state probabilities or qubit expectation values. Higher number of shots worsen the training time of QML models on real hardware and the access cost. Higher number of shots also increase the simulation-based training time. In this paper, we propose shot optimization method for QML models at the expense of minimal impact on model performance. We use classification task as a test case for MNIST and FMNIST datasets using a hybrid quantum-classical QML model. First, we sweep the number of shots for short and full versions of the dataset. We observe that training the full version provides 5-6% higher testing accuracy than short version of dataset with up to 10X higher number of shots for training. Therefore, one can reduce the dataset size to accelerate the training time. Next, we propose adaptive shot allocation on short version dataset to optimize the number of shots over training epochs and evaluate the impact on classification accuracy. We use a (a) linear function where the number of shots reduce linearly with epochs, and (b) step function where the number of shots reduce in step with epochs. We note around 0.01 increase in loss and ~4% (1%) reduction in testing accuracy for reduction in shots by up to 100X (10X) for linear (step) shot function compared to conventional constant shot function for MNIST dataset, and 0.05 increase in loss and ~5-7% (5-7%) reduction in testing accuracy with similar reduction in shots using linear (step) shot function on FMNIST dataset. For comparison, we also use the proposed shot optimization methods to perform ground state energy estimation of different molecules and observe that step function gives the best and most stable ground state energy prediction at 1000X less number of shots.
Quantum Machine Learning (QML) is a new domain of Machine Learning (ML) and Quantum Computing (QC) that uses a trainable Variational Quantum Circuit (VQC) to solve various learning problems. Classical data is transformed to the quantum Hilbert space using an embedding scheme and the VQC performs quantum operations on this transformed quantum data using parametric quantum gates. In this work, we present a new embedding scheme using Quantum Random Access Memory (QRAM) that takes address as input, and gives data as the output similar to a classical Random Access Memory (RAM). We propose a basic circuit-based QRAM architecture and its application in, (a) storage for ML usage and, (b) storage of binary data. For ML, we store images of digits dataset into the QRAM and perform binary classification using a Quantum Neural Network (QNN). We observe that QNN driven by the proposed QRAM converges faster (at 6 th epoch) with 100% classification accuracy compared to QNN with normal embedding and classical Fully Connected Neural Network (FCNN) setups which converge with same accuracy at around 10 th and 15 th epochs, respectively. We obtain ≈ 63% classification accuracy on real hardware from IBM. For storage of binary data, we measure Hamming Distance (HD) and percentage correct predictions for quality evaluation of the proposed QRAM. We observe that the HD and percentage correct predictions worsens as the number of address lines increases. We propose two techniques to improve the QRAM accuracy namely, (i) clustering of raw data and allocating one QRAM per cluster and, (ii) bit splitting to divide wider data into smaller chunks and allocating one QRAM per split. Clustering provides best results by improving the HD for 9-address QRAM by ≈1.95X than pure QRAM and ≈1.74X improvement compared to bit-split (which provides ≈1.11X improvement than pure QRAM).
Identification of molecular properties, like side effects, is one of the most important and time-consuming steps in the process of molecule synthesis. Failure to identify side effects before submission to regulatory groups can cost millions of dollars and months of additional research to the companies. Failure to identify side effects during the regulatory review can also cost lives. The complexity and expense of this task have made it a candidate for a machine learning-based solution. Prior approaches rely on complex model designs and excessive parameter counts for side effect predictions. Reliance on complex models only shifts the difficulty away from chemists rather than alleviating the issue. Implementing large models is also expensive without prior access to high-performance computers. We propose a heuristic approach that allows for the utilization of simple neural networks, specifically the GRU recurrent neural network, with a 98+% reduction of required parameters compared to available large language models while obtaining near identical results as top-performing models.