Lecture 2 (Bergmann) Flashcards
What is Machine learning?
it’s a subset from computer science focusing on developing algorithms that lean from a dataset to solve problems.
What is a “black box”, and what is the problem with it?
It contradicts the scientific method, which involves hypothesis formulaton. The black box is unknown.
State the Historical AI approaches.
Making a mind: focused on pre programming algorithms based on simplifications. Failed due to underestimation of fomalizing intellignce
Modelling the brain: Centered around neural networks without predefined rules.
What are the three learning Architectures
Supervised: Used for Tasks like image calssification
Unsupervised: Applied clustering and detecting complex relationships. Frauds and stock detections.
Reinforcement learning: Stocktrading, AlphaGo Zero and auto driving.
What are the characteristics of ML task?
Suited for complex, unstructured decision making problems accounting for subtle statistical patterns.
What are the limitations of ML?
Not completely representative of future data scenarios
What are the ML good at and bad at?
It produces accurate predictions for complex tasks, however, lack the explainability.
What is chain of though prompting:
A method to make LLM demonstrate their reasoning process
Define Deep hedging.
A method using neural networks to develop hedging strategies in financial model, Particularly useful in complex, High-dimensional scenarios
What is neural network Parametrization?
It refers to Networks processing various inputs like current asset prices and past strategies at each time point, to increase the hedging
When is Delta Hedging vital?
when Traditional methods, like Delta heding are ineffective. Especially in model with multiple assets, transaction costs and incomplete markets.
Define optimization problems.
:Involves minimizing the difference between the payoff of an option and the return from hedging strategy. Depicted by a loss fucntion involving stochastic integrals.
How is training data supplied and why is it needed?
Generated from sample paths of asset prices, this data is used to train the neural networks to optimze the hedging strategy.
What is black scholes model and what does it show?
Shows that deep heding can be applied in a standard model setup. With neural network-based strategy.
