arXiv AI
Jul 21
Normalized Rewards for Preference Optimization
arXiv:2607. 16240v1 Announce Type: cross Abstract: Direct Alignment Algorithms (DAAs) such as DPO have become a common way to post-train and align LLMs with human preferences.
Truncate Bad, Upweight Good: BoN-Style Distillation via Rank-Based Classification
arXiv:2608. 19748v1 Announce Type: cross Abstract: Inference-time selection methods, such as Best-of-N, improve generation by sampling a pool of candidates and selecting the top-ranked completion according to a reward model.
Related stories
arXiv Machine Learning
Aug 11
CODS: Iterative Bellman-Residual Data Selection for Reusable Offline Reinforcement Learning
arXiv:2608. 07719v1 Announce Type: new Abstract: Offline reinforcement learning repeatedly trains policies from a fixed transition pool, making redundant data costly across seeds and hyperparameters, while naive subsampling can remove rare transitions needed for long-horizon credit assignment.
arXiv AI
Sep 3
On-Policy Distillation Meets Off-Policy GRPO: Training Compact Instruction-Following Rerankers
The paper introduces a two‑stage training framework for compact instruction‑following rerankers. Stage 1 strengthens a 4B teacher reranker with off‑policy GRPO using LLM‑judge feedback on 88K examples, while Stage 2 trains a 1B student by sampling its own rankings and receiving soft teacher‑derived rewards, blending exploration with knowledge transfer. The method achieves superior nDCG and MRR scores on MAIR‑11 and MAIR‑Full benchmarks, outperforming offline distillation baselines and larger RL‑trained rerankers.
Hugging Face Trending Papers
Sep 1
On-Policy Distillation Meets Off-Policy GRPO: Training Compact Instruction-Following Rerankers
Compact instruction-following rerankers are attractive for deployment, but conventional distillation pipelines typically train students by offline imitation of teacher outputs on a fixed set of exampl...
arXiv Machine Learning
Jul 31
LoRA Scaffolded Policy Optimization (LSPO): A Sampling-Time Low-Rank Scaffold for Recovering Reinforcement-Learning Gradient on Zero-Reward Cliff Prompts
arXiv:2607. 27787v1 Announce Type: new Abstract: Reinforcement learning from verifiable rewards (RLVR) for mathematical reasoning suffers from a structural blind spot: on "cliff" prompts-those on which every sampled rollout in a group fails-the group-normalized advantage is identically zero, so GRPO produces no gradient on precisely the prompts at the frontier of the model's capability.
arXiv AI
Jun 17
Beyond the Sampled Token: Preserving Candidate Support in RLVR
arXiv:2510. 14807v3 Announce Type: replace Abstract: We revisit exploration collapse in reinforcement learning with verifiable rewards (RLVR), from the perspective of the \emph{candidate distribution} for next-token prediction.
arXiv AI
2d ago
RankBuffer: Efficient Ranking-Based Rewards for Open-Ended Generation
arXiv:2609.36652v1 Announce Type: new Abstract: Open-ended generation lacks canonical answers, making pointwise rewards difficult to calibrate for group-based reinforcement learning. Directly ranking...
Hugging Face Trending Papers
Jul 30
LoRA Scaffolded Policy Optimization (LSPO): A Sampling-Time Low-Rank Scaffold for Recovering Reinforcement-Learning Gradient on Zero-Reward Cliff Prompts
Reinforcement learning from verifiable rewards (RLVR) for mathematical reasoning suffers from a structural blind spot: on "cliff" prompts-those on which every sampled rollout in a group fails-the group-normalized advantage is identically zero, so GRPO produces no gradient on precisely the prompts at the frontier of the model's capability. We introduce LoRA Scaffolded Policy Optimization (LSPO), a sampling-time mechanism that recovers this lost gradient.
arXiv Computation and Language
2d ago
AdaTutoRank: Learning to Rerank Document Sets via Adaptive Tutoring Optimization for RAG and Deep Research
AdaTutoRank introduces a setwise document reranker that uses Adaptive Tutoring Optimization (ATO) to provide graded supervision across nine rubric dimensions. By generating hint‑based silver labels, reinforcement rewards, and distillation cues tailored to each rollout’s quality, the method improves credit assignment for individual documents within a set. Experiments on ten benchmarks covering Retrieval‑Augmented Generation (RAG), deep research, and setwise evaluation show that AdaTutoRank achieves superior overall performance while reducing the number of retrieval calls.
arXiv Computation and Language
Aug 27
Less can be More: Relieving RAG Bottlenecks via Evidence Frontloading and Pressure-Adaptive Budgeting
The paper introduces “PACE”, a training‑free framework that tackles bottlenecks in Retrieval‑Augmented Generation by frontloading evidence and adaptively budgeting reranking. It first reorders candidate documents based on marginal evidence coverage—prioritizing query‑relevant, complementary, and chain‑forming documents—providing a $(1-1/e)$ approximation guarantee. Then it dynamically adjusts the reranking budget according to the relative pressure of the reranker and the language model, improving evidence recall and reducing p95 latency in multi‑hop QA workloads.
arXiv Machine Learning
Jul 15
Beyond Logit Adjustment: A Residual Decomposition Framework for Long-Tailed Reranking
arXiv:2604. 01506v2 Announce Type: replace Abstract: Long-tailed classification, where a small number of frequent classes dominate many rare ones, remains challenging because models systematically favor frequent classes at inference time.
arXiv Machine Learning
Jun 2
BLISS: A Lightweight Bilevel Influence Scoring Method for Data Selection in Language Model Pretraining
arXiv:2510. 06048v4 Announce Type: replace Abstract: Effective data selection is essential for pretraining large language models (LLMs), enhancing efficiency and improving generalization to downstream tasks.