Policy optimisation, reward modelling and RLHF — how models are trained by feedback rather than by labels.
arXiv:2608.30719v1 Announce Type: new Abstract: Productive dialogue alignment requires distinguishing \emph{surface coordination} (acknowledgments and smooth task progression) from \emph{epistemic al...
arXiv:2608.29247v1 Announce Type: new Abstract: Deep neural networks remain highly vulnerable to adversarial perturbations, and adversarial training (AT) has become a widely used approach for improvi...
SkillForge is a framework that breaks down formal code synthesis into reusable atomic skills, each handling a specific subtask such as specification inference, body synthesis, invariant generation, error diagnosis, or repair. A verification-driven harness coordinates these skills by submitting candidates to the Dafny verifier, diagnosing failures, and routing them deterministically to the appropriate repair skill until correctness is achieved or a budget is reached. On a curated benchmark, SkillForge outperforms state‑of‑the‑art agentic and iterative baselines, requiring fewer tokens and lower latency, with ablation studies showing each skill’s measurable contribution and rapid convergence.
TEMPO is a unified model that adds temporally‑grounded capabilities to large audio‑language models, enabling timestamping of events, speakers, and sounds in audio, speech, and music. It introduces a supervised fine‑tuning stage featuring atomic timestamp tokens, a time‑aware projector with sinusoidal encodings, and a distance‑aware Gaussian loss, trained via a synthetic‑to‑real curriculum. Additionally, TEMPO employs reinforcement learning (GRPO) as a refinement step, and achieves state‑of‑the‑art performance on a benchmark of 10K samples across five timestamping tasks, surpassing Audio Flamingo Next and Qwen3‑Omni.
arXiv:2606.22394v3 Announce Type: replace Abstract: Consistency distillation has significantly accelerated diffusion-model inference, but its sampling dynamics remain underexplored. We reveal an asym...
arXiv:2607.20372v2 Announce Type: replace Abstract: Humans distill experience into reusable abstractions, e.g., strategies and cautionary reminders, and apply them to gradually solve problems more ef...
arXiv:2608.29461v1 Announce Type: new Abstract: Sandbagging models strategically underperform on evaluations while retaining the capabilities being measured. The evaluations that guide frontier-model...
arXiv:2210.00580v4 Announce Type: replace Abstract: This paper builds bridges between two families of probabilistic algorithms: (hierarchical) variational inference (VI), which is typically used to m...
arXiv:2608.30109v1 Announce Type: new Abstract: Large Language Models (LLMs) trained on extensive scientific research are increasingly integrated as assistants for scientific discovery. However, most...
arXiv:2512.23927v3 Announce Type: replace-cross Abstract: Fitted \(Q\)-iteration (FQI) is a standard regression-based method for optimal control in offline reinforcement learning, but its stability u...
arXiv:2608.29831v1 Announce Type: new Abstract: Localizing issue-relevant code regions is a critical step in automated software engineering. However, due to their reliance on sparse trajectory-level...
The paper introduces LFPG‑RL, a reinforcement‑learning approach that estimates dynamic origin‑destination matrices in real time by integrating link‑flow propagation guidance into proximal policy optimization. LFPG‑RL transforms aggregate link‑flow errors into OD‑specific advantages, enabling a single forward pass during deployment. Evaluated on Melbourne arterial network data, it achieves low RMSE, MAPE, and high correlation, outperforming existing calibration methods.
arXiv:2410.01432v3 Announce Type: replace Abstract: Amortized inference is the task of training a parametric model, such as a neural network, to approximate a distribution with a given unnormalized d...
arXiv:2310.04363v3 Announce Type: replace Abstract: Autoregressive large language models (LLMs) compress knowledge from their training data through next-token conditional distributions. This limits t...
arXiv:2608.31079v1 Announce Type: new Abstract: Sycophantic agreement refers to a behavior in which language models excessively affirm the user, often at the cost of factual accuracy. Although sycoph...
arXiv:2608.30528v1 Announce Type: new Abstract: Reinforcement learning (RL) is used to improve the reasoning abilities of LLMs, while training data span heterogeneous tasks. However, most RL post-tra...
arXiv:2608.30457v1 Announce Type: new Abstract: Multimodal geometry reasoning requires VLMs to extract precise visual relations and preserve them through multi-step deduction. Existing free-form trac...
arXiv:2608.30442v1 Announce Type: new Abstract: Recent offline reinforcement learning (RL) studies report policies that outperform physician decisions on clinical outcomes. We conduct a systematic, p...
arXiv:2605.11467v2 Announce Type: replace-cross Abstract: Reasoning models post-hoc rationalize answers they have already committed to internally, producing chains of *reasoning theater*: deliberativ...
arXiv:2605.06387v4 Announce Type: replace-cross Abstract: On-policy distillation (OPD) trains a student on its own trajectories with token-level teacher feedback and often outperforms off-policy dist...