Project case study
Dyson — Modeling & Algorithm R&D
AI/ML
Financial market models spanning statistical methods through deep neural networks and probabilistic reasoning
TensorFlowKerasScikit-learnstatsmodelsCNNsRNNsTransformersBayesian Neural NetworksARIMA
Evolving collection of models and algorithms for financial markets, including equities, options, futures, and crypto, spanning traditional statistical modeling through deep neural networks and probabilistic reasoning.
Dyson's sequence-based models produce regression and classification predictive inferences of future asset prices, combining deep ensemble learning with custom optimizers and loss functions built for the unique complexities of time-series market data.
Key capabilities:
- Model families spanning CNNs, RNNs, Transformers, Temporal Convolutional Networks, Autoencoders, Bayesian neural networks, ARIMA, and structural time-series models
- Extensive exploratory data analysis (EDA), feature engineering, scaling/normalization, resampling, and principal component analysis (PCA)
- Advanced generalization and regularization techniques that lift model performance beyond the training data, specialized for financial time series