This talk explores the cutting-edge field of quantum computing and generative AI, examining its current state and future potential. The presentation provides an overview of quantum machine learning and its implementation on NISQ hardware, specifically Quantum Support Vector Machines (QSVM), Quantum Neural Networks (QNN), and Quantum Generative Adversarial Networks (QGAN). In addition to discussing hybrid architectures that combine both quantum and classical neural networks, the talk also highlights the capabilities of Tensor Networks in increasing the explainability, expressibility, and controllability of Large Language Models (LLMs). Finally, the session highlights the potential real-world applications of generative AI in the pharmaceutical industry, with a focus on the development of de novo drug design using QGAN.

A tensor is a multidimensional array of numbers that can be used to store data, encode a computational relation and represent quantum entanglement. In this sense a tensor can be viewed as valuable resource whose transformation can lead to an understanding of structure in data, computational complexity and quantum information. In order to facilitate the understanding of this resource, we propose a family of information-theoretically constructed preorders on tensors, which can be used to compare tensors with each other and to assess the existence of transformations between them. We hope that this work will provide a useful vantage point for exploring tensors in applied mathematics, physics and computer science, but also from a purely mathematical point of view.

The Quantum Technology & Application Consortium (QUTAC) is a collaboration among leading German companies that aims to industrialize quantum computing as soon as feasible. The core activities of QUTAC are carried out by scientific working groups that try to identify the latest academic advancements and test their practicality in applied settings. In this talk, we showcase the operating mode of the Materials Science Working Group and communicate recent findings from investigations on the quantum-enhanced Quantum Monte Carlo (QC-QMC) proposed by Huggins et al.

During the past decade, many new techniques for addressing hard optimization problems were proposed by research groups in academia and industry. Most of them can heuristically sample solutions for QUBO (Quadratic Unconstrained Binary Optimization) problems or for an equivalent Ising model formulation. Yet, the structural simplicity of QUBO models and the heterogeneous landscape of novel specialized hardware put some challenges for those attempting to build applications that rely on such methods. In this presentation, we will describe the fundamental building blocks for QUBO.jl: a Julia package ecosystem that allows relevant industrial applications to be constructed with specialized hardware for optimization. First, it is necessary to imagine a uniform interface through which the many devices can communicate. This was achieved by integrating many of these solvers to the JuMP ecosystem, where users can find a friendly experience for interacting with those solvers. Next, we will depict the creation of a compiler for mathematical programming that can translate more general, high-level models, into low-level abstractions that are ready to be run in the novel target architectures. This is an essential piece for taking optimization models developed in industrial settings and using them to prototype and deploy applications based on quantum and other physics-inspired devices. Moreover, we are going to point out some of the relevant implications of this work, including potential use cases such as testing and benchmarking solvers. Finally, we will consider the extension of the QUBO formalism into QUMO (Quadratic Unconstrained Mixed-Integer Optimization), by allowing the presence of continuous variables. Even though most of these solvers operate using QUBO models, recent advances on the theory and implementation of QUMO solvers call for a generalization step in the proposed software suite.

In recent years, quantum computing has transitioned from an academic curiosity to a topic of interest for industrial biomanufacturers. The central question now is: What value can Quantum Computing bring to the bioprocess landscape? Our research focuses on leveraging Time Series Generative Adversarial Neural Networks (TS-GANs) in a hybrid approach. We propose that Quantum Generative Artificial Intelligence (QGAI) can rapidly and accurately generate synthetic data. This capability will be instrumental in scaling up challenging tasks, including the global optimization of bioprocess systems. Inspired by recent advancements in Quantum Drug Design, we explore a novel approach. We feed latent-space data from a GAN into a Parameterized Quantum Circuit (PQC). The PQC’s output informs the generator, enhancing the learning of the prior distribution. This iterative process improves the discriminator’s ability to distinguish between passing and failing data generated by the model. Currently, we employ the DoppleGANger model to generate synthetic data based on available process data. However, we anticipate that incorporating a PQC will allow us to create accurate synthetic data even from sparse observations. This advancement promises faster training times, reduced computational costs, and higher-quality outputs. Ultimately, it benefits industries seeking Generative Enhanced Optimization (GEO) for their industrial bioprocesses.

Modern AI methods based on deep neural networks are promising for accelerating the development of quantum computing. Most implementations to date have been small-scale; however, the power of deep neural networks arises when significant data and compute are invested. We present two approaches to quantum circuit compilation using deep reinforcement learning and language models, in collaboration with teams from DeepMind and META. The problem of circuit compilation is of fundamental importance in quantum computation, as reducing the quantum resources necessary for an application accelerates the path towards achieving quantum advantage. We find that by combining domain knowledge from human experts with large-scale generative models, we obtain significant improvements over the state-of-the-art in standard benchmark sets of circuits. These promising results add to and reinforce the motivation for the application of AI methods in the full stack of quantum computing.

One of the major challenges in computer science is to establish lower bounds on the resources, typically time, that are needed to solve computational problems, especially those encountered in practice. A promising approach to this challenge is the study of fine-grained complexity, which employs special reductions to prove time lower bounds for many diverse problems based on the conjectured hardness of key problems. For instance, the problem of computing the edit distance between two strings, which is of practical interest for determining the genetic distance between species based on their DNA, has an algorithm that takes O(n^2) time. Through a fine-grained reduction, it can be demonstrated that a faster algorithm for edit distance would imply a faster algorithm for the Boolean Satisfiability (SAT) problem. Since faster algorithms for SAT are generally considered unlikely to exist, this implies that faster algorithms for the edit distance problem are also unlikely to exist. Other problems used for such reductions include the 3SUM problem and the All Pairs Shortest Path (APSP) problem. The quantum regime presents similar challenges; almost all known lower bounds for quantum algorithms are defined in terms of query complexity, which offers limited insight for problems where the best-known algorithms take super-linear time. Employing fine-grained reductions in the quantum setting, therefore, represents a natural progression. However, directly translating classical fine-grained reductions to the quantum regime poses various challenges. In this talk, I will present recent results in which we overcome these challenges and prove quantum time lower bounds for certain problems in BQP, conditioned on the conjectured quantum hardness of, for example, SAT (and its variants), the 3SUM problem, and the APSP problem. This presentation is based on joint works with Andris Ambainis, Bruno Loff, Florian Speelman, and Subhasree Patro.