2025/12/22 Joint CQSE-NCTS-CACIB Seminars
2025/12/22 Joint CQSE-NCTS-CACIB Seminars
Seminar 1
Time: Dec. 22, 11:00 ~ 12:00Title: Quantum-Classical Hybrid Algorithms for Cancer Biomarker Discovery
Speaker: Dhirpal Shah, PhD candidate (Computer Science, University of Chicago)
Place: NCTS Physics Lecture Hall, 4F, Chee-Chun Leung Cosmology Hall, NTU
Online Link: https://nationaltaiwanuniversity-zbh.my.webex.com/nationaltaiwanuniversity-zbh.my/j.php?MTID=me030079a9c897b96fe518c8b0249c073
Abstract
Feature selection on dense, highly-correlated data sets is a combinatorially hard task that can have critical real-world relevance, making it a promising domain for demonstrating utility in quantum computing. In this work, we present a domain-specific, full-stack approach that connects application structure, algorithm choice and hardware limitations. Working within depth and connectivity constraints, our approach implements the recursive quantum approximate optimization algorithm (RQAOA) for feature selection by coupling topology-aware sparsification--integrating problem construction and qubit mapping to reduce circuit depth-- with application-tailored compilation and error mitigation. Using cancer biomarker discovery as a case study, these optimizations reduce circuit depths by up to 40$\times$, enabling a small-scale experiment on current IBM Heron devices. In a run of RQAOA on a 10 qubit circuit with two-qubit gate depth of 89 and depth $p{=}1$, the experimental observations match predictions from numerical simulation across consecutive iterations, making this the first demonstration of successful RQAOA on IBM quantum hardware. Using noiseless simulations to probe larger system sizes, we show that for selection from $\gtrsim$20 features, the hybrid classical-quantum approach can improve the results of a purely classical one. We benchmark state-of-the-art classical solvers and show that these begin to struggle in problems selecting from $\sim$100--200 features. Our resource estimates, tailored to the bivariate bicycle code for demonstrations within this regime, show that our approach goes beyond NISQ devices and into the early fault-tolerant era. This work paves the way for a demonstration of this application with the potential for empirical quantum advantage.
Biography
Dhirpal Shah is a Computer Science PhD candidate from Taipei at the University of Chicago where he also completed a joint BA/MS in Computer Science and a BA in Economics. His research focuses on co-designed quantum-classical pipelines for biomarker discovery, aligning algorithms, hardware, and clinical datasets to move toward quantum utility in cancer diagnostics as part of the Q4Bio collaboration with UChicago Medicine, MIT, and Infleqtion. He has contributed to projects on noisy quantum simulation and quantum circuit decomposition and scheduling for neutral atom architectures, as well as earlier work on SIMD dataflow optimization for neural network inference and green materials for pollutant remediation. His publications span quantum computing, computer architecture, and environmental applications.
Seminar 2
Time: Dec. 22, 14:30 ~ 15:30Title: From Quantum Circuits to Quantum Agents: Towards Scalable and Self-Programming Quantum AI
Speaker: Dr. Samuel Yen-Chi Chen (Wells Fargo, New York, NY 10017, USA)
Place: NCTS Physics Lecture Hall, 4F, Chee-Chun Leung Cosmology Hall, NTU
Online Link: https://nationaltaiwanuniversity-zbh.my.webex.com/nationaltaiwanuniversity-zbh.my/j.php?MTID=mefd006807e6161a0efe315d668e0a657
Abstract
This talk presents a unified vision for advancing quantum machine learning from static variational models to scalable, adaptive, and ultimately self-programming quantum agents. Beginning with Variational Quantum Circuits (VQCs) as the fundamental computational primitives, we introduce several recent frameworks that extend the expressive and structural capacity of quantum models: Quantum Architecture Search (QAS) including evolutionary, RL and differentiable methods for learnable quantum circuit design, the Quantum Fast Weight Programmer (QFWP) for meta-level parameter generation, and the Quantum Train (QT) paradigm that couples quantum and classical networks for hybrid learning.
These components form the basis for next-generation architectures such as Quantum LSTM (QLSTM) and Quantum Reinforcement Learning (QRL), enabling temporal modeling, decision-making, and adaptive behavior. I will also discuss their applications in climate forecasting, biomedical signal analysis, and communication networks, illustrating how quantum circuits can evolve toward more autonomous, agent-like intelligence. The talk concludes with future directions on distributed quantum learning and self-referential quantum AI systems.
Biography
Dr. Samuel Yen-Chi Chen is a researcher specializing in Quantum Machine Learning and Quantum AI, with prior appointments at Brookhaven National Laboratory and ongoing collaborations with universities and national laboratories worldwide. He received his Ph.D. and B.S. in Physics and an M.D. in Medicine from National Taiwan University. He pioneered several influential quantum AI frameworks-including Quantum LSTM (QLSTM), Quantum Reinforcement Learning (QRL), Differentiable Quantum Architecture Search (DiffQAS), and the Quantum Fast Weight Programmer (QFWP), which have shaped emerging directions in quantum temporal models, meta-learning, and quantum architecture search. His current research focuses on scalable quantum-enhanced agents, distributed quantum learning, and self-programming quantum AI systems.
