Welcome to the Quantum Realm. 

Enjoy today’s breakdown of news, research, & events within quantum.

🔥 What’s quantum and neuromorphic and photonic all at once? An energy-efficient powerhouse — that’s what we like to hear. Plus, Hamiltonian QuGANs, error correction that doesn’t leave a single qubit behind, and the global trend that’s concerning for collaboration.

🗓️ UPCOMING

Sunday, July 7 | QTM-X Quantum Education Series 6 of 10

📰 NEWS QUICK BYTES

🛠️ Developing the Quantum Error Correction Stack for Diverse Qubit Types: While all quantum computers require error correction, error correction is not one size fits all; different qubit types require different strategies. That’s why Riverlane is committed to developing Deltaflow, a modular QEC Stack adaptable across qubit modalities. Deltaflow's development involves three tracks with each track using specialized decoding techniques to handle the challenges specific to different qubit types. With the combined benefit of advanced decoding methods and a centralized platform for error correction, Riverlane is actively working towards scalability, especially when considering the future potential for modularity.

✔️ Breakthrough in Fault-Tolerant Quantum Computing with Dual-Rail Qubit: Quantum Circuits, Inc. recently announced the development of the superconducting Dual-Rail Qubit that features built-in error detection, record fidelities of 99.99% for SPAM (state preparation and measurement), and superior coherence with phase errors and bit-flip errors occurring much less frequently as compared to standard superconducting qubits. This in turn lowers the resources needed for error correction and improves capabilities in terms of near-term algorithms.

🔒️ Identical Quantum Computer Export Bans Lack Scientific Rationale and Stem from Secret International Talks: Worldwide, governments are imposing export controls on quantum computers. The UK, France, Spain, and the Netherlands have set limits on exporting quantum computers with 34 or more qubits and low error rates, based on discussions under the Wassenaar Arrangement, which controls dual-use technologies. However, the rationale behind these specific limits remains undisclosed, leading to concerns about stifling innovation in quantum research. Attempts to obtain explanations through freedom of information requests have been denied on national security grounds, and while some officials claim the limits are based on scientific analysis, details of such analyses have yet to be released. This secrecy is sparking debates within the quantum computing industry about the underlying motivations as well as impacts of these restrictions.

👩‍🔬 Hamburg Quantum Computing Project Launched to Advance Quantum Technology and Applications in Industry: The "Hamburg Quantum Computing" project has been launched at the University of Hamburg's Center for Optical Quantum Technologies. This six-year, EU-funded joint project between the University of Hamburg and the Hamburg University of Technology is on a mission to develop software and hardware solutions for quantum computers, including the construction of the neutral atom Rymax One quantum computer demonstrator, specifically designed to solve optimization problems. Germany is now second only to China in public investment in quantum computing.

🎉 Host or join the Qiskit Fall Fest 2024 for a global celebration of quantum collaboration: The Qiskit Fall Fest returns this year for its fourth installment with the chance for participants to experience diverse and enriching quantum activities such as quantum challenges, hackathons, coding competitions, workshops, and social events. The theme for 2024, "World of Quantum," underscores the event's international scope and the rapid expansion of the global quantum community. Prospective event hosts and participants are encouraged to register for informational sessions by August 7, with the event roster announced in early September and events held in October and November. While most events are student-led, anyone can host or participate in events.

☕️ FRESHLY BREWED RESEARCH

Hamiltonian quantum generative adversarial networks: Hamiltonian quantum generative adversarial networks are proposed as a more performant framework for generating unknown quantum states and for their versatility in learning quantum dynamics. HQuGANs directly control the Hamiltonian parameters, which simplifies implementation and reduces errors as compared to parameterized quantum circuits. By using quantum optimal control techniques like GRAPE, HQuGANs successfully generate complex quantum states such as GHZ states and Haar random states with high fidelity. Breakdown here.

Hybrid Quantum-Classical Photonic Neural Networks: A hybrid quantum-classical photonic neural network integrates continuous variable quantum neural networks into neuromorphic photonic systems. The hybrid network achieves the same performance as classical networks twice their size and demonstrates resilience to noise and higher trainability. This has the potential to serve as a scalable and energy-efficient solution for advanced AI applications. Breakdown here.

High-rate intercity quantum key distribution with a semiconductor single-photon source: The first intercity quantum key distribution experiment is conducted using a deterministic single-photon source based on semiconductor quantum dots. A high-rate secret key transmission over a 79km optical fiber link between Hannover and Braunschweig, Germany, is achieved demonstrating a record-high secret key rate per pulse and low quantum bit error ratio. This experiment highlights the potential of semiconductor quantum dots for advancing secure quantum communication over long distances.

Error mitigation with stabilized noise in superconducting quantum processors: A study on stabilizing noise in superconducting quantum processors is conducted by modulating qubit-TLS interactions. Controlling qubit-TLS interactions is shown to reduce noise instabilities and improve error mitigation performance. They tested two strategies, optimized and averaged noise channels, showing that these methods improve the stability and accuracy of observable estimations in quantum computations.

State Transfer in Noisy Modular Quantum Networks: Quantum state transfer in noisy modular quantum networks is investigated, comparing the effects of two noise models: network topology noise and oscillator frequency noise. The study finds modular networks to be generally more robust to noise than monolithic ones. Noise in network topology mainly affects higher normal modes, while noise in oscillator frequencies impacts the slowest normal mode. The authors propose methods to mitigate these noises and increase the fidelity of quantum state transfer while emphasizing the importance of modular network structures.

