Quantum Deep Hedging vs BloombergGPT

1.
GPT-4를 두고 전 세계가 떠들썩 합니다. AGI의 위협으로부터 인류사회를 보호하기 위하여 ‘멈추자’고 하는 분들과 ‘그렇지않다’라고 하는 분들이 논쟁중입니다. 저도 ChatGPT로 이런 저런 시도를 시험을 해봅니다.

최근 몇 년동안 인류에게 도전이면서 위기인 것이 두가지입니다. 첫째 양자역학입니다. 유튜브를 보면 양자역학을 다루는 영상들이 무척 많습니다. 최근에 본 영상에서는 양자충첩을 설명하면서 스타트렉에서 볼 수 있는 ‘순간이동’의 가능성을 설명합니다.
둘째는 인공지능입니다. ChatGPT가 대표적인입니다. 한동안 기계학습을 주제로 다양한 방법들을 소개하였는데 ChatGPT이후 Large-scale Language Model이 모든 것을 집어 삼기는 듯 합니다.

이상의 흐름이 금융과 결합할 때 어떤 변화가 등장할지를 다루는 두가지 논문이 나왔습니다.

첫째는 Quantum Deep Hedging

Download (PDF, 857KB)

이미 JP Morgan은 Deep Hedging을 개발하여 업무에 적용하였습니다.

JP Morgan의 Deep Hedging
JP Morgan의 Deep Hedging 둘째

위에서 사용했던 방법을 양자컴퓨터를 이용한 모델로 발전시키기 위한 모델을 만들었다고 합니다.

In our work we look at the problem of hedging where deep reinforcement learning offers a powerful framework for real markets. We develop quantum reinforcement learning methods based on policy-search and distributional actor-critic algorithms that use quantum neural network architectures with orthogonal and compound layers for the policy and value functions. We prove that the quantum neural networks we use are trainable, and we perform extensive simulations that show that quantum models can reduce the number of trainable parameters while achieving comparable performance and that the distributional approach obtains better performance than other standard approaches, both classical and quantum. We successfully implement the proposed models on a trapped-ion quantum processor, utilizing circuits with up to 16 qubits, and observe performance that agrees well with noiseless simulation. Our quantum techniques are general and can be applied to other reinforcement learning problems beyond hedging

저자중 El Amine Cherrat은 퀀텀컴퓨터를 이용한 강화학습모델을 계속 연구한 분입니다. 그리고 JP Morgan소속인데 Ben Wood는 Deep Hedging 논문을 발표했던 사람이기도 합니다. El Amine Cherrat는 Quantum Reinforcement Learning via Policy Iteratio을 주제로 논문을 발표하였습니다.

Download (PDF, 379KB)

ChatGPT에 위 논문에 대해 물어보았습니다. 알고리즘 트레이딩이나 프라이싱에 사용할 수 있다고 평가합니다. 물론 양자컴퓨터가 있어야 검증합니다.(^^)

Q:Can the quantum techniques developed for hedging be applied to other reinforcement learning problems, and if so, what are some potential applications?

A:Yes, the quantum techniques developed for hedging can be applied to other reinforcement learning problems beyond hedging. The proposed quantum reinforcement learning methods are general and can be used for other problems such as algorithmic trading or option pricing. The paper suggests that it would be interesting to develop specific quantum methods for such problems. It is important to note that in these use cases, the training data can be produced efficiently, removing the bottleneck of loading large amounts of data onto the quantum computer.

2.
대용량언어모델(LLM)을 이요한 금융서비스가 새롭게 나왔습니다. 출시한 곳은 Bloomberg입니다.

Bloomberg’s ML Product and Research group collaborated with the firm’s AI Engineering team to construct one of the largest domain-specific datasets yet, drawing on the company’s existing data creation, collection, and curation resources. As a financial data company, Bloomberg’s data analysts have collected and maintained financial language documents over the span of forty years. The team pulled from this extensive archive of financial data to create a comprehensive 363 billion token dataset consisting of English financial documents.

