Sparse Rewards Can Self-Train Dialogue Agents (2025.findings-acl)

Copied to clipboard

Challenge: Recent advances in large language models have been driven by supervised fine-tuning and high-quality human feedback. however, acquiring meaningful human feedback has become increasingly challenging and costly.
Approach: They propose a method that empowers LLM agents to enhance their performance without external feedback.
Outcome: The proposed method improves tool-based interactions while preserving general model capabilities across diverse benchmarks.

Similar Papers

Bootstrapping LLM-based Task-Oriented Dialogue Agents via Self-Talk (2024.findings-acl)

Copied to clipboard

Challenge: Large language models (LLMs) are powerful dialogue agents, but specializing them towards fulfilling a specific function can be prohibitive in terms of feasibility, time, and resources.
Approach: They propose a method for training large language models by enabling "self-talk" they propose supervised fine-tuning of LLMs to improve quality of dialogues .
Outcome: The proposed method generates training data via "self-talk" of LLMs that can be refined and utilized for supervised fine-tuning.
Aligning Dialogue Agents with Global Feedback via Large Language Model Multimodal Reward Decomposition (2025.findings-emnlp)

Copied to clipboard

Challenge: a large language model is used to decompose global feedback into a lightweight reward model.
Approach: They propose a large language model based reward decomposition framework for dialogue agents . they use a frozen large language modeling framework to decompose global feedback .
Outcome: The proposed framework infers fine-grained local rewards from a single session-level feedback signal.
Prospector: Improving LLM Agents with Self-Asking and Trajectory Ranking (2024.findings-emnlp)

Copied to clipboard

Challenge: Existing LLMs are limited in their ability to incorporate feedback from an environment.
Approach: They propose an LLM agent that consists of an Actor and a Critic.
Outcome: The proposed agent outperforms existing LLMs on benchmark environments and shows that it can generate diverse trajectories and pick the most rewarding trajectory.
Meta-Rewarding Language Models: Self-Improving Alignment with LLM-as-a-Meta-Judge (2025.emnlp-main)

Copied to clipboard

Challenge: Existing methods for improving large language models have focused on improving model responses rather than judgment capabilities, resulting in rapid saturation during iterative training.
Approach: They propose an iterative Meta-Rewarding step where the model judges its own judgements and uses that feedback to refine its judgment skills.
Outcome: The proposed model improves Llama-3-8B-Instruct from 22.9% to 39.4% on AlpacaEval 2 and 20.6% to 29.1% on Arena-Hard.
Demystifying Mixed Outcomes of Self-Training: Pre-training Analyses on Non-Toy LLMs (2026.findings-eacl)

Copied to clipboard

Challenge: Recent studies on self-training report seemingly contradictory outcomes.
Approach: They use OLMo-2 models as non-toy LLMs and perform multiple rounds of continual pre-training using self-generated text with different prompting strategies and data filtering.
Outcome: The proposed model collapse is inherent to the training procedure itself, while self-improvement is likely owes its success to human-designed, strategic synthetic pipelines that inject external intelligence.
Dynamic Rewarding with Prompt Optimization Enables Tuning-free Self-Alignment of Language Models (2024.emnlp-main)

Copied to clipboard

Challenge: Empirical evaluations on eight recent LLMs reveal that DRPO significantly enhances alignment performance, enabling base models to outperform their SFT/RLHF-tuned counterparts.
Approach: They propose a tuning-free approach to self-alignment called Dynamic Rewarding with Prompt Optimization (DRPO) it leverages a dynamic rewarding mechanism to identify and rectify alignment weaknesses .
Outcome: The proposed approach outperforms existing methods and is highly adaptable to various alignment challenges.
Global Reward to Local Rewards: Multimodal-Guided Decomposition for Improving Dialogue Agents (2024.emnlp-main)

Copied to clipboard

Challenge: Existing methods for asynchronous dialogue agents only use a single global score at the end of the session.
Approach: They propose a method for aligning an LLM-based dialogue agent for long-term social dialogue . they use local implicit feedback to decompose a human-provided global Explicit reward .
Outcome: The proposed approach improves the turn-level utterance generation across conversational metrics compared to baseline methods.
Teaching LLM to be Persuasive: Reward-Enhanced Policy Optimization for Alignment from Heterogeneous Rewards (2026.acl-industry)

Copied to clipboard

Challenge: a large language model (LLM) is used as a business development agent for persuasive price negotiation in online travel agencies.
Approach: They propose a reward-enhancing policy optimization method that integrates three complementary reward sources-a preference-trained reward model and an LLM-as-a-judge.
Outcome: The proposed method improves average dialogue rating to 4.63 (+0.33 over GRPO) and raises share of conversations with at least one excellent response to 66.67% (+23.34 pp over grepo).
ReSo: A Reward-driven Self-organizing LLM-based Multi-Agent System for Reasoning Tasks (2025.emnlp-main)

Copied to clipboard

Challenge: Multi-agent systems (MAS) are limited by poor flexibility and scalability, with underdeveloped optimization strategies.
Approach: They propose a task graph generation and a reward-driven two-stage agent selection process to integrate multi-agent systems to improve their reasoning capabilities.
Outcome: The proposed model outperforms existing methods on Math-MAS and SciBench-MAS SciBech, while other methods completely fail.
AgentGym: Evaluating and Training Large Language Model-based Agents across Diverse Environments (2025.acl-long)

Copied to clipboard

Challenge: Large language models (LLMs) are promising foundations to build generally-capable agents . however, the community lacks a unified interactive framework that covers diverse environments for comprehensive evaluation of agents.
Approach: They propose a framework that features 7 real-world scenarios, 14 environments, and 89 tasks for unified, real-time, and concurrent agent interaction.
Outcome: The proposed framework features 7 real-world scenarios, 14 environments, and 89 tasks for unified, real-time, and concurrent agent interaction.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations