Linganisha mbinu
Pitia mbinu ulizochagua bega kwa bega; safu zinazotofautiana zinaangaziwa.
| Ujifunzaji wa Kuimarisha Uliosafishwa (Fine-Tuned Reinforcement Learning)× | Ujifunzaji wa Uimarishaji Unaojisimamia Kwenyewe× | |
|---|---|---|
| Nyanja | Ujifunzaji wa Kina | Ujifunzaji wa Kina |
| Familia | Machine learning | Machine learning |
| Mwaka wa asili≠ | 2017–2022 | 2020 |
| Mwanzilishi≠ | Christiano, P. et al.; Ouyang, L. et al. | Laskin, M.; Srinivas, A.; Abbeel, P. (and contemporaries) |
| Aina≠ | Policy adaptation via fine-tuning | Self-supervised auxiliary-task learning for RL |
| Chanzo asilia≠ | Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., Schulman, J., Hilton, J., Kelton, F., Miller, L., Simens, M., Askell, A., Welinder, P., Christiano, P., Leike, J., & Lowe, R. (2022). Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, 35, 27730–27744. link ↗ | Laskin, M., Srinivas, A., & Abbeel, P. (2020). CURL: Contrastive Unsupervised Representations for Reinforcement Learning. Proceedings of the 37th International Conference on Machine Learning (ICML), PMLR 119, 5639–5650. link ↗ |
| Majina mbadala | RL fine-tuning, policy fine-tuning, RLHF, reinforcement learning from human feedback | SSL-RL, self-supervised RL, representation-based reinforcement learning, auxiliary-task RL |
| Zinazohusiana≠ | 5 | 4 |
| Muhtasari≠ | Fine-Tuned Reinforcement Learning adapts a pre-trained policy or model to a new task or behavioral objective using reinforcement signals — including human feedback — rather than retraining from scratch. Popularized by RLHF, it is the core technique behind aligning large language models and adapting deep RL agents to specialized environments with minimal additional data. | Self-supervised Reinforcement Learning (SSL-RL) augments standard RL training with self-supervised auxiliary objectives — such as contrastive, predictive, or data-augmentation-based tasks — applied to the agent's own experience. These objectives improve the quality of learned representations without requiring extra human labels, enabling faster convergence and better sample efficiency, especially in high-dimensional observation spaces like raw pixels. |
| ScholarGateSeti ya data ↗ |
|
|