dpo
Provably Mitigating Overoptimization in RLHF: Y our SFT Loss is Implicitly an Adversarial Regularizer
Then it fine-tunes the LLM to maximize the learned reward using RL techniques.
Country:
- Asia > Middle East > Jordan (0.04)
- North America > United States > California > Santa Clara County > Palo Alto (0.04)
- Asia > Myanmar > Tanintharyi Region > Dawei (0.04)
Genre:
- Research Report > New Finding (1.00)
- Research Report > Experimental Study (1.00)
Technology:
- Information Technology > Artificial Intelligence > Representation & Reasoning (1.00)
- Information Technology > Artificial Intelligence > Natural Language > Large Language Model (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Learning Graphical Models (0.93)
- Information Technology > Artificial Intelligence > Machine Learning > Neural Networks > Deep Learning (0.85)
Genre:
- Research Report > Experimental Study (0.93)
- Research Report > New Finding (0.67)
Technology:
- Information Technology > Artificial Intelligence > Representation & Reasoning (1.00)
- Information Technology > Artificial Intelligence > Natural Language > Large Language Model (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Neural Networks > Deep Learning (0.96)
- Information Technology > Artificial Intelligence > Natural Language > Chatbot (0.70)
Country:
- South America > Argentina (0.04)
- North America > United States > Virginia (0.04)
- South America > Uruguay (0.04)
- (5 more...)
Genre:
- Research Report > Experimental Study (1.00)
- Research Report > New Finding (0.93)
Technology:
- Information Technology > Artificial Intelligence > Representation & Reasoning (1.00)
- Information Technology > Artificial Intelligence > Natural Language > Large Language Model (1.00)
- Information Technology > Artificial Intelligence > Natural Language > Chatbot (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Neural Networks > Deep Learning (1.00)
Genre:
- Research Report > Experimental Study (1.00)
- Research Report > New Finding (0.67)
Industry:
- Health & Medicine > Pharmaceuticals & Biotechnology (1.00)
- Health & Medicine > Therapeutic Area > Immunology (1.00)
Technology:
- Information Technology > Artificial Intelligence > Representation & Reasoning (1.00)
- Information Technology > Artificial Intelligence > Natural Language > Large Language Model (1.00)
- Information Technology > Artificial Intelligence > Natural Language > Chatbot (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Neural Networks > Deep Learning (1.00)
Stepwise Alignment for Constrained Language Model Policy Optimization Akifumi Wachi Thien Q. Tran Rei Sato Takumi Tanabe Y ouhei Akimoto L Y Corporation University of Tsukuba
Safety and trustworthiness are indispensable requirements for real-world applications of AI systems using large language models (LLMs).
Country:
- Asia > Japan > Honshū > Kantō > Ibaraki Prefecture > Tsukuba (0.40)
- North America > United States > California (0.14)
- North America > United States > Alaska (0.04)
- (10 more...)
Industry:
Technology:
- Information Technology > Artificial Intelligence > Natural Language > Large Language Model (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Neural Networks > Deep Learning (0.70)
- Information Technology > Artificial Intelligence > Representation & Reasoning > Optimization (0.68)
Preference Learning Algorithms Do Not Learn Preference Rankings
Preference learning algorithms (e.g., RLHF and DPO) are frequently used to steer LLMs to produce generations that are more preferred by humans, but our understanding of their inner workings is still limited.
Genre:
- Research Report > Experimental Study (1.00)
- Research Report > New Finding (0.93)
Industry:
- Media (0.92)
- Information Technology (0.67)
- Leisure & Entertainment (0.67)
Technology:
Author Contributions
A.1 Deriving the Optimum of the KL-Constrained Reward Maximization Objective In this appendix, we will derive Eq. 4. Analogously to Eq. 3, we optimize the following objective: max
Country:
- Asia > Japan (0.05)
- North America > United States > Virginia (0.04)
- North America > United States > Hawaii (0.04)
- (5 more...)
Industry:
- Government (1.00)
- Education (0.93)
- Law (0.69)
Technology:
Country:
- Oceania > Australia > Victoria > Melbourne (0.04)
- North America > United States > Oregon > Multnomah County > Portland (0.04)
- North America > United States > California > Santa Clara County > Palo Alto (0.04)
- (5 more...)
Technology:
- Information Technology > Artificial Intelligence > Representation & Reasoning (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Neural Networks > Deep Learning (1.00)
- Information Technology > Artificial Intelligence > Natural Language > Large Language Model (0.97)
- Information Technology > Artificial Intelligence > Machine Learning > Reinforcement Learning (0.96)
Country:
- North America > United States > Pennsylvania > Allegheny County > Pittsburgh (0.04)
- North America > United States > Illinois > Cook County > Chicago (0.04)
- Asia > Middle East > Jordan (0.04)
Technology:
- Information Technology > Artificial Intelligence > Machine Learning > Neural Networks > Deep Learning (0.98)
- Information Technology > Artificial Intelligence > Natural Language > Large Language Model (0.95)
- Information Technology > Artificial Intelligence > Natural Language > Chatbot (0.69)
- (2 more...)