ActiveSAM is a training-free, zero-shot framework that enhances SAM 3 for open-vocabulary semantic segmentation by identifying an image-conditioned active class set. It improves speed-accuracy tradeoff, outperforming SegEarth-OV3 by +1.4 mIoU on average and running up to 5.5x faster on large-vocabulary datasets, with strong robustness to image corruption.
ExpRL introduces a novel mid-training approach for LLMs using human-written question-answer data as reward scaffolds. Instead of imitating reference solutions, it constructs problem-specific grading rubrics to reward intermediate reasoning steps, enabling better initialization for sparse-reward RL and outperforming SFT, sparse-reward GRPO, and self-distillation on math reasoning tasks.
Hierarchical Advantage-Weighted Behavior Cloning (HABC) enhances online RL fine-tuning of vision-language agents by using separate critic heads for viability and efficiency. It combines their outputs via a state-adaptive gate and applies per-transition weights, while intervention-aware credit assignment prevents supervision leakage. In real-robot experiments, HABC boosts success rates to 92%, 88%, and 38% on three bimanual tasks, surpassing SFT baselines of 36%, 44%, and 12%.
HalBench evaluates 29 open-source LLMs on a custom benchmark for sycophancy and hallucination. Qwen 3.6 and Gemma 4 outperform larger models, with Qwen 3.6 achieving 36.6% pushback—higher than GPT-5.4 and Gemini 3.1 Pro. Model size does not correlate with honest responses, indicating that architecture and training data matter more than parameters.
Public AI evaluation archives show that a single terminal result can arise from two distinct pre-terminal histories, with estimated times to reach 95% of performance ceilings at 23.03 or 75.13. A candidate selection-aware frontier model fails synthetic recovery and uncertainty calibration, and is rejected by fixed audit gates. An archive-and-adjudication protocol verifies timing boundaries and falsifies unsupported frontier claims.
TuneJury is an open, instance-level pairwise reward model that predicts music preference scores from text prompts and audio clips. It is trained on diverse human-preference data and demonstrates strong generalization, with anchor calibration enabling efficient post-hoc alignment for music generation systems.
TokenPilot reduces inference costs by 61% to 87% in both isolated and continuous modes, outperforming prior systems in cost efficiency while maintaining competitive performance. It uses ingestion-aware compaction and lifecycle-aware eviction to preserve prompt cache continuity and minimize token footprint without introducing prefix mismatches.
Image classifiers like PRISM2D, GFNet, and ViT-B/16 show that phase, not magnitude, drives predictions in hidden layers. ResNet-50 reveals a latent sign code in late blocks, indicating phase/sign identity exists across architectures, though expressed differently due to activation and readout mechanisms.
Factorized Neural Operators (FaNO) decompose spectral representations into equivariant dynamic and invariant persistent responses. This factorized structure enables better interpretability, generalization, and consistent predictions across scales, domains, and physical regimes.
CEAP, a new circuit discovery method, substantially reduces resampling variance compared to EAP-IG. The paper shows that rephrasing variance arises from prompt templates activating different circuits, suggesting LLMs are inherently hard to steer across diverse inputs. Sample-wise variance is largely benign, as poor unfaithfulness scores result from selective contribution scaling, not circuit defects.
We propose a new functional gradient descent algorithm that adapts its representation during optimization. The method achieves convergence to a stationary point under smooth losses and to a global minimizer under smoothness and a Polyak-Lojasiewicz condition, despite using finite-dimensional approximations. It outperforms both fixed-approximation FGD and neural network baselines in regression, PDE solving, and computer vision tasks.
This paper proposes a unified causal-origin taxonomy for distributional shifts in reinforcement learning, linking ID/OOD generalization to non-stationary settings. It decomposes the agent-environment interaction using a POMDP framework, identifying internal, agent-driven, and external, environment-driven shifts, with explicit, implicit, and hybrid types defined by the shifted-time boundary. The work introduces an evaluation framework to measure shift impact through performance degradation and recovery metrics, enabling systematic analysis of RL robustness.
CircuitLasso enables scalable circuit learning in large language models by using sparse linear regression. It recovers circuits with structural accuracy matching state-of-the-art methods at significantly lower computational cost, and demonstrates human-interpretable semantic propagation through model components. The learned circuits achieve comparable performance on a domain-generalization task with reduced cost.
A model-agnostic auditing framework detects and distinguishes true and phantom disclosures in synthetic data. It uses only synthetic outputs and a held-out control set to perform statistical testing, offering tighter privacy leakage bounds than prior methods without requiring model access or additional training.
A residual learning approach using directional task-error supervision achieves stable five-ball juggling on real robots, converging from the second attempt. The system outperforms human practice timelines and relies on both directional feedback and an informative prior, with a fixed-Jacobian Newton update proving most reliable.
A measurement study finds that 26 semantic post-hoc operators do not improve held-out accuracy over Best-of-N in frozen small code models. While some operators reduce compute usage or recover correct programs, none outperform BoN in accuracy, due to systemic limitations like coverage walls and consensus traps. An expression-layer recovery (M1) improves performance on HumanEval+ by 12 tasks, with no harm or leakage, and shows consistent results across model cells.
TuneJury is an open, instance-level pairwise reward model that predicts music preference scores from text prompts and audio clips. It is trained on diverse human-preference data and demonstrates strong generalization, with anchor calibration enabling efficient post-hoc alignment for music generation systems.
TokenPilot reduces inference costs by 61% to 87% in both isolated and continuous modes, outperforming prior systems in cost efficiency while maintaining competitive performance. It uses ingestion-aware compaction and lifecycle-aware eviction to stabilize prompt prefixes and manage context segments efficiently.
KVEraser enables efficient localized context erasing in large language models by replacing only the KV cache states of an erased span with learned steering states. It achieves near-full-recomputation performance on in-domain tasks and offers a 24% latency increase versus a 17.6x increase for full recomputation, with up to 3--4x speedup on long-document QA tasks.
A new attack, RING, exploits differential privacy in federated learning to conceal backdoor signals while maximizing impact. It achieves 90.3% attack success against state-of-the-art defenses, up to 26.08x over baseline methods, and reveals a critical security gap in DP-FL due to inherent masking of malicious updates.