Pre-Initial Public Offering(Pre-IPO) investment is an extremely important investment method in private equity investment. It invests in the latest stage of the company's listing time. In the context of Chinese stock issuance registration system reform, a large number of small and medium-sized enterprises with technological attributes and high growth potential have begun to conduct IPOs. As the last opportunity for companies to replenish their "ammunition" before going public, Pre-IPO investment has gradually become a hot topic in the practical and academic circles. This article collected four financial indicators that may affect the Pre-IPO investment income(including profitability, solvency, growth potential and governance structure) and used a multiple regression model to examine the mechanism. This article has three research conclusions: First, the Pre-IPO investment still has considerable returnsin the current market environment, but the rate of return is gradually decreasing. Secondly, most of the Pre-IPO return is brought about by the successful listing of the invested company, and the return on total assets, current ratio, and total asset growth rate have a significantly positive impact on primary market investment income, while the controlling shareholder's shareholding ratio has a negative impact. The income from secondary market is much less than primary market, and it is mainly related to the profitability and solvency of the invested company. Profitability has a negative impact on the stock price, while solvency has a positive impact; Thirdly, with the launch of the Science and Technology Innovation Board and the implementation of the Registration System, investment institutions need to conduct a comprehensive inspection of the invested companies, such as the company's growth and internal governance structure. Companies listed on the Science and Technology Innovation Board need to pay more attention to their profitability and growth,while others need to pay more attention to the debt solvency and governance structure.

Kungfu bonds are bonds denominated in dollars issued by mainland companies in overseas markets. In the past ten years, the issuance of Chinese dollar bonds has been in full swing. The Kungfu bond market is booming with China's economic development and the deepening of the capital market. Since 2019, the regulatory policies for the domestic real estate industry have gradually become stricter. Developers led by Evergrande and Country Garden have extremely tight cash flow, and their domestic and overseas financing channels are greatly restricted. In addition, due to the repeated impact of the epidemic, the risk exposure of Chinese-funded US dollar bonds and real estate debts has intensified. Due to the wave of defaults induced by the forced deleveraging of domestic regulators, overseas credit bonds with poor credit quality have been sold by investors. This paper looks into the interest rate level of Kungfu bonds as the research object, and examines the information asymmetry as the research entry point to conduct an in-depth quantitative study of the additional costs that mainland real estate companies need to pay for new bonds issued in the international market, and what factors may exacerbate or alleviate information asymmetry.This study found that Mainland real estate companies need to pay an additional 1.2238% interest rate difference on average when issuing bonds in Hong Kong for the first time. In addition, Mainland real estate companies that have issued bonds in the international market pay a lower credit premium on average, which means that issuing bonds in the international market can significantly enhance the company's reputation and alleviate information asymmetry among institutional investors, thereby reducing financing costs. To sum up, this paper analyzes in depth the pricing problem of Chinese dollar bonds issued in the international market through the method of quantitative regression, enriches the related research on bond issuance pricing, and provides information necessary for practitioners to make investment decisions and for listed companies to make financing arrangements. It provides valuable suggestions, which may be applied to other industries.

Recently, a well-designed and well-trained neural network can yield state-of-the-art results across many domains, including data mining, computer vision, and medical image analysis. But progress has been limited for tasks where labels are difficult or impossible to obtain. This reliance on exhaustive labeling is a critical limitation in the rapid deployment of neural networks. Besides, the current research scales poorly to a large number of unseen concepts and is passively spoon-fed with data and supervision.

To overcome the above data scarcity and generalization issues, in my dissertation, I first propose two unsupervised conventional machine learning algorithms, hyperbolic stochastic coding, and multi-resemble multi-target low-rank coding, to solve the incomplete data and missing label problem. I further introduce a deep multi-domain adaptation network to leverage the power of deep learning by transferring the rich knowledge from a large-amount labeled source dataset. I also invent a novel time-sequence dynamically hierarchical network that adaptively simplifies the network to cope with the scarce data.

To learn a large number of unseen concepts, lifelong machine learning enjoys many advantages, including abstracting knowledge from prior learning and using the experience to help future learning, regardless of how much data is currently available. Incorporating this capability and making it versatile, I propose deep multi-task weight consolidation to accumulate knowledge continuously and significantly reduce data requirements in a variety of domains. Inspired by the recent breakthroughs in automatically learning suitable neural network architectures (AutoML), I develop a nonexpansive AutoML framework to train an online model without the abundance of labeled data. This work automatically expands the network to increase model capability when necessary, then compresses the model to maintain the model efficiency.

In my current ongoing work, I propose an alternative method of supervised learning that does not require direct labels. This could utilize various supervision from an image/object as a target value for supervising the target tasks without labels, and it turns out to be surprisingly effective. The proposed method only requires few-shot labeled data to train, and can self-supervised learn the information it needs and generalize to datasets not seen during training.

Spin-orbit interactions are important in determining nuclear structure. They lead to a shift in the energy levels in the nuclear shell model, which could explain the sequence of magic numbers in nuclei. Also in nucleon-nucleon scattering, the large nucleon polarization observed perpendicular to the plane of scattering needs to be explained by adding the spin-orbit interactions in the potential. Their effects change the equation of state and other properties of nuclear matter. Therefore, the simulation of spin-orbit interactions is necessary in nuclear matter.

The auxiliary field diffusion Monte Carlo is an effective and accurate method for calculating the ground state and low-lying exited states in nuclei and nuclear matter. It has successfully employed the Argonne v6' two-body potential to calculate the equation of state in nuclear matter, and has been applied to light nuclei with reasonable agreement with experimental results. However, the spin-orbit interactions were not included in the previous simulations, because the isospin-dependent spin-orbit potential is difficult in the quantum Monte Carlo method. This work develops a new method using extra auxiliary fields to break up the interactions between nucleons, so that the spin-orbit interaction with isospin can be included in the Hamiltonian, and ground-state energy and other properties can be obtained.