Arishi Orra

Proposes DARL, integrating denoising diffusion probabilistic models with DRL to generate synthetic stress scenarios and enhance robustness of portfolio strategies.

Introduces an adaptive DRL framework that balances exploitation of an existing investment universe with exploration of new opportunities using two cooperating DRL agents.

Develops a risk-adjusted DRL approach combining three DRL agents trained with different reward functions (log returns, differential Sharpe ratio, maximum drawdown) and fuses their actions via a CNN to produce a unified risk-aware policy; tested across multiple market indices.

Proposes a volatility-guided DRL framework that uses GARCH-based volatility forecasting to pre-select assets according to investor risk profiles (aggressive/moderate/conservative) and trains a DRL agent to learn investment policies; evaluated on Dow 30 stocks.

Investigates reward shaping via expert feedback to improve DRL-based stock trading.

Presents a dynamic reinforced ensemble method for stock trading using Bayesian optimization.