Error-corrected Hadamard gate simulated at the circuit level: The article describes the simulation of an error-corrected Hadamard gate at the circuit level for a logical qubit encoded in a surface code. To implement this gate a method using a transversal Hadamard gate and patch deformation, and a method using a domain wall to interchange logical operators are explored. The circuits are optimized to maintain effective code distance under circuit-level noise and demonstrate comparable performance between the logical Hadamard gate and standard quantum memory experiments.

UNTIL TOMORROW.

BREAKDOWN

Hamiltonian quantum generative adversarial networks

🔍️ SIGNIFICANCE: 

• GANs, or generative adversarial networks, are essentially two neural networks playing an adversarial game in which the generator network has a strategy to maximize its ability to create false data that appears real and the discriminator network has a strategy to minimize the number of times it misclassifies data as real (minimax). Over time, each network improves its strategy and ultimately they converge to a point where both perform well, but never enough to overcome the other — the zero-sum game. GANs are particularly useful for generating high-dimensional and complex data, which has a broad set of use cases, including video generation and data augmentation.

• QuGANs adapt this concept for the generation of unknown quantum states. These are particularly useful for simulating and learning more about quantum systems.

• This research takes it a step further by proposing the Hamiltonian QuGAN which directly modifies the parameters of the Hamiltonian, versus the parameterized quantum circuits in “traditional” QuGANs. By directly modifying the Hamiltonian, the HQuGAN can reduce the errors and complexities that stem from control pulses and gate errors.

• Since direct control of the Hamiltonian parameters leads to continuous control, overparameterization becomes another feature of HQuGANs. This overparameterization actually improves the performance of the HQuGans in terms of convergence — it can navigate the optimization landscape more effectively and reach global saddle points more efficiently.

🧪 METHODOLOGY: 

• Two competing quantum optimal controls are used to iteratively update the control parameters of their Hamiltonians. The generator’s goal is to produce a quantum state that mimics the target state, while the discriminator’s goal is to be able to distinguish between the generated state and the target state.

• Since this is a minimax optimization problem, the cost function is designed to prevent mode collapse while encouraging convergence.

• Numerical simulations are used to demonstrate the effectiveness of HQuGANs in generating quantum states, including generalized GHZ states and Haar random states.

• The GRAPE algorithm, gradient ascent pulse engineering, is used for optimizing control parameters.

📊 OUTCOMES & OUTLOOK: 

• In terms of state generation, HQuGANs were able to generate highly entangled states such as GHZ states up to six qubits and Haar random states while consistently achieving a high fidelity (0.999) across the experiments.

• Using quantum optimal control techniques such as GRAPE encouraged better convergence as compared to circuit-based methods. The continuous nature of control parameters led to improved overparameterization and faster training.

• HQuGANs were extended to learn unknown unitary transformations using both Choi matrices and pairs of input-output quantum states to show its versatility in different quantum learning tasks.

• Overall, by addressing the limitations of traditional QuGANs and providing a more hardware-friendly approach, HQuGANs are a useful tool to add to the toolbox in order to generate and control complex quantum states.

BREAKDOWN

Hybrid Quantum-Classical Photonic Neural Networks

🔍️ SIGNIFICANCE: 

• Neuromorphic computing is another computing paradigm that takes both computational and architectural inspiration from the brain. One of the main advantages of neuromorphic is its energy efficiency — more complex computations can be performed while using less power as compared to conventional computers. One subarea of neuromorphic is neuromorphic photonics which uses light-based technologies to emulate the neural networks present in the brain. While neuromorphic photonics are faster and even more energy efficient, they are also limited by their physical size, which means that large and complex neural networks are particularly difficult.

• Continuous variable quantum neural networks are a type of QNN that already use photonic hardware and technologies. They are also relatively simple to integrate into hybrid networks since they can be trained using backpropagation, a technique used for classical neural networks. As compared to other QNNs, they are relatively resilient to error and noise.

• Bringing together all of the above, the researchers propose a hybrid network that combines the advantages of neuromorphoc, photonics and quantum into one. This network uses a CVQNN in place of a classical network in the hidden layer of a hybrid neural network.

• This integration enables the hybrid networks to achieve the same performance as classical networks twice their size, even under conditions of reduced bit precision. The hybrid network also demonstrates resistance to noise, making them more practical for real-world applications.

🧪 METHODOLOGY: 

• Both hybrid and fully classical neural networks were constructed to use for comparison. The main difference between these models is that the classical hidden layers in the neural networks are replaced with CV quantum neural networks in the hybrid models.

• To be quantumly processed, classical information is encoded into quantum modes (qumodes) using a series of displacement, squeezing, and Kerr gates.

• Both classical and quantum layers are trained using conventional gradient descent methods to solidify the compatibility and efficiency of the training process across hybrid networks.

• The training dataset was synthetically generated and consists of 1000 samples divided into four classes, each with eight features. This dataset was normalized to simulate the input transmission characteristics of photonic circuits.

📊 OUTCOMES & OUTLOOK: 

• For classical networks twice their size in terms of paramenters, the hybrid networks showed equivalent performance. For instance, a hybrid network with 120 parameters performed as well as a 235-parameter classical network.

• The hybrid networks maintained its superior accuracy even when noise was introduced. Specifically, hybrid networks required only 6.3 bits of precision to achieve near-ideal accuracy, which is within the capabilities of current photonic platforms.

• Hybrid networks were consistently more trainable than classical networks of similar size. For networks with fewer than 316 parameters, hybrid models outperformed classical models.

• The integration of quantum layers into photonic neural networks not only improves their computational capacity but also provides a way to successfully scale neuromorphic photonic systems. Plus, it’s possible that hybrid quantum-classical networks may begin to play a more significant role in advancing neuromorphic hardware, which would in turn enable high-performance AI with less energy use.