Serdecznie zapraszamy 4 marca 2024 r. o godzinie 12:00 na Seminarium Centrum Marka Kaca pod tytułem Reinforcement learning assisted quantum architecture search for near-term quantum algorithms, które wygłosi mgr Akash Kundu z Instytutu Informatyki Teoretycznej i Stosowanej PAN. Seminarium odbędzie się w sali D-2-43 Wydziału Fizyki, Astronomii i Informatyki Stosowanej UJ.
Abstract
In the noisy intermediate-scale quantum (NISQ) era of quantum computing, a major challenge is finding effective quantum circuits that work within the limitations of current hardware. Variational quantum algorithms (VQAs) were developed to tackle this challenge, but their success depends on factors such as: the initialization of variational circuits , its structure (which is called ansatz), and the cost function configuration. In this talk, I focus on improving VQAs by using reinforcement learning (RL) to automatically find the best circuit structures. This process of automating search for optimal structures is known as quantum architecture search (QAS), is often studied in noiseless environments, but we explore its effectiveness in noisy scenarios. We evaluate circuits based on depth, gate and parameter count, and accuracy in solving problems. Our approach includes using tensor-based circuit encoding, controlling environment dynamics, halting episodes to find shorter circuits, and employing a double deep Q-network (DDQN) with an $\epsilon$-greedy policy for stability. Our experiments demonstrate that our RL-based QAS performs better than existing methods across various VQAs, and our techniques can be adapted for different quantum algorithms.