arXiv subjects

Zhipeng Liang

Publications and source records attributed to Zhipeng Liang.

22 records · Page 2Linked to original sources

arXiv2022

Channel Polarization of Two-dimensional-input Quantum Symmetric Channels

Being attracted by the property of classical polar code, researchers are trying to find its analogue in quantum fields, which is called quantum polar code. The first step and the key to design quantum polar code is to find out for the quantity which can measure the quality of quantum channels, whether there is a polarization phenomenon which is similar to classical channel polarization. Coherent information is believed to be the quantum analogue of classical mutual information and the quantity to measure the capacity of quantum channel. In this paper, we define a class of quantum channels called quantum symmetric channels, and prove that for quantum symmetric channels, under the similar channel combining and splitting process as in the classical channel polarization, the maximum single letter coherent information of the coordinate channels will polarize. That is to say, there is a channel polarization phenomenon in quantum symmetric channels.

quant-ph↗

arXiv2021

Improved Polar-code-based Efficient Post-processing Algorithm for Quantum Key Distribution

Combined with one-time pad encryption scheme, quantum key distribution guarantees the unconditional security of communication in theory. However, error correction and privacy amplification in the post-processing phase of quantum key distribution result in high time delay, which limits the final secret key generation rate and the practicability of quantum key distribution systems. To alleviate this limitation, this paper proposes an efficient post-processing algorithm based on polar codes for quantum key distribution. In this algorithm, by analyzing the channel capacity of the main channel and the wiretap channel respectively under the Wyner's wiretap channel model, we design a codeword structure of polar codes, so that the error correction and privacy amplification could be completed synchronously in a single step. Through combining error correction and privacy amplification into one single step, this efficient post-processing algorithm reduces complexity of the system and lower the post-processing delay. Besides, the reliable and secure communicaiton conditions for this algorithm has been given in this paper. Simulation results show that this post-processing algorithm satisfies the reliable and secure communication conditions well.

quant-ph↗

arXiv2021

Generalized Linear Bandits with Local Differential Privacy

Contextual bandit algorithms are useful in personalized online decision-making. However, many applications such as personalized medicine and online advertising require the utilization of individual-specific information for effective learning, while user's data should remain private from the server due to privacy concerns. This motivates the introduction of local differential privacy (LDP), a stringent notion in privacy, to contextual bandits. In this paper, we design LDP algorithms for stochastic generalized linear bandits to achieve the same regret bound as in non-privacy settings. Our main idea is to develop a stochastic gradient-based estimator and update mechanism to ensure LDP. We then exploit the flexibility of stochastic gradient descent (SGD), whose theoretical guarantee for bandit problems is rarely explored, in dealing with generalized linear bandits. We also develop an estimator and update mechanism based on Ordinary Least Square (OLS) for linear bandits. Finally, we conduct experiments with both simulation and real-world datasets to demonstrate the consistently superb performance of our algorithms under LDP constraints with reasonably small parameters $(\varepsilon, δ)$ to ensure strong privacy protection.

stat.ML↗

arXiv2018

Adversarial Deep Reinforcement Learning in Portfolio Management

In this paper, we implement three state-of-art continuous reinforcement learning algorithms, Deep Deterministic Policy Gradient (DDPG), Proximal Policy Optimization (PPO) and Policy Gradient (PG)in portfolio management. All of them are widely-used in game playing and robot control. What's more, PPO has appealing theoretical propeties which is hopefully potential in portfolio management. We present the performances of them under different settings, including different learning rates, objective functions, feature combinations, in order to provide insights for parameters tuning, features selection and data preparation. We also conduct intensive experiments in China Stock market and show that PG is more desirable in financial market than DDPG and PPO, although both of them are more advanced. What's more, we propose a so called Adversarial Training method and show that it can greatly improve the training efficiency and significantly promote average daily return and sharpe ratio in back test. Based on this new modification, our experiments results show that our agent based on Policy Gradient can outperform UCRP.

q-fin.PM↗