Sophon-3: AI that keeps learning
Explore Finforge’s research into continually learning models for screening markets, forecasting returns and constructing portfolios.
Markets change. Finforge’s Sophon-3 research explores models that keep learning as new data arrives, connecting market screening, forecasting and portfolio construction.
From signals to portfolios
Sophon-3 brings specialized models into a shared investment workflow. A screener narrows the asset universe. Where included, a forecast model then evaluates the selected candidates before the results flow into portfolio construction.
The agents use different combinations of inputs and model nodes. Their configurations offer concrete examples of how the research can be applied.
Methods in the open
Finforge publishes its research so that the approach and its limitations can be read, questioned and evaluated. The full paper is titled Sophon-3: Continual-Learning Deep Models for Non-Stationary Financial Markets.
This overview introduces the research described on Finforge’s website. Consult the paper for the full methodology, evidence and limitations.
Explore the agents
- Sophon Apex — current ratio and revenue.
- Sophon Core — current assets and solvency.
- Sophon Pillar — company equity.
- Sophon Surge — price and price-to-book.