Event Agenda: 23 January 2024 – Day 2
08:30 - 09:00
09:00 - 09:05
09:05 - 09:10
Day 1 Wrap-up
09:10 - 09:50
FinTech - Hybrid Quantum Applications
In today's informative session, IonQ will showcase an application of Generative Learning for analyzing correlation structures in financial data using quantum copulas. Copulas are mathematical tools for modeling joint probability distributions. The fact that copulas can be expressed as maximally entangled quantum states has revealed a promising approach to practical quantum advantages. You'll also hear about other financial applications that your competitors are exploring. IonQ is expanding globally to meet the growing interest in quantum applications, so it will be our pleasure to share the latest technical advances in IonQ's trapped-ion quantum computers and the commercially focused applications that are compelling our clients to begin their quantum journeys with IonQ.
09:50 - 10:30
Towards large-scale quantum combinatorial optimization solvers with few qubits
Combinatorial optimizations are a major promise of quantum computation, but practical applicability before fault-tolerant devices appear is unclear. I present two methods that drastically reduce the experimental requirements for near-term quantum solvers. In the first part, I discuss a quantum-classical solver for binary optimizations of size m polynomially larger than the number of qubits n used. This features several technical advantages that lead to unprecedented performances. For instance, for m=2000 vertices, an experiment with n=17 trapped-ion qubits featured MaxCut approximation ratios beyond the hardness threshold 0.941. To our knowledge, this is the highest quality attained experimentally on such sizes. More generally, constrained optimizations require first reducing them to Quadratic Unconstraint Binary Optimizations (QUBOs). However, finding ‘good' QUBO reformulations can be hard. In the second part, I thus present an efficient classical algorithm for QUBO reformulations with significant advantages in terms of Hamiltonian spectral properties. For portfolio optimization, the quantum runtime using the novel reformulation is orders of magnitude smaller than with existing approaches. Our findings offer novel heuristics for quantum-inspired solvers as well as a promising route towards solving commercially-relevant problems on near term quantum devices.
10:30 - 11:00
11:00 - 11:30
Quantum feature maps - aspects of theory and hardware implementation
- Quantum kernels and quantum feature maps
- Analogy to other classical models
- The ansatz on NISQ devices
- A simple applied example
11:30 - 12:00
Quantum computing for probability distributions classification
- Definition of the problem of probability distributions classification and its relevance in finance
- Classical methods and definition of the quantum computing approach to the problem
- Executions on Rigetti's quantum computing platform and results
- Conclusions and outlook
12:00 - 12:30
Saving resources with Quantum Agents
Agents often execute complex strategies - continually adapting their reactions to input stimuli to synergize with past actions. As society pushes to automate ever more complex tasks, the computational resource requirements of such agents are growing in tandem – contributing to chip shortages, and a growing energy footprint of such technologies. Indeed, with the rapid advances in large language models, the memory resources costs required by such technologies has been doubling every 3-4 months and energetic costs are growing in tandem. In this talk we determine the fundamental limits on the amount of memory and energy required for executing a complex strategy classically. We demonstrate that a quantum agent can use less memory and lower energetic cost to execute such tasks, implying that it is more efficient to make decisions quantum mechanically.
12:30 - 13:30
13:30 - 14:00
Time-series forecasting with quantum-enhanced signature kernels
- Predicting time series is an ubiquitous task in quantitative finance and, in this talk, we examine a problem defined on real-world financial time-series
- We illustrate a technical approach to time-series predictions based on classical signature kernels combined with quantum feature maps
- We present the results we obtained running the method on Rigetti's quantum computing platform and comparing against classical benchmarks
- We complete the discussion with conclusions and considerations on future potential enhancements and extensions
14:00 - 14:30
Quantum-enhanced stochastic modelling and its potential implications for finance
Stochastic modelling and simulation are critical components of time-series analysis. They provide systematic means for us to make inferences about future occurrences and their potential likelihoods based on data available in the present and simultaneously help us isolate what elements of the past are most pertinent for predicting future behaviour. Yet, as the process we wish to model becomes ever more non-Markovian, and the futures we want to infer are ever more rare, the memory and time resources needed by traditional techniques can grow in tandem. In this presentation, I outline the potential for quantum models to mitigate such resource requirements. I will introduce quantum stochastic models, illustrating their potential to (i) generate equally accurate future predictions while tracking less data from the past and (ii) generate such predictions in quantum superposition. I then outline how these ideas can be combined for enhanced quantum stochastic analysis - allowing potential for quadratic speed-up in estimating the likelihood of rare events and quantum-enhanced dimensional reduction with reduced model distortion.
14:30 - 15:00
Randomized semi-quantum matrix processing
Quantum computers have the potential to speed up important matrix-arithmetic tasks. A prominent framework for that is the quantum singular-value transformation (QSVT) formalism, which uses Chebyshev approximations and coherent access to the input matrix via a unitary block encoding to design a target matrix function. Nonetheless, physical implementations for useful end-user applications require large-scale fault-tolerant quantum computers. Here, we present a hybrid quantum-classical framework for Monte-Carlo simulation of generic matrix functions more amenable to early fault-tolerant quantum hardware. We apply our technique to four specific use cases: partition-function estimation via quantum Markov-chain Monte Carlo and via imaginary-time evolution; end-to-end linear system solvers; and ground-state energy estimation. For these cases, we prove advantages on average circuit depths.
15:00 - 15:30
15:30 - 16:00
Design Space Exploration and Applications of Quantum Machine Learning
Quantum Computing (QC) claims to improve the efficiency of solving complex problems, compared to classical computing. When QC is applied to Machine Learning (ML) applications, it forms a Quantum Machine Learning (QML) system. After discussing the basic concepts of QC and its advantages over classical computing, this talk presents the key aspects of QML in a comprehensive manner, followed by an overview of different QML algorithms and their domain applicability, quantum datasets, hardware technologies, software tools, simulators, and applications such as finance, cryptography, and sensing. Afterward, this talk presents an analysis and exploration of the design space of hybrid Quantum Neural Networks by investigating different architectural permutations and quantum hyperparameters.
16:00 - 16:30
Hybrid quantum-classical reservoir computing for solving chaotic systems
Inspired by the reservoir computing classical paradigm of machine learning, and its ability to exploit non-linear dynamics to learn linear mapping of such dynamics to complex functions, we look at opportunity to explore the dynamics of quantum systems, as represented by sequential runs of quantum circuits specified via measurement results, for short term forecasting of time-series (nowcasting). We observe that even few noisy qubits on superconducting processors and simple entangling circuits could provide the richness of dynamics required for nowcasting chaotic system benchmarks. This work could pave the way to more complex applications in precipitation forecasting.