Machine Learning for Crypo-Assets
Download as PPTX, PDF1 like1,170 views
This document discusses using machine learning for analyzing crypto assets. It begins with an introduction to machine learning and the differences between machine learning and statistics. It then discusses using machine learning for scenarios like blockchain deanonymization, investor profiling, factor identification, and price predictions. For each scenario, it provides an example machine learning model and the potential results. The document also discusses challenges for crypto data like lack of labeled datasets and introduces techniques like semi-supervised learning, transfer learning and neural architecture search that are relevant for crypto machine learning.
Machine Learning for Crypo-Assets
- 1. Machine Learning for Crypto-Assets What Works, What Doesn’t and How to Do It Without Going Crazy
- 2. Agenda ❖ A gentle intro to machine learning ❖ Machine learning for crypto assets ❖ Scenarios and solutions
- 3. INTRODUCING There is a lot of hype around machine learning….
- 6. Machine Learning vs. Statistics Machine Learning is an algorithm that can learn from data without relying on rules-based programming. Statistical modeling is a formalization of relationships between variables in the data in the form of mathematical equations.
- 7. INTRODUCING Machine learning programs are able to learn and improve over time…
- 8. Asset-Based Factor-Based Machine Learning vs. AI vs. Deep Learning
- 9. Asset-Based Factor-Based ML and DL in Crypto Asset Scenarios Machine Learning Deep Learning Crypto Analytics Scenarios
- 10. Types of Machine Learning
- 11. The Traditional Way of Thinking
- 12. Reality is More Complicated
- 13. Machine Learning for Crypto Assets
- 15. Asset-Based Factor-Based Social Datasets Blockchain Datasets Exchange Datasets Heterogeneous Datasets
- 16. INTRODUCING Some killer challenges in crypto that can be addressed using machine learning….
- 18. Blockchain Deanonymization The Problem: Classifying blockchain addresses as known actors in the crypto ecosystem (exchanges, miners, OTC). Why is This Relevant: Understanding actors allows us to develop domain- specific analytics. Why is This Machine Learning: Programs need to learn from existing classified addresses in order to identify similar ones.
- 19. Example: Classification for exchange labeling • The input is a collection of blockchain addresses • A classifier is trained in a collection of labeled addresses • A decision tree infers the rules for classifying different exchanges
- 21. Investor Profiling The Problem: Classifying individual wallets based on their blockchain activity. Why is This Relevant: Understanding individual investor behaviors helps to analyze potential market movements. Why is This Machine Learning: Programs need to infer patterns from blockchain activity and drive specific conclusions.
- 22. Example: Clustering for Investor Position Classification • The input is a collection of blockchain addresses and transactions • A clustering models segments the most relevant groups based on proximity • The outputs are clusters of addresses based on similar characteristics
- 24. Factor Identification The Problem: Identifying individual factors that are relevant to market momentum. Why is This Relevant: Individual factors are the cornerstone of quant trading strategies. Why is This Machine Learning: Programs need to learn computations that are relevant to price movements and back test the specific inferences.
- 25. Example: Linear Regression for Factor Identification • The input is a collection of individual attributes • A regression model is trained in matrix of price variations and predictors • A linear regression model infers the relevance of individual factors in price movements in certain market conditions
- 27. Price Predictions The Problem: Predict price trends in crypto assets. Why is This Relevant: Intelligent price predictions is the holy grail of machine learning in capital markets. Why is This Machine Learning: Programs need to understand market activity to predict new price movements.
- 28. Example: CNN+LSTM for Order Book Data • The input is an image of the order book from multiple exchanges • A CNN layers detects features for the order book data • A time distributed layer identifies patterns for a specific timeframe • A LSTM layer aggregates all the individual patterns into a target prediction 28
- 30. INTRODUCING Novel machine learning techniques to consider when analyzing crypto datasets…
- 31. Semi-Supervised Learning • Enables the training of models with minimum datasets • It’s relevant to crypto because the lack of labeled datasets for crypto-assets.
- 32. Transfer Learning • Allows to reuse models across different tasks • It’s relevant to crypto in order to build models that reuse other trained models.
- 33. Neural Architecture Search • Uses machine learning to test different machine learning models for a given problem. • It’s relevant to crypto to discover sophisticated models for specific tasks.
- 34. 34 Summary ● Machine learning will play a key role in the development of market intelligence for crypto assets. ● Blockchain datasets are a unique dataset that can be used by machine learning models. ● The pseudonymous characteristic of the blockchain requires the application of machine learning models to extrapolate meaningful intelligence.