Hardware war, computing power competition, genius game: game-breakers in the era of quantitative involution

This is the best era for quantitative investment. Industry recognition continues to rise, outstanding talents are gathering together, and more technical players are exploring uninhabited lands...

This is also the "worst" era for quantitative investment In this era, more intense competition rules and cruel elimination mechanisms mean that only by constantly climbing over layers of technological peaks can we "see the mountains and the small mountains at a glance."

In 2023, the quantitative private equity industry, which has entered the "trillion" era, has begun an involution known as hardware wars, computing power competitions, and genius games. From barbaric growth to intensive farming, how to face the industry ecosystem where technology and intelligence competition are constantly escalating?

Kunde Investment, which has been steadily operating in the domestic market for ten years, is using its own experience to solve problems. Let's take a look at the answers given by the Kuande people.

The following video comes from: WizardQuant Investment With a rigorous academic attitude, when technology takes off

When Elric from Tsinghua University’s math and science class chose a major, the world was facing In the era of rapid data expansion, statistics, as an important basic part of data science, is increasingly used in real life applications.

While there are diversified options, how can you strongly connect your own interests with the needs of academia and industry? I listened to several lectures by returned teachers who shared academic and industry cutting-edge lectures, and exchanged learning experiences with my exchange mentors at UCLA. Especially after I learned about the inextricable connection between statistics and the hot machine learning, Elric The starting point of the voyage is locked.

Statistics is about exploring unknown patterns from massive amounts of data and finding order in seemingly disordered conditions, which requires long-term rigor and persistence in scientific research. Timothy Gowers once made an interesting analogy. He said that most of the far-reaching contributions in mathematics were made by "tortoises" rather than "rabbits." This means that in the long-term research time scale, only through repeated discussions with data and models, whether with joy or pain, can those breakthrough achievements come slowly.

Later, during his several years studying biostatistics at Berkeley, Elric not only trained in the scientific qualities necessary for academic research, but also made in-depth explorations in many frontier fields. Multiple hypothesis testing, Bayesian models, bioinformatics, pharmacogenomics, graph neural networks, and non-convex learning theory. His research scope extends from statistics and computational biology to machine learning.

When Elric was studying abroad, the domestic quantitative trading market was developing rapidly. Although overseas markets have more complete trading tools, friendlier market factors, and more mature and rational capital, they are also faced with many problems such as increasing homogeneity of strategies, obvious Matthew effect, and over-reliance on leverage and low-cost financing. In comparison, although the development of domestic market trading is in a relatively early stage, the market is more dynamic and the industry is in the ascendant. This actually provides more imagination and room for growth for quantitative trading, especially for young practitioners.

From a technical point of view, the top domestic quantitative trading institutions are rapidly catching up with top overseas institutions, and even have their own unique advantages in segmented technical directions. Although it started late, domestic institutions have obvious advantages in being latecomers. Whether it is the application of SOTA deep learning technology, the construction of large-scale AI computing platforms, and the design of high-performance software and hardware trading systems, domestic institutions represented by Kuande Quantitative trading institutions are at the forefront. This east wind of technology has set sail for Elric to set sail in Kuande.

软 All trading software and frameworks of Kuande are independently developed, and can make dynamic adjustments with markets and exchanges trading rules and policies.

The original intention of pursuing perfection

The needs of the times are like traffic lights guiding the orderly flow of talents. With the acceleration of technological innovation, more and more financial institutions in the country Employs machine learning algorithms to predict market trends and select investment portfolios. At present, the research and application of traditional machine learning (such as kernel methods, tree models, etc.) and the broader and rapidly developing deep learning have become a must-have for financial institutions, especially the quantitative industry.

After graduating from PhD, Elric joined Cundall as a machine learning researcher. Currently, Elric's work focuses on data mining, designing model solutions and strategies, using machine learning methods to solve various puzzles in the financial market, and discovering high-dimensional, non-linear connections in massive data. Quantification has given him a broader perspective Show heaven and earth.

"Cundak has many independent data sources, ultra-large-scale computing resources and strong computing power support, which has created a superior working environment for us. For example, I am equipped with sufficient A100, I don’t have to worry about it like I was a student We are also worried about insufficient resources, and work efficiency has been greatly improved," Elric said.

The field of research is constantly expanding. How can we turn the tools in our hands into "sharp weapons"?
Elric believes that the more a team can combine the strengths of everyone and work together, the longer it will last.

Kunde’s team includes colleagues who specialize in theoretical mathematics and theoretical physics, as well as members who study engineering, statistics, and numerical calculations. In Kuande, he met outstanding colleagues such as IPhO gold medalists, game masters + PhDs in mathematics, senior veterans of Flower Street, talented interns, etc. They often consulted with each other in their free time, and provided more complete and comprehensive services from their respective professional perspectives. solution ideas.

"Many of our "physics gold medalists" often think about financial laws with insight into the laws of physics. This kind of collision can often lead to inspiration. For example, the famous Feynman-Kac formula in financial engineering, It was created by the famous physicist Richard Feynman when he was describing the distribution of high-energy particles. Even though he is also a mathematics major, his PhD research direction is the quasi-Monte Carlo method. He is doing modeling, training, and They often give me new ideas when reasoning." These insights from different fields help Elric broaden his horizons in research and avoid being stuck in a single thinking mode.

年 The young talent team with a variety of backgrounds is more cohesive and competitive.

Keep your mind open to new ideas and new achievements, Elric is like a sponge absorbing water, constantly enriching himself. Cundak has built a series of internal knowledge systems. Whether it is about the latest statistical methods, hot machine learning, artificial intelligence algorithms, programming paradigms, the use of tool chains, etc., courses are regularly arranged to help colleagues self-precipitate and achieve Knowledge Sharing.

"Students studying statistics in school still need to expand a considerable amount of knowledge in the field of computer science. Recently, I am working hard to improve my relatively weak engineering abilities. From basic programming skills to understanding the entire ecology of computer science , such as data structures, algorithms, computer composition principles, operating systems, databases, compilation principles, etc. "Elric also spends some time every day reading the latest articles on arXiv and the industry. "Some articles about GNN and Transformer models have been published. It inspired me to maintain my sense of the academic world and not to lose some homework just because I started working. "

Using the latest machine learning technology to solve practical problems and create real and measurable value is the key. The eternal pursuit of technical talents is also their long-term pursuit of ultimate technology.

And Kuande is the "ivory tower" that protects these true intentions. Its exploration of diversity and great tolerance inspire the research motivation of talents in the "post-student era". At the same time, Cundak has equipped this power with mature tools and scientific methods, allowing everyone to devote themselves wholeheartedly to studying the beauty of technology in finance.

I have benefited a lot from exploring the harmony of technology and value

Today, machine learning, deep learning, reinforcement learning and other technologies are greatly reshaping the technical landscape of the quantitative trading industry, providing The development of quantitative trading opens up a new situation. For example, the quantitative trading field will draw extensively on SOTA ML/DL related research. But absorption does not mean appropriation. How to use a certain deep learning technology correctly? When the signal-to-noise ratio is low, how to determine whether it is a limitation of the data or a bias in the training process? These are all adjustments that the quantitative trading industry needs to make to new scenarios and new problems.
You should gradually expand the boundaries of cognition from the fields you are familiar with, lay a solid foundation step by step, and strive to "know what is happening but also why."

With the deepening of work practice, Elric gradually realized that quantification is an organic combination of "specific technical implementation" and "unique value to the market". He quoted the speech delivered by Dr. Feng Xin, founder of Cundak, at the World Artificial Intelligence Conference: "Quantitative investment can establish and utilize extensive correlations among stocks to build the entire stock market into an organic whole with liquidity sharing and impact sharing. From the perspective of machine learning From a perspective, we can embed stocks from a very sparse high-dimensional space such as one-hot encoding to a dense low-dimensional space. This process is called embedding. Based on this, attention (attention mechanism) can be used to study point-to-point relationships between stocks. The relationship between stocks is converted into the relationship between coordinates in this space. This is the core idea of ​​the GAT (Graph Attention Networks) paradigm. After introducing this mechanism, individual stocks are woven into one sheet. Information flows in this network. Single points are often easy to break through, but in this network, each node is strengthened, and the pressure on a certain point will be shared across the entire network. The market also shares liquidity. Under this network, the impact cost of transactions is greatly reduced, making the value of stocks redeemable in the secondary market more stable and sufficient.”

                                                                                                                                                   ^{-                                                                                                            .}

"This example given by Dr. Feng reflects a harmony that integrates elegant technology and its market significance. Through the deep learning method of expressing the relationship between objects like GNN, we can establish a kind of relationship between stocks. The mutually beneficial and complementary relationship has established a more stable and effective structure for the market, as if thousands of stocks have been forged into a whole. "Elric said: "Mr. Feng, like me, has a background in statistics. He has 20 years of quantitative experience at home and abroad. I have benefited a lot from practical experience and accumulated experience in aspects such as model interpretability and how to deal with overfitting. It is in this collision between veterans and young people that Kuande continues to generate sparks of innovation."

In the market competition where every second counts, Kuande has such a group of geeks who are steady and constantly seeking, and they will surely stand at the forefront when convolution strikes.

Highlights of Kuande Hot Recruitment Projects

The above is the detailed content of Hardware war, computing power competition, genius game: game-breakers in the era of quantitative involution. For more information, please follow other related articles on the PHP Chinese website!