This paper proposes Lite-LLM4Rec, a hierarchical LLM architecture that eliminates beam search, reducing inference time by 97.28% and boosting performance by 46.8%.

Hanbing Wang, Xiaorui Liu, Wenqi Fan et al.

Proposes a causal reasoning framework based on reinforcement learning, utilizing two key lemmas to quantify causality in multivariate stochastic processes.

Mehdi Fatemi, Sindhu Gowda

DoRA enhances LoRA's learning capacity via weight decomposition, outperforming LoRA on multiple tasks.

Shih-Yang Liu, Chien-Yi Wang, Hongxu Yin et al.

InfoRM introduces an information bottleneck for reward modeling, effectively mitigating reward hacking and improving generalization in RLHF.

Yuchun Miao, Sen Zhang, Liang Ding et al.

Proposes PNSIS framework, integrating necessity and sufficiency probabilities to extract invariant subgraphs, boosting graph OOD generalization.

Xuexin Chen, Ruichu Cai, Kaitao Zheng et al.

SLGM model enhances structure prediction via loss calibration and formatted decoding, significantly boosting <1B parameter model performance.

Minho Lee, Junghyun Min, Yerang Kim et al.

Introduces Target Score Matching (TSM), leveraging known target scores to improve low-noise score estimates, enhancing diffusion models’ accuracy.

Valentin De Bortoli, Michael Hutchinson, Peter Wirnsberger et al.

Proposes an equitable R&D framework based on Scrum, integrating tools like model cards and diversity metrics to mitigate bias in AI and robotics projects.

Andrew Hundt, Julia Schuller, Severin Kacianka

Introduced THE COLOSSEUM benchmark, evaluating 5 manipulation models across 14 environmental perturbations, with success rates dropping 30-50%, over 75% in combined perturbations.

Wilbert Pumacay, Ishika Singh, Jiafei Duan et al.

Introduces granular hyperparameter G and scaling laws for MoE, optimizing training to outperform dense Transformers under various compute budgets.

Jakub Krajewski, Jan Ludziejewski, Kamil Adamczewski et al.

PIVOT uses iterative visual prompting to enable zero-shot spatial reasoning and robotic control with VLMs, achieving 75% success in navigation tasks.

Soroush Nasiriany, Fei Xia, Wenhao Yu et al.

AIR-Bench introduces a hierarchical benchmark for large audio-language models, evaluating 19 tasks and 2000 open-ended questions with GPT-4 scoring, revealing current limitations.

Qian Yang, Jin Xu, Wenrui Liu et al.

Proposes Rolling Diffusion with sliding window scheduling, outperforming standard diffusion in complex temporal data, validated on Kinetics-600 and fluid simulations.

David Ruhe, Jonathan Heek, Tim Salimans et al.

Proposes parallel budget-controlled A/B testing with market segmentation and FPPE bias correction, improving estimation accuracy under market interference.

Luofeng Liao, Christian Kroer, Sergei Leonenkov et al.

GALA3D employs LLM-generated layouts and Gaussian splatting for high-fidelity complex scene synthesis, outperforming prior methods with CLIP scores of 35.052.

Xiaoyu Zhou, Xingjian Ran, Yajiao Xiong et al.

By conducting semi-structured interviews and observations with 23 guide dog handlers and 5 trainers, this study identifies key interaction challenges and design principles for guide dog robots.

Hochul Hwang, Hee-Tae Jung, Nicholas A Giudice et al.

Proposes debate framework to improve weak model evaluation of strong models, achieving 76% accuracy with non-expert judges.

Akbir Khan, John Hughes, Dan Valentine et al.

InternLM-Math unifies chain-of-thought, reward modeling, formal proof, and code interpretation, achieving SOTA on multiple benchmarks.

Huaiyuan Ying, Shuo Zhang, Linyang Li et al.

iDEM employs diffusion-based energy matching for high-dimensional Boltzmann sampling, achieving 2-5× faster training without MCMC samples.

Tara Akhound-Sadegh, Jarrid Rector-Brooks, Avishek Joey Bose et al.

SPHINX-X scales data and parameters, achieving state-of-the-art multi-modal performance across diverse tasks, from 1.1B to 56B parameters.

Dongyang Liu, Renrui Zhang, Longtian Qiu et al.