SALT uses subspace-aligned training to recover high-rank accuracy with ultra-low-rank residuals, achieving 18.5% accuracy improvement.

Xiang Li, Pengcheng Wang, Huazheng Wang et al.

Proposes Approximate Speculative Decoding (ASD), a zero-training verifier that boosts throughput by up to 15.26% via budgeted longest-prefix selection.

Yuannuo Feng, Zegang Peng, Yuxin Xie et al.

Maglev combines sliding-window attention with fixed-size recurrent memory via a parallel-trained predictor, boosting long-sequence modeling by 1-2% accuracy.

Bo Liu, Qiang Liu

Z-PEFT employs layer-wise spectral features for zero-shot backdoor detection, achieving AUROC 0.9433 on unseen attacks.

Nicola Pitzalis, Donald Shenaj, Giacomo Cignoni et al.

HindSearch introduces trajectory-level hindsight critique with on-policy distillation, boosting average EM to 39.4% on seven QA benchmarks, outperforming prior methods.

Haowei Liu, Jiamian Wang, Hsin-Tai Wu et al.

AlphaG-OPD enhances reliability in symbolic factor discovery via structured on-policy distillation, showing strong cross-market performance.

Yaoyu Su

AOSpec jointly speculates actions and observations, reducing end-to-end latency by 11.8%-32.5% using lossless verification and EV-based observation drafting.

Hao Mark Chen, Jinnan Guo, Wayne Luk et al.

Agentic Bayesian Optimization via Sara excels in dynamic settings, surpassing traditional methods.

Paul Brunzema, Louis Tiao, Nhat Le et al.

AutoCause automates multi-method causal discovery in environmental time series, improving reproducibility and auditability.

Marco Ruiz, Miguel Arana-Catania, David R. Ardila et al.

OnlineCache enhances diffusion inference efficiency with dynamic caching and error correction, achieving nearly 3x speedup on FLUX.1-dev.

Zhikang Xie, Xichen Ye, Yifan Wu et al.

DASH-OPD introduces a discrepancy-aware hysteretic switching mechanism for on-policy distillation, improving multi-turn agent training efficiency and performance.

Yuchen Xia, Qianguo Sun, Chao Song et al.

CFTD uses MACE features to evaluate material generation models for quality and novelty.

Paul Hagemann, Katharina Ueltzen, Simon Müller et al.

LM-GRASP employs online imitation learning with Transformer to generate instance-specific solutions, outperforming traditional heuristics with 28.4 units improvement on PFSP.

Mohand Mezmaz, Grégoire Danoy

Flux-OPD uses evolving contexts and reverse KL decomposition to stabilize open-domain distillation, outperforming existing methods with +2-3 score improvements.

Yuran Wang, Zekun Wang, Bohan Zeng et al.

Proposes an expand-then-compress framework using multi-teacher policy distillation, improving Qwen3-1.7B's reasoning by 2.0%-8.3%.

Songshuo Lu, Zhi Chen, Yaohua Tang

Proposes a dual-objective surrogate model guiding neural ensemble search, improving accuracy and diversity simultaneously.

Alexandr Udeneev, Petr Babkin, Oleg Bakhteev

Proposes 'horizon residual' metric, comparing full-task success with short-stage predictions to diagnose long-horizon failures.

Chao Peng, Zhiheng Lyu, Peijie Dong et al.

Gradient, evolutionary, and one-shot NAS methods improve traffic prediction model automation, achieving up to 8% RMSE reduction.

Truong Giang Vu, Li Yang, Richard W. Pazzi

Using a diffusion-based generative framework for urban air quality reconstruction from sparse observations, achieving high performance on Paris dataset.

Abhishek A. Sabnis, Mihai Mitrea, Lya Lugon et al.

MemSFT uses external parametric memory to mitigate alignment tax, boosting domain performance with minimal impact on general skills.

Jiarui Wang, Xiang Shi, Jiaqi Cao et al.