Research map / Reinforcement learning / Policy gradient and actor-critic
Deep Recurrent Optimal Stopping
NIRANJAN DAMERA VENKATA, Chiranjib Bhattacharyya
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Fit the stopping policy directly (OSPG) instead of fitted Q-iteration with a DNN | DNN · Bermudan max-call d=20 p0=100 | average return: 50.92 -> 51.02 | Improved |
| Use a recurrent policy instead of feed-forward on Markovian data | OSPG · Bermudan max-call d=100 p0=100 | average return: 82.96 -> 82.74 | No clear effect |
| Use RRLSM with fixed random RNN features instead of OSPG | RNN · Bermudan max-call d=20 p0=100 | average return: 50.99 -> 49.68 | Hurt |
| Use recurrent OSPG instead of DNN-OSPG on non-Markovian fBm | OSPG · Fractional Brownian motion h=0.05 | average return: 1.14 -> 1.28 | Improved |
| Use OSPG instead of FQI with an RNN on fBm | RNN · Fractional Brownian motion h=0.80 | average return: 0.18 -> 0.27 | Improved |
| Extend DOS state with process history (DOS-ES) | DOS · Fractional Brownian motion h=0.05 | average return: 0.70 -> 1.18 | Improved |
| Directly fit policy (DNN-OSPG) vs value-recursion methods DOS and DNN-FQI | DNN · Fractional Brownian motion h=0.05 | average return: 0.87 -> 1.14 | Improved |
| Use DNN-OSPG instead of DNN-FQI on American geometric-call option | DNN · American geometric-call d=7 s0=110 | average return: 15.0487 -> 15.9699 | Improved |
| Use OSPG versus Longstaff-Schwartz at higher asset count | DNN-OSPG · American geometric-call d=13 s0=90 | average return: 5.5962 -> 5.7436 | Mixed |
| Use RNN-OSPG instead of RRLSM for early stopping classifier | RNN · UCR CBF early classification | average cost: 0.192 -> 0.186 | Improved |
| Use RNN-OSPG instead of RNN-FQI for early stopping | RNN · UCR CBF early classification | average cost: 0.308 -> 0.186 | Improved |
| Share parameters across time-steps (DNN-OSPG) vs DOS per-step networks | DNN · Bermudan max-call d=20 | model size (params): 9,225 -> 1,047 | Improved |
Closest papers by meaning
- Learning to Stop: Deep Learning for Mean Field Optimal Stopping (ICML 2025 · similarity 0.88)
- Efficient Recurrent Off-Policy RL Requires a Context-Encoder-Specific Learning Rate (NeurIPS 2024 · similarity 0.86)
- When to Learn and When to Stop: Quitting at the Optimal Time (Student Abstract) (AAAI 2025 · similarity 0.85)
- Deep Learning for Computing Convergence Rates of Markov Chains (NeurIPS 2024 · similarity 0.85)
- Recurrent Reinforcement Learning with Memoroids (NeurIPS 2024 · similarity 0.85)
- A Study of Bayesian Neural Network Surrogates for Bayesian Optimization (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