This data was augmented with a 345 billion token public dataset to create a large training corpus with over 700 billion tokens. Using a portion of this training corpus, the team trained a 50-billion parameter decoder-only causal language model. The resulting model was validated on existing finance-specific NLP benchmarks, a suite of Bloomberg internal benchmarks, and broad categories of general-purpose NLP tasks from popular benchmarks (e.g., BIG-bench Hard, Knowledge Assessments, Reading Comprehension, and Linguistic Tasks). Notably, the BloombergGPT model outperforms existing open models of a similar size on financial tasks by large margins, while still performing on par or better on general NLP benchmarks.
Introducing BloombergGPT, Bloomberg’s 50-billion parameter large language model, purpose-built from scratch for finance중에서

오랜 동안 금융과 관련한 데이타를 축정한 블룸버그이기 때문에 가능한 작업입니다. 위 설명중 제가 모르는 단어가 있네요. Token Data… 무슨 의미일까 찾아보았습니다.

Token classification is a natural language understanding task in which a label is assigned to some tokens in a text.

토큰화는 민감한 데이터를 고유한 식별 데이터로 마스킹 또는 대체하지만, 데이터에 대한 필수적인 정보를 유지한다. 이런 동등한 고유 대체 데이터를 토큰이라 부른다.

아래와 같은 예제 설명도 있네요.

토큰의 기준을 단어(word)로 하는 경우, 단어 토큰화(word tokenization)라고 합니다. 다만, 여기서 단어(word)는 단어 단위 외에도 단어구, 의미를 갖는 문자열로도 간주되기도 합니다.예를 들어보겠습니다. 아래의 입력으로부터 구두점(punctuation)과 같은 문자는 제외시키는 간단한 단어 토큰화 작업을 해봅시다. 구두점이란 마침표(.), 컴마(,), 물음표(?), 세미콜론(;), 느낌표(!) 등과 같은 기호를 말합니다.

입력: Time is an illusion. Lunchtime double so!

이러한 입력으로부터 구두점을 제외시킨 토큰화 작업의 결과는 다음과 같습니다.

출력 : “Time”, “is”, “an”, “illustion”, “Lunchtime”, “double”, “so”

이 예제에서 토큰화 작업은 굉장히 간단합니다. 구두점을 지운 뒤에 띄어쓰기(whitespace)를 기준으로 잘라냈습니다. 하지만 이 예제는 토큰화의 가장 기초적인 예제를 보여준 것에 불과합니다.

보통 토큰화 작업은 단순히 구두점이나 특수문자를 전부 제거하는 정제(cleaning) 작업을 수행하는 것만으로 해결되지 않습니다. 구두점이나 특수문자를 전부 제거하면 토큰이 의미를 잃어버리는 경우가 발생하기도 합니다. 심지어 띄어쓰기 단위로 자르면 사실상 단어 토큰이 구분되는 영어와 달리, 한국어는 띄어쓰기만으로는 단어 토큰을 구분하기 어렵습니다. 그 이유는 뒤에서 언급하겠습니다.
딥 러닝을 이용한 자연어 처리 입문 02. 텍스트 전처리(Text preproc… 02-01 토큰화(Tokenization) 중에서

다시금 블룸버그 자료로 가보면 700 billion tokens라는 표현이 나옵니다. 엄청납니다. 여기에 50-billion parameter라고 설명합니다.

블룸버그는 이상의 결과물을 BloombergGPT: A Large Language Model for Finance로 발표하였습니다.

Download (PDF, 1.07MB)

예전에 금융의 선진기법이라고 하면 상품의 설계를 이야기하는 경우가 많았습니다. 이제는 선진기법=과학기술입니다. 격차가 느껴집니다.

Leave a Comment

이메일 주소는 공개되지 않습니다. 필수 필드는 *로 표시됩니다

이 사이트는 스팸을 줄이는 아키스밋을 사용합니다. 댓글이 어떻게 처리되는지 알아보십시오